133 116 17MB
English Pages 360 [356] Year 2019
The African Wild Dog
MONOGRAPHS IN B E H AV I O R A N D E C O L O G Y Edited by John R. Krebs and Tim Clutton-Brock Foraging Theory, by David W. Stephens and John R. Krebs Dynamic Modeling in Behavioral Ecology, by Marc Mangel and Colin W. Clark The Evolution of Parental Care, by T. H. Clutton-Brock Parasitoids: Behavioral and Evolutionary Ecology, by H.C.J. Godfray Sexual Selection, by Malte Andersson Polygyny and Sexual Selection in Red-Winged Blackbirds, by William A. Searcy and Ken Yasukawa Leks, by Jacob Hoglund ¨ and Rauno V. Alatalo Social Evolution in Ants, by Andrew F. G. Bourke and Nigel R. Fransk Female Control: Sexual Selection by Cryptic Female Choice, by William G. Eberhard Sex, Color, and Mate Choice in Guppies, by Anne E. Houde Foundations of Social Evolution, by Steven A. Frank Parasites in Social Insects, by Paul Schmid-Hempel Levels of Selection in Evolution, edited by Laurent Keller Social Foraging Theory, by Luc-Alain Giraldeau and Thomas Caraco Model Systems in Behavioral Ecology: Integrating Conceptual, Theoretical, and Empirical Approaches, edited by Lee Alan Dugatkin Sperm Competition and Its Evolutionary Consequences in the Insects, by Leigh W. Simmons The African Wild Dog: Behavior, Ecology, and Conservation, by Scott Creel and Nancy Marusha Creel
The African Wild Dog Behavior, Ecology, and Conservation SCOTT CREEL AND NANCY MARUSHA CREEL Princeton University Press Princeton and Oxford
Copyright 䉷 2002 by Princeton University Press Published by Princeton University Press, 41 William Street, Princeton, New Jersey 08540 In the United Kingdom: Princeton University Press, 3 Market Place, Woodstock, Oxfordshire OX20 1SY All Rights Reserved Library of Congress Cataloging-in-Publication Data Creel, Scott, 1962– The African wild dog : behavior, ecology, and conservation / Scott Creel and Nancy Marusha Creel. p. cm. — (Monographs in behavior and ecology) Includes bibliographical references (p. ). ISBN 0-691-01655-0 (alk. paper) — ISBN 0-691-01654-2 (pbk. : alk. paper) 1. African wild dog. I. Creel, Nancy Marusha. II. T itle. III. Series. QL737.C22 C74 2002 599.77—dc21
2001051034
This book has been composed in T imes Roman Printed on acid-free paper. ⬁ www.pupress.princeton.edu Printed in the United States of America 10
9
8
7
6
5 4
3
2
1
10 9 (Pbk.)
8
7
6
5 4
3
2
1
For Andie and Bridget
Contents
1
2
3
4
Preface
xi
History and Natural History
1 3 4 7 7
1.1 1.2
Taxonomy and Phylogeny Social Organization
1.3
Ecology
1.4
Conservation Issues
1.5
Issues Addressed by the Research and Organization of the Book
The Selous, the Study Population, and General Methods 2.1
The Selous Game Reserve
2.2
The Study Area and Population
2.3
General Methods
Home Ranges and Habitat Selection
11 15 15 23 25 36 36 39
3.1
Specific Methods
3.2
Description of Home Ranges
3.3
Exclusive Areas, Overlaps and Territorial Defense
3.4
Den Locations and Characteristics
3.5
Pack Size and Range Size
3.6
Habitat Selection
41 50 51 52
3.7
Effect of Prey Distribution on Habitat Selection and Home Range Properties
55
3.8
Comparison with Other Wild Dog Populations
3.9
Summary
Cooperative Hunting and the Evolution of Sociality 4.1
Specific Methods
4.2
Hunting and Foraging Success
4.3
Prey Selection and Hunting Success
4.4
Cooperative Hunting Behavior
4.5
Characteristics of Kill Sites
59 65 67 69 73 74 76 84
viii ▪ C O N T E N T S 4.6
Quantitative Effects of Pack Size on Hunting Benefits and Costs
4.7
Optimal Hunting Pack Size
4.8
Net Rate of Food Intake vs. Efficiency
4.9
Effects of Group Size Unrelated to Hunting
4.11 Other Wild Dog Populations
95 96 97
4.12 Communal Hunting and Group Size: Comparisons with Other Species
98
4.10 Variance in Foraging Success
5
6
7
8
84 88 89
Prey Selection 5.1
Prey Availability and Encounter Rates
5.2
Encounters and Hunts
5.3
Hunts and Kills
5.4
Combined Effects of Encounter, Hunting, and Killing Probabilities on Prey Selection
5.5
Quantitative Models of Prey Selection
5.6
Summary
Ungulate Herd Sizes and the Risk of Predation by Wild Dogs
103 105 109 111 112 114 122 124 126
6.1
Probability of Being Encountered
6.2
The Probability of Being Hunted upon Encounter
6.3
Hunting Success
130 130
6.4
Kills per Encounter, Dilution of Risk, and Combined Measures of Vulnerability
133
Demography—Survival and Reproduction 7.1
Survival Rates
7.2
Reproduction
7.3
Density Dependence
7.4
Genetic Effective Population Size
7.5
Demographic Effective Population Size
Dispersal 8.1
Defining Dispersal in Social Carnivores
8.2
Number and Size of Dispersing Groups
8.3
Rates of Dispersal
8.4
Size of Dispersing Groups
145 145 159 173 175 176 179 181 184 184 184
C O N T E N T S ▪ ix 8.5
Linear Dispersal Distance
186
8.6
The Duration and Circumstances of Floating
187
8.7
Comparison with Dispersal in Other Wild Dog Populations
8.8
Mortality Risk of Dispersal
8.9
Dispersal and Escape from Reproductive Suppression
8.10 Dispersal and Escape from Inbreeding 8.11 Integrating Forces that Drive Dispersal
9
Reproductive Suppression, Social Stress, and the Behavioral and Endocrine Correlates of Rank
194 195 200
9.1
Are Dominants More Aggressive?
201 205
9.2
Do Dominants Mate More Often or More Effectively?
207
9.3
Do Hormonal Differences Accompany Behavioral Differences?
9.4
Nonbreeder Lactation
210 214
9.5
Does Social Stress Mediate Reproductive Suppression of Subordinates?
215
9.6
How Effective Is Reproductive Suppression of Subordinates?
216
9.7
Similarities and Differences between the Sexes in the Correlates of Rank
9.8
Interspecific Comparisons
9.9
Dominance and Stress
9.10 Do the Correlates of Rank Relate to Dispersal and Social Organization?
10
190 191
217 218 218 222
Patterns of Relatedness and the Fitness Consequences of Dispersal, Philopatry, and Reproductive Suppression 223 10.1
Age-specific Relatedness of Natal and Immigrant Subordinates to Breeders
10.2
Inclusive Fitness of Nondispersers
10.3
Inclusive Fitness of Dispersers
226 231 238
10.4
Incomplete Reproductive Suppression: Breeding by Subordinates
240
x ▪ CONTENTS
11
12
13
Interspecific Competition with Larger Carnivores 11.1
Specific Methods
245 246
11.2
Carnivore Densities and Distributions in Selous
248
11.3
Correlations between Species Densities
11.4
Diet Overlap
11.5
Direct Competition at Kills
11.6
Interactions Away from Kills
11.7
Impact of Interspecific Competition
253 257 259 263 265
11.8
Adaptations to Interspecific Competition
266
Infectious Diseases 12.1
Canine Distemper Virus
12.2
Rabies Virus
12.3
Anthrax
12.4
Canine Parvovirus
12.5
Other Pathogens
12.6
Behavior and Epidemiology
269 271 274 277 279 281 284
12.7
Impact of Diseases on Population Dynamics and Density
286
Extinction Risk and Conservation 13.1
Analysis of Extinction Risk with Leslie Matrix Projections
13.2
Stochastic Individual-Based Modeling of Extinction Risk
13.3
Sensitivity Analysis and Results
13.4
Summary and Recommendations
288 290 295 298 308
References
311
Index
339
Preface The theorist imagines his companions as a naive Romeo imagined his ideal Juliet. The experimenter’s lovers sweat, complain, and fart. —James Gleick, Chaos
In the last fifteen years, African wild dogs have made a lot of progress. After many years of languishing in the shadows of Africa’s better-known carnivores, wild dogs have recently been the focus of many studies and conservation efforts scattered across the continent. Much of the recent interest in wild dogs has been spurred by the realization that wild dogs face an uncertain future. In the 1980s, George and Lori Frame conducted a continentwide survey and concluded that only some 5,000 wild dogs survived in subSaharan Africa. Worse, the species’ range was contracting and some populations were shrinking. For reasons that were not well understood at the time of the Frames’s survey. wild dogs always live at low densities, and frequently disappear from protected areas that function perfectly well for the conservation of lions, hyenas, and leopards. Spurred by the recognition that wild dogs are at risk of extinction, masses of information have been gathered in wild dog studies by Gus Mills, Martyn Gorman, and Anthony Maddock in South Africa; Tico McNutt in Botswana; Josh Ginsberg and Greg Rasmussen in Zimbabwe; and a long series of scientists in Northern Tanzania, including George and Lori Frame, Hugo van Lawick, James Malcolm, Clare Fitzgibbon, John Fanshawe, Todd Fuller, Pieter Kat, Karen Laurenson, Stephen Lelo, and Roger Burrows. Reintroduction efforts in South Africa, Tanzania, and Namibia have met with mixed success, but generally provide hope that the wild dogs’ decline can not only be stopped, but reversed. In addition to answering questions posed by conservation concerns, field studies of wild dogs have made basic contributions to the fields of ecology and animal behavior. In particular, wild dogs have been excellent subjects for studies of cooperation and group living, hunting behavior, social behavior, and interspecific competition. Against this background, the intention of this book is to present the data from our study of African wild dogs in the Selous Game Reserve, which ran from 1991 to 1997. We compare our results to other studies of wild dogs and to studies of other carnivores and social species. The book is aimed at pro-
xii ▪ P R E FA C E
fessional biologists and students, but we hope it will be intelligible to others with an interest in wild dogs. The Selous is a tough place, and we are grateful to the many people and organizations who helped us there. First of all, our sincere thanks to Dr. Markus Borner and the late Dr. Richard Faust of the Frankfurt Zoological Society, for encouraging us to work on wild dogs in Selous and providing continuous, enthusiastic support of the project. Mohamed Kachui and Rehema Nguwasi made life much easier and more enjoyable. Willness Minja and was a real pleasure to work with as Sector Manager for Northern Selous. The game scouts of the northern sector and their families in Matambwe were always friendly and helpful. Gerald Bigurube and Benson Kibonde, both chief wardens of Selous, provided critical cooperation—our particular thanks to Gerald (later of Tanzania National Parks) for his open-minded approach during the early stages of our work. Sal Arsene provided pleasant occasional breaks at Mbuyuni Camp. Rex Jackson was a good friend in Matambwe, even after he broke his ultralight plane (and his back). Goran Spong was a good companion in the field. Rolf Baldus, Ludwig Siege, and everyone at the GTZ office in Dar es Salaam helped with mail and office space. Finally, our thanks to Peter Waser for providing the opportunity to begin working in Tanzania. For permission to conduct the research we are grateful to the Tanzania Division of Wildlife of the Ministry of Lands, Tourism and Natural Resources; the Tanzania Commission for Science and Technology; the Tanzania Wildlife Research Institute (formerly SWRI); and the management of the Selous Game Reserve. For funding, we are grateful to the Frankfurt Zoological Society, the National Science Foundation, the Smithsonian Institution, Rockefeller University, and the NSF-NATO fellowship program.
The African Wild Dog
1
History and Natural History
The scientific name of the African wild dog (Lycaon pictus) means painted wolf, a reference to their patchwork coats of brown, black, and white, which Angier (1996) aptly called “a furred version of combat fatigues.” Their shape follows the general canid body plan, with modifications accumulated over 3 million years of divergence from the rest of the dog family. For example, wild dogs have only four toes, having lost the fifth toe that persists as a vestigial dewclaw in most canids. Compared to wolves or coyotes, they are lean and tall, with outsized ears that complement their quiet vocalizations. Altogether, the wild dog is a unique and beautiful animal (Figure 1.1). Wild dogs stand 65 to 75 cm at the shoulder, and weigh from 18 to 28 kilograms (Smithers 1983). Though they have been described as sexually monomorphic (Malcolm 1979; Girman et al. 1993), males are from 3–7 percent larger than females in linear measures of body size (Table 2.3). The original suggestion that wild dogs are monomorphic was probably based on measurements of body mass, which is extremely variable, because a hungry wild dog can consume 8–9 kg of meat (about 1⁄3 of its own weight). Wild dogs have sparse hair, though there is variation among individuals. Part of this variation is related to age—yearlings have more hair than adult dogs, and old dogs can become almost hairless. Hair is particularly lost on the head, which begins to appear gray as the skin shows through. Captive wild dogs in cold climates also tend to have more hair. The color patterns of wild dogs are extraordinarily variable, and they appear to recognize one another individually at distances of 50 to 100 meters, suggesting that they make use of the information that coat variation provides. For example, when two packs encounter one another, dogs chase members of the other pack. The scene rapidly becomes chaotic, but we never saw dogs pursuing members of their own pack. Chases are often initiated from distances of 50 to 100 meters, so it seems likely that individuals are recognized by sight, though olfaction may also be involved. Most of the variation in color is on the trunk and legs. Patterns on the face are relatively invariant, with a black muzzle shading to brown on the cheeks and forehead, a black line extending up the forehead, and blackish-brown on the backs of the ears. Some dogs have a brown teardrop on the muzzle below the eyes. There is never white on the head, and the posterior part of the head and the dorsal surface of the neck are consistently brown or yellow. Colors on the body and legs are unpredictable. There is often a white patch
2 ▪ CHAPTER 1
Figure 1.1 African wild dogs are unusual canids, with a variegated coat, long legs, and large rounded ears.
just behind the forelegs, and dogs with little or no white elsewhere may have white on their forelegs or on the ventral surface of their neck or chest. The tail is almost always tricolored, with brown at the root, a black band, and a white tip. Some dogs have two black tail-bands, or black dots, or a black tip below the white, and a few have no white at all. Coat patterns are not bilaterally symmetrical. The asymmetry is great enough that photographs of a
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 3
dog’s right side cannot be matched to photographs of its left side without additional information. Wild dogs have only four toes on their forelegs, while members of the genus Canis have a vestigial fifth toe. The pads of the middle two toes are usually fused at the posterior edge, although in Selous we observed several individuals with unfused toes. The dental formula is 3 1 3 3 (upper), 3 1 4 3 (lower). The last lower molar is vestigial. The canine teeth are narrow for their length, in comparison to other carnivores (Van Valkenburgh 1989). In a set of 23 canids, felids, and hyenids, wild dogs had the largest premolars (relative to body mass) of all carnivores other than hyenas (Van Valkenburgh 1989). This suggests that wild dogs eat bone regularly, although they have a reputation for eating meat almost exclusively. In Selous, wild dogs often eat leg bones, ribs, vertebrae, and skulls. The droppings of wild dogs sometimes turn white with age due to a high proportion of digested bone, similar to the droppings of spotted hyenas.
1.1 Taxonomy and Phylogeny Fossil evidence does not resolve the origin of African wild dogs. Undisputed Lycaon fossils come from the mid-Pleistocene (about 1 million years ago), and are very similar to modern wild dogs (Savage 1978). There is some debate over the geographic range for fossils of Lycaon. Kurten (1968) suggests that skull fragments from the genus are found in late Pleistocene sites in Europe, but Thenius (1972) and Malcolm (1979) believe that these fragments came from wolves (Canis). If so, Lycaon may always have been restricted to Africa. Within Africa, identification of the oldest Lycaon is complicated by the difficulty of distinguishing Lycaon fossils from those of an early Pleistocene wolf, Canis africanus. The current view of fossil evidence is that wild dogs arose 2–3 million years ago, in Africa (Savage & Russell 1983). The first taxonomic description of a wild dog was by Temminck (1820), who considered it to be a type of hyena (and named it Hyena picta). Matthew (1930) placed wild dogs in a subfamily of the Canidae, the Simocyoninae, together with the dhole (Cuon alpinus) and the bush dog (Speothos venaticus). This group was proposed on the basis of the shape of the lower carnassial molar, which in these three species has a short blade and no basined cusp (Van Valkenburgh 1989). Lycaon, Cuon, and Speothos are not particularly similar in other respects. Bush dogs look nothing like dholes and wild dogs. Wild dogs and dholes are similar in morphology, behavior, and ecology (Johnsingh 1982; Venkataraman 1995), but Thenius (1954) described an Asian fossil lineage that leads from a jackal of the early Pleistocene to the dhole. Today, similar carnassial molars within the Simocyoninae are considered analogous rather than homologous, and the subfamily is no longer recognized (Wozencraft 1989).
4 ▪ CHAPTER 1
The wild dog has the same number of chromosomes as the domestic dog (Canis familiaris) and similar neuroanatomy (Radinsky 1973). The myoglobins of wild and domestic dogs differ by one amino acid, compatible with a single-point mutation (Romero-Herrera et al. 1976). Girman et al. (1993) sequenced 736 base-pairs of the cytochrome b gene in wild dogs and other canids. These sequence data suggest that wild dogs are phylogenetically distinct from the other wolflike canids (wolves, jackals, and coyotes), justifying their current placement in a monotypic genus. Wild dogs showed 11.3– 13.7% sequence divergence from the other species, and the single most parsimonious phylogenetic tree placed the divergence of the wild dog just basal to the radiation of the Canis clade. Girman et al. (1993) also noted 1% sequence divergence within the species, and proposed that two geographically isolated subspecies occupy eastern and southern Africa. This suggestion was based on samples from three widely separated locations (Kruger, Hwange, and Serengeti National Parks, respectively located in South Africa, Zimbabwe, and Tanzania). With samples from more locations, and with the addition of data on nuclear microsatellite genotypes and mtDNA control region sequences, the picture has changed (Girman et al. 1997). There are no geographically distinct subspecies, though there is substantial genetic variation among populations. Parsimony analysis of mtDNA control region haplotypes suggests that there are two clades of wild dogs, but the clades are geographically mingled. Unique mtDNA haplotypes are found at the northern and southern extremes of the sampled range (Serengeti and Kruger), but the genetic affinities of intervening populations are not clearly related to geography. In Selous, for example, the predominant mtDNA haplotype is most similar to a haplotype found only in Kruger, but not in the intervening populations of Zimbabwe and Botswana (Figure 1.2). Nuclear microsatellites also reveal gene flow among populations, but the patterns from nuclear and mitochondrial DNA do not match. For example, dogs from the Selous and Serengeti ecosystems share microsatellite alleles that are not found elsewhere, but mtDNA places these populations in different clades (Girman & Wayne 1997). The data, though extensive, leave open some questions about genetic divergence among wild dog populations. In general, continentwide genetic patterns are consistent with a history of radiations north and south from the miombo woodland belt (extending from the latitude of southern Tanzania in the north, to the latitude of northern Zimbabwe and northern Botswana in the south).
1.2 Social Organization Wild dogs live in permanent packs of 2 to 27 adults and yearlings, though packs of 5 to 15 adults and yearlings are most common. Excluding yearlings, packs held 6.6 Ⳳ 0.8 adults, taking the average for six populations (Table
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 5
Figure 1.2 Geographic range of the wild dog. (a) Distribution based on sightings in the past 15 years with major populations noted. Dark shading indicates populations within protected areas, stippling indicates populations outside of protected areas. Based on data from Fanshawe et al. (1991) and Ginsberg (1994). (b) Historical distribution. based on data from Smithers (1983), Fanshawe et al. (1991), and Ginsberg (1993).
3.9). Mean pack size varies from 4–5 adults in Kruger N. P. and Masai Mara N.R. to 8–9 adults in Moremi and Selous (Reich 1981; Fuller et al. 1992a; McNutt 1996; Mills & Gorman 1997). Within a pack, there is a clear dominance hierarchy among males, and another among females. The dominant female is usually the oldest in the pack, but old males often lose their rank to prime-aged males, so many packs include one or more old, formerly dominant males (Chapter 7). Only the dominant female is assured of breeding, though subordinate females do occasionally become pregnant. Reproduction is also largely monopolized by the alpha male, but the pups of a single litter can have more than one father, as in most carnivores (Girman et al. 1997; Chapter 8). The simplest pack structure is a set of related males and a set of related females, with no genetic relationship (or a distant relationship) between the males and females (Frame et al. 1979; Malcolm & Marten 1982; Girman et al. 1997). This structure becomes more complicated if offspring born within the pack are recruited. Individuals of either sex may stay in their natal pack well beyond the age of maturity. When this occurs, some individuals are related to pack members of the opposite sex. Pack structure can also be complicated by immigration. Generally, successful immigrants evict samesexed residents. These pack takeovers replace the lineage of one sex, but do not alter the basic structure of unrelated male and female lineages. Occasionally, unrelated immigrants join a pack without evicting all of the samesexed residents, and this dilutes relatedness within that sex. In Selous, unfa-
6 ▪ CHAPTER 1
miliar and apparently unrelated individuals of both sexes have immigrated successfully without evicting residents. In short, variations in the patterns of immigration, emigration, and breeder turnover may produce a complex web of genetic relatedness within packs. The coefficient of relatedness between packmates averages 0.25–0.35, but for a specific pair of individuals can range from 0 to 0.5 or above in the case of mild inbreeding (Frame et al. 1979; Reich 1981; Girman & Wayne 1997; Chapters 8, 10). Short-distance dispersal can also create genetic ties between neighboring packs (McNutt 1996; Girman et al. 1997). Females are more likely than males to disperse in some populations, including Selous (Frame & Frame 1976; Chapter 8). In other populations, dispersal is not sex-biased, or is male-biased (Reich 1981; McNutt 1996). Emigrants of both sexes are likely to disperse as yearlings or two-year-olds, and usually disperse as a single-sex group of littermates, or as a group composed of two cohorts born one year apart (McNutt 1996; Chapter 8). Many populations, including Selous, have an adult sex ratio biased in favor of males. For populations in which dispersal is female biased, the biased adult sex ratio may result from mortality during dispersal. In captivity, however, the sex ratio of a large sample of pups was also male-biased (Malcolm 1979), and pup sex ratios are male-biased in some wild populations (Fuller et al. 1992a). The sex ratio of pups is 1:1 for some populations, including Selous and Kruger (Maddock & Mills 1994; Chapter 7). Differences among populations in pup sex ratios might be related to rates of alpha female turnover, because primiparous females produce a high proportion of sons, while multiparous females produce a high proportion of daughters (Creel et al. 1998). No unaided pair of wild dogs has been observed to raise pups, and in Selous no pack smaller than five adults raised pups to independence (Chapters 7, 10). Subordinates of both sexes help to raise the pack’s young, which, as we mentioned, are normally produced by the dominant pair. The most important help comes in the form of food. For the first three months after they are born, pups cannot move quickly enough to follow a hunting pack, and are confined to a den. Most of the pack leaves to hunt twice a day, but one or more dogs remain behind as guards (Malcolm & Marten 1982). The alpha female normally guards the pups by herself, but in some cases another dog (usually a female) will remain with her. When the hunters return, both the pups and the mother solicit food, and dogs of both sexes (and all ages) respond by regurgitating meat (Malcolm & Marten 1982). Less often, dogs will carry a portion of a carcass to the den, usually a leg, to gnaw on. In addition to feeding pups, nonbreeding helpers take part in protecting the pups, from lions, leopards, and spotted hyenas. When the pups begin moving with the pack at about three months of age, they are often bivouacked during a hunt—left behind and later recovered (as with wolves; Mech 1970). One or more dogs of either sex may remain with the pups
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 7
under these circumstances. If no adults remain with the bivouacked pups, dogs of either sex may go back to retrieve them and lead them to the kill. Pups are allowed to eat first at carcasses (though adults sometimes eat hastily until the pups arrive), followed by yearlings and then adults (Malcolm 1979; unpublished observations).
1.3 Ecology Wild dogs rely almost exclusively on mammalian prey that they have killed for themselves. They hunt prey as small as hares (1–2 kg), and as large as adult zebra or juvenile buffalo and eland (about 200 kg), but concentrate on prey between 10 and 120 kg, with larger packs taking larger prey (Chapter 4). Impala and wildebeest are an important part of their diet in most ecosystems. The remainder of the diet is made up of species smaller than wildebeest that are locally abundant, such as greater kudu, warthogs, and duikers. Wild dogs rarely scavenge, probably to avoid risky encounters with larger carnivores (Kruuk & Turner 1967; Creel & Creel 1996). Where the density of spotted hyenas is high or visibility is good, kleptoparasitism by hyenas at wild dog kills is common (Estes & Goddard 1967; Malcolm 1979; Fanshawe & Fitzgibbon 1993). Predation on wild dogs by lions has been seen in most populations, and lion predation is the most common known cause of death in some populations (Ginsberg et al. 1995; Mills & Gorman 1997). Altogether, interference competition with larger carnivores is an important force shaping the behavior, number, and distribution of wild dogs (Creel & Creel 1996; Mills & Gorman 1997).
1.4 Conservation Issues Conflict with Humans Like most large carnivores, the single most important conservation problem for wild dogs is conflict with an expanding human population. Wild dogs formerly had a wide distribution across sub-Saharan Africa, excepting only rainforest (Smithers 1983). Like many species, wild dogs have become patchily distributed as the human population has expanded (Figure 1.2). Wild dogs now live mainly in protected areas, and few areas are known to hold more than a hundred individuals (Fanshawe et al. 1991). As a landscape is settled and moves into agricultural use, prey populations are depleted so that carnivore populations cannot maintain themselves. If carnivores persist, they are often killed to remove threats to livestock and people. It is occasionally suggested that wild dogs kill people (Leakey 1983), but we know of no documented cases. Wild dogs are wary of people
8 ▪ CHAPTER 1
unless they have been habituated to tourism, and, in our experience, villagers near protected areas do not fear them. Wild dogs will kill unattended sheep and goats (Rasmussen 1996), but do not attack livestock that are attended by a shepherd. Wild dogs in our study area often moved out of the reserve through areas with scattered rice farms and small dirt tracks. In these areas, they skirted around people, and we never saw a direct interaction other than the dogs running from a person who had approached them without being detected. Like some other carnivores, notably spotted hyenas, wild dogs were actively persecuted by wildlife managers for much of the 20th century. In general, wildlife managers shot them whenever possible. In Zimbabwe, 3,404 wild dogs were shot for “vermin control” between 1956 and 1975 (Childes 1988). In Namibia, 156 wild dogs were killed over 19 months in 1965–1966 (Anonymous 1967). Most game scouts in Selous recall shooting wild dogs up to the mid 1980s, and it is likely that hundreds were shot, though there are no accurate records. In 1977, the South African Red Data Book stated “[wild dogs are] still considered vermin and are shot on sight even on nature reserves. . . . [They are] likely to get little sympathy from farmers” (Skinner et al. 1977, p. 11). Dislike of wild dogs can easily be seen in writings from the 1900s through the 1970s. Some examples: It will be an excellent day for African game and its preservation when means can be devised for [wild dogs’] complete extermination.— Maugham (1914) Although wild dogs, when present in large numbers, are a scourge to the game, killing, terrifying, and scattering it all over the country, they still find a useful place in Nature’s economy, and the Kruger National Park would certainly be the better for a considerably larger number than exists.—Stevenson-Hamilton (1947) The wild dog is the only animal of the veldt that is always feared. The lion is not. Many a hunter has watched a full-fed lion walk in plain sight of a herd of antelopes.—Hubbard (1954) Wild dogs hunt in packs, killing wantonly far more than they need for food, and by methods of the utmost cruelty.—Bere (1956) In a later annotation, Bere noted, “This is now known to be nonsense.” The rapacious appetite of these foul creatures is staggering.—Hunter (1960) Though some of these authors had a grudging respect for the dogs (Stevenson-Hamilton’s 1947 book is a good example), there were two broad reasons for their persecution. The major problem was with the dogs’ method of killing prey. Because they are small relative to their prey, and do not have a
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 9
specialized killing bite, wild dogs kill their prey by pulling it to a halt and disemboweling it. Large prey can take a half-hour to die (though most die in minutes), and empathy for the prey led to antipathy toward the predator. A second strike against the dogs was the perception that they disrupted prey populations more than other predators. Because wild dogs are cursorial hunters that rely on an open chase to catch their prey, it is certainly true that a wild dog hunt can set a large number of prey in motion, especially in open habitat. However, it is also true that calm returns quickly to an area in which the dogs have hunted. Wildebeest herds often resume grazing in plain sight of wild dogs feeding on a herdmate. Prey show little fear of wild dogs at rest, just as with other predators. Anyone who watches a pack of wild dogs for a day will undoubtedly see prey herds moving past or feeding nearby, aware of the dogs but unbothered. Zebra and wildebeest sometimes approach resting dogs and harass them. Some early naturalists must have known that relations between wild dogs and prey were much like those of other carnivores. Active persecution decreased as field studies described the wild dogs’ ecology and behavior. By the mid 1980s wild dogs were legally protected in the six nations that hold significant numbers (Botswana, Kenya, South Africa, Tanzania, Zambia, and Zimbabwe). Road accidents kill wild dogs in areas that are transected or bordered by high-speed roads (Fanshawe et al. 1991; Drews 1995). The rain-rutted dirt tracks in Selous do not allow high-speed driving, and we recorded no road kills. By contrast, Hwange National Park borders a high-speed highway between two large cities, Bulawayo and Victoria Falls, and road kills were the most common known cause of death (Ginsberg et al. 1995). In Mikumi National Park (Tanzania), traffic on the Tanzania-Zambia highway is estimated to kill between 3% and 12% of the wild dog population annually (Drews 1995; Creel & Creel 1998). Wire snares set for game species can unintentionally catch carnivores, and this is a surprisingly common cause of death in some places (Hofer et al. 1993). In Selous, snaring and poisoning by illegal game hunters caused 11% of 45 known-cause deaths. Snaring and shooting accounted for 18% of 57 deaths in Kruger (van Heerden et al. 1995), and 29% of 31 deaths in Hwange (Ginsberg et al. 1995). Though its force varies among populations, human impacts on wild dogs are substantial even in large protected areas. Low Density within Protected Areas and Interspecific Competition with Larger Carnivores If conflict with humans was the only problem that wild dogs faced, they would not be endangered. Many African nations have set aside large areas for wildlife, and these parks hold a great many lions, leopards, and hyenas. All three of these species pose a greater threat to livestock (and people) than
10 ▪ C H A P T E R 1 Table 1.1 Densities of wild dog populations throughout Africa (adults/1,000 km2) Population
Density
Source
Selous, Tanzania Moremi, Botswana Hluhluwe, South Africa Kruger, South Africa Hwange, Zimbabwe Ngorongoro, Tanzania Serengeti, Tanzania Botswana/Namibia/Zambia Northern Kenya
38 40 33 16.7 15 0–26 0–15 2–3a 2–3a
Creel & Creel 1996 McNutt 1996 Maddock 1993 Mills & Gorman 1997 Childes 1988; Ginsberg 1993 Estes & Goddard 1967 Malcolm 1979; Burrows et al. 1994 Ginsberg 1993 Ginsberg 1993
a
Populations largely outside of protected areas.
wild dogs do, but they remain abundant and widespread. Although their ecological needs are similar, a fundamental difference between wild dogs and these larger carnivores is that wild dogs remain at low population density under all conditions. It seems likely that competition between wild dogs and larger carnivores explains this pattern (Creel & Creel 1996; Mills & Gorman 1997; Gorman et al. 1998). Frame (1985) described wild dog-hyena interactions in Serengeti: “Hyenas typically assembled behind wild dog packs as they hunted, and we recorded periods of weeks at a time in which hyenas stole almost all kills made by the dogs before the latter finished eating” (p. 3). Mills & Gorman (1997) showed that lions account for 43% of natural wild dog deaths in Kruger. Interference competition with lions and spotted hyenas also has a strong impact on cheetahs (Laurenson 1995; Durant 1998; cf. Crooks et al. 1998), and considerable data suggest that interspecific competition has strong effects on many carnivore populations (Palomares & Caro 1999; Creel et al. in press). We discuss interspecific competition in Chapter 11. Analyses of carnivores’ distributions (within ecosystems and among ecosystems) suggest that interspecific competition limits wild dogs in number and distribution (Creel & Creel 1996; Mills & Gorman 1997). Regardless of the cause, wild dog densities are spectacularly low (Table 1.1). The highest population density on record is from the northern Selous, with an average of 1 adult/26.0 km2 over six years. A more typical density is 1 adult per 60– 100 km2. Even at maximal density, an area of 1,000 km2 holds a population of only 40 adults, which is unlikely to be viable in the long run. As a result, small parks will play a small role in wild dog conservation, unless they are actively managed. In the end, conservation of wild dogs comes down to understanding the causes and consequences of their invariably low population densities.
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 11
Infectious Diseases The literature on wild dogs often states that they are “particularly sensitive to disease” (Fanshawe et al. 1991, p. 140), or that infectious diseases have played “a main role in the numerical and distributional decline of African wild dogs” (Kat et al. 1995, p. 229). This idea is based almost exclusively on data from the Serengeti ecosystem. There, wild dogs declined to local extinction while experiencing recurrent outbreaks of rabies and possibly canine distemper (Schaller 1972; Malcolm 1979; Gascoyne et al. 1993; Alexander & Appel 1994). The data from Serengeti clearly shows that viral diseases can cause substantial mortality in wild dogs, and can contribute to a local extinction. However, the Serengeti population was probably vulnerable to extinction for other reasons. First, the population was small enough (less than 30 dogs) to be vulnerable to a knockout blow, regardless of the cause (Ginsberg et al. 1995). Second, the Serengeti dogs faced intense competition from larger carnivores (Frame & Frame 1981). Finally, Serengeti held a diverse suite of carnivores, many at high densities, that were known to carry rabies virus and/or canine distemper virus (Maas 1993; Alexander & Appel 1994; Alexander et al. 1994, 1995; Roelke-Parker et al. 1996). Under these conditions, it is expected that spillover transmission from high-density species will endanger species living at lower density (Grenfell & Dobson 1995). For these reasons, it might not be justified to generalize the conclusion that wild dogs are especially vulnerable to diseases. Little is known about the regulatory role of diseases in other wild dog populations, but current data suggest that disease is not a major factor for all populations. Several dogs have died of infection with the bacterium Bacillus anthracis, in the Luangwa valley, Kruger N. P. and Selous (Turnbull et al. 1991; Creel et al. 1995; van Heerden et al. 1995). In Kruger and Selous there have not been detectable disease-related population declines over periods of 22 and 6 years, respectively (Reich 1981; van Heerden et al. 1995; Creel et al. 1997c). Combining demography, serology, post-mortems and veterinary examinations, van Heerden et al. (1995) concluded that “disease could not be incriminated as an important cause of death” (p. 18) in Kruger. In summary, current data are compatible with a wide range of views on the role of infectious disease in wild dog population dynamics. We discuss infectious diseases in Chapter 12.
1.5 Issues Addressed by the Research and Organization of the Book This book moves between results that are relevant to conservation and results that are relevant to behavioral ecology. Most chapters are more closely aligned to one of these fields than the other, but some data are relevant to
12 ▪ C H A P T E R 1
both fields. For example, we use data on hunting success to address the evolution of cooperative hunting, but also test whether hunting success is a limiting factor for some populations. For this reason, some results appear in different forms in more than one chapter, with discussions aimed at different goals. Chapter 2 gives a description of the Selous Game Reserve, the study site and population, and our general methods. We give narrower descriptions of some specific methods in other chapters for the sake of coherence. In Chapter 3 we discuss habitat selection, determinants of home range size, and overlap of home ranges. These analyses are relevant to conservation, because (all else equal) large home ranges lead to low population density. Wild dogs can have home ranges larger than 1,000 km2, among the largest reported for carnivores (Frame et al. 1979; Mills & Gorman 1997). Understanding why wild dog packs require large ranges is central to understanding why they are endangered. Chapters 4–6 deal with hunting. In Chapter 4 we present basic data from wild dog hunts in Selous and analyze the energetic costs and benefits of hunting in different pack sizes. A prominent question in behavioral ecology has been whether cooperative hunting favors life in groups or is simply an unselected consequence of life in groups. Among carnivores, this question has been addressed by studies of lions (Schaller 1972; Packer et al. 1990; Stander 1992), spotted hyenas (Kruuk 1972; Mills 1990; Holekamp et al. 1997), cheetahs (Caro 1994), wolves (Schmidt & Mech 1997), and wild dogs (Estes & Goddard 1967; Fanshawe & Fitzgibbon 1993; Fuller et al. 1995; Creel & Creel 1995b; Creel 2001). The question has also been studied in other taxa, notably chimpanzees (Boesch 1994), Harris’s hawk (Bednarz 1988), and killer whales (Baird & Dill 1996). Conclusions have varied, depending on the species studied and on the currency used to measure foraging success (Packer & Caro 1997; Creel 1997). Some authors argue that cooperative hunting has not been an important force in the evolution of sociality in carnivores (Packer et al. 1990; Caro 1994), but for wild dogs, it is clear that foraging success depends on group size. We feel that this important issue remains unresolved for carnivores in general (Creel 2001). In Chapter 5 we focus on prey selection. We use our hunting data to measure the profitability of each prey species, then test whether the proportion of each prey species in wild dogs’ diet is correlated to its profitability. The diet might also be determined by the availability of prey types, rather than their profitability alone. Because we have data on the abundance of each prey species, encounter rates between dogs and prey, and measures of profitability, we consider availability and profitability together. Chapter 6 digresses to take the perspective of the prey. In particular, we examine how herd size affects the vulnerability of impala and wildebeest to wild dogs. Grouping could reduce vulnerability to predation in several ways (Caro & Fitzgibbon 1992; Fitzgibbon & Lazarus 1995; Lima 1995a, 1995b).
H I S T O R Y A N D N AT U R A L H I S T O R Y ▪ 13
Most studies of group size and vulnerability to predation have involved stalking predators, focusing on the benefit of detecting predators before they can come close enough to make a kill (fish: Krause & Godin 1995; birds: Lima & Zollner 1996; mammals: Fitzgibbon 1990). However, stalkers and coursers hunt in very different ways, and our analyses suggest that many mechanisms that reduce vulnerability to stalking predators do not reduce vulnerability to coursing predators. Chapters 7–10 describe social organization and behavior. In Chapter 7 we present basic data on demography and population dynamics, including the following topics: (1) life tables, (2) the effects of social rank on survival and reproduction, (3) the effectiveness of helpers, (4) sex-ratio evolution, (5) population dynamics, and (6) effective population size (Wright 1969; Nunney & Elam 1994). Chapter 8 describes patterns of immigration and emigration. Early studies showed that female wild dogs were the primary dispersers in Serengeti (Frame & Frame 1976). Female-biased dispersal is rare among carnivores (Waser 1996) and among mammals in general (Chepko-Sade & Halpin 1987). Even among wild dogs, female-biased dispersal may not be the general rule, as McNutt (1996) found that all dogs of both sexes dispersed in Moremi. In Selous, females are substantially more likely to disperse than males. Chapter 8 describes patterns of dispersal, and discusses the effect of dispersal on genetic relationships within and among packs. We compare predicted patterns of relatedness to data from mitochondrial DNA and microsatellites (Girman et al. 1997). Dominant wild dogs do not disperse unless evicted by immigrants, but subordinates of both sexes commonly leave their pack. This suggests that escape from reproductive suppression is a driving force behind dispersal (Waser 1996). In Chapter 9, we discuss the behavioral and endocrine mechanisms that prevent reproduction in social subordinates, addressing six questions: (1) What is the effect of social subordination on mating rates? (2) To what degree is reproductive suppression of subordinates due to aggression from dominants? (3) Is reproductive suppression of subordinates strictly a behavioral process, or is it physiologically mediated by depressed sex-steroid levels? (4) Is suppression of subordinates mediated by stress? (5) How do mechanisms of suppression differ between males and females? (6) How do behavioral and endocrine patterns relate to wild dogs’ social organization? We then take a comparative perspective, asking how the physiological and behavioral correlates of reproductive suppression differ among social carnivores, and among cooperative breeders in general. In Chapter 10 we examine reproductive suppression from an evolutionary standpoint, asking why social subordinates tolerate reproductive suppression and help to raise the young of others. A gargantuan literature addresses this question from a theoretical perspective (Hamilton 1964; Brown 1987) or with empirical data from birds (Brown 1987; Stacey & Koenig 1990), in-
14 ▪ C H A P T E R 1
sects (Bourke & Franks 1995), and, to a lesser extent, mammals (Solomon & French 1997). Rather than attempting to review this subject (a book in itself), we take a narrow focus and apply a quantitative model for the evolution of reproductive suppression (Vehrencamp 1983). We compare patterns predicted by the model to data on mating rates and reproductive physiology, and to direct data on maternity and paternity. Chapters 11–13 turn to conservation, addressing the issues discussed earlier in this chapter. In Chapter 11 we examine interspecific competition between wild dogs and larger carnivores. Across ecosystems, the density of wild dogs is negatively correlated with the densities of lions and spotted hyenas, and considerable data suggest that interference competition and predation cause this correlation. In Chapter 12, we discuss the effects of infectious diseases. In Chapter 13, we provide an overview of six factors that may limit wild dogs in number or distribution: intraspecific competition, interspecific competition, prey limitation, disease, genetic problems, and human activities. We then use simulations to model the probability of local extinction in Selous.
2
The Selous, the Study Population, and General Methods
2.1 The Selous Game Reserve The Selous Game Reserve, covering 43,600 km2, is the largest protected area in Africa, and one of the largest in the world. The Selous ecosystem spans 78,650 km2, about half of the Rufiji River basin that drains southeastern Tanzania. Four major rivers (the Ruaha, Kilombero, Luwegu, and Mbarangandu) join within the reserve, flowing into the Rufiji River and on to the Indian Ocean (Figure 2.1). The Selous is among the oldest wildlife reserves in Africa. The nucleus of the reserve was gazetted in 1905 by the German colonial administration. This was expanded in 1912 to create three reserves: Mahenge, Mohoro, and Matundu. In 1922, the British administration again redrew the boundaries to link the three smaller areas into a single reserve, now named the Selous Game Sanctuary in memory of Frederick Selous, who was killed in World War I just north of the Rufiji River. Selous was a naturalist best known for his exploration of the area that would later become Rhodesia and then Zimbabwe. Selous also led U.S. President Theodore Roosevelt on an expedition through East Africa to collect specimens for the American Museum of Natural History. He was buried where he fell after being shot in the head, and his grave lies at Beho Beho, in an area where the outline of World War I trenches can still be seen. Prior to 1945, the reserve was inhabited by hunters, fishermen, and shifting cultivators, primarily from the Wangindo, Wapogoro, and Warufiji tribes (Rodgers 1979). Human settlement was sparse because poor soils inhibited dirt farming, while abundant tsetse flies (Glossina morsitans) limited the keeping of livestock by transmitting trypanosomiasis. Between 1945 and 1947, residents were resettled outside the reserve, ostensibly to reduce the impact of trypanosomiasis on human health. Following resettlement, the borders were adjusted and the reserve took its present shape. Beginning in 1964, the newly independent Tanzanian government developed the reserve for hunting, establishing more than 7,000 km of dirt tracks, airstrips, bridges, and ferries. Much of this infrastructure is now in disrepair. The current boundary of the reserve was written into law in 1976. Some of the boundary description relies on place names that are slightly ambiguous, so statements of the reserve’s size vary from 43,000 to 44,800 km2. We have adopted an estimate
16 ▪ C H A P T E R 2
Figure 2.1 The Selous Game Reserve and adjoining protected areas. SGR: Selous Game Reserve, MNP: Mikumi National Park, a: Gonabisi Open Area (O.A.), b: Kisarawe O. A. , c: Tapika O. A. , d: Kilwa O. A. , e: Liwale North O. A., f: Liwale South O. A., g: Kilombero Game Controlled Area (north and south), h: Mahenge South O. A., i: Mahenge North O.A.
(43,600 km2) from a digital map prepared by plotting the boundary from its legal description onto 1:250,000 base maps (TWCM 1990). The Selous ecosystem includes the Selous Game Reserve itself, Mikumi National Park (3,140 km2), Udzungwa National Park (2,000 km2), Kilombero Game Controlled Area (5,300 km2), and eight “Open Areas” (⬃35,000 km2). Human settlement is not allowed in the Reserve or in National Parks, but is
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 17
allowed within Game Controlled Areas and Open Areas. Tourist hunting is allowed in Game Reserves, but not in National Parks. Hunting by residents is allowed in Open Areas, which can also be leased for tourist hunting in designated Community Conservation Areas. The boundary of the ecosystem is vaguely defined, and recent aerial censuses have set the ecosystem’s area as low as 73,000 km2, and as high as 92,000 km2 (TWCM 1994). We have adopted an estimate of 78,650 km2, based on an aerial survey conducted in 1989 (TWCM 1990). The Selous rises from 80 meters above sea level in the northeast to 750 meters in the south, topped by the 1,300m Mbarika mountains in the southwest. Five rivers—the Njenje, Mbarangandu, Luwegu, Luhombero, and Kilombero—flow down this inclined plane from southwest to northeast, to join the Great Ruaha River. Below this confluence lies the Rufiji river, which bisects the reserve and carries 54,000 cubic meters of water per minute into the Indian Ocean (FAO 1961). Geologically, the oldest formations in Selous are composed of quartzite, limestone, gneiss, and schist from the Usagarian system of basement rocks, but the landscape is mostly derived from Karoo sandstone deposited in the Jurassic and Cretaceous periods (Haldeman 1962). Erosion of these formations through the Tertiary and deposition of alluvial sand in the Pleistocene epoch produced a rolling landscape cut by minor ridges and floodplains (Haldeman 1962; Rodgers 1979). At 7⬚S to 11⬚S, the Selous experiences one dry season and one wet season annually. The wet and dry seasons are of equal length, with the rains beginning in November or December, and lasting through early June. Satellite data on cloud cover suggest that annual rainfall ranges from 750 to 1300 mm in a gradient from light in the northeast to heavy in the southwest (Tanzania Department of Wildlife 1995). Our data suggest that these values are substantially too low, at least for the northern portion of the reserve. We recorded rainfall and temperature between 1992 and 1995 at Matambwe, in the northwest of the reserve. There, annual rainfall averaged 1406 mm, concentrated between January and June (Figure 2.2). August was the coldest month (mean minimum ⳱ 19.4 Ⳳ 0.3⬚C; mean maximum ⳱ 32.4 Ⳳ 0.2⬚C) and January was the hottest (mean minimum ⳱ 24.8 Ⳳ 0.2⬚C; mean maximum ⳱ 36.9 Ⳳ 0.2⬚C). Daily minimum temperatures ranged from 15.6⬚C in June and August, to 32.2⬚C in February. Daily maxima ranged from 21.1⬚C in June, to 41.1⬚C in February. Temperatures varied more in the wet season than in the dry season (Figure 2.3). Temperatures were higher and rainfall was heavier than Rodgers (1979) reported for Kingupira in eastern Selous. Like most of southern Tanzania, Selous is dominated by miombo woodland. Miombo is defined by trees of two genera, Brachystegia and Julbernardia, which are deciduous and have no thorns. In Selous the dominant species are Brachystegia spiciformis and Julbernardia globifera. The understory is dominated by Diplorhynchus condylocarpon in thick miombo, and by Com-
18 ▪ C H A P T E R 2
Figure 2.2 Monthly means for (a) daily minimum and daily maximum temperatures, and (b) rainfall, recorded at Matambwe, 1992–1995.
bretum in more open areas. The miombo of Selous also holds many coastal trees, particularly in riverine areas and thickets. The sparse grass layer is dominated by Panicum infestum (Rodgers 1979). Miombo is maintained by fire, but miombo grades into chipya woodland in areas that are very heavily burnt. Chipya is more open than miombo, with smaller trees and dense, coarse grass. The dominant trees vary among loca-
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 19
Figure 2.3 Variability in temperature, shown by plotting the coolest and hottest daily maximum recorded in each month, and the coolest and hottest minimum. Data from 1992–1995, collected at Matambwe. Temperatures were more variable in the wet season (January–June).
tions, and include Pterocarpus, Pteleopsis Pseudolachnostylis, Millettia, Combretum, and Terminalia. Dominant grasses are Andropogon schirensis and species of Hyparrhenia (Rodgers 1979). On our study area in the north of the reserve, early successional chipya dominated by Combretum and Terminalia sericea was common. The boundaries between miombo, chipya, and Combretum woodland are graded and difficult to define, so our vegetation map of northern Selous (Figure 2.4) combines these into a single category (deciduous woodland). In addition to the thornless deciduous woodlands, northern Selous has substantial areas of thorn woodland (Figure 2.5), dominated by Terminalia spinosa and Acacia drepanolobium. In thorn woodlands, the grass is generally low and sparse, including Sporobolus ioclados, Digitaria milanjiana, and Panicum infestum. These areas do not burn as frequently as deciduous woodland because the grasses provide less fuel. Patches dominated by Acacia nigrescens and long coarse grasses occur within Terminalia thorn woodland. Seasonally flooded long grass plains occupy much of northern Selous. The grass is dense and commonly grows to 3 meters. Common species include Setaria sphacelata, Andropogon gayanus, and Echinochloa haploclada (Rodgers 1979). Large tracts of floodplain are “tussock grass,” commonly including Sporobolus pyramidalis, which creates sharply defined humps of soil alternating with small channels. These tussocks make travel difficult for most animals (including people!) other than elephants. Most long grass areas burn every year.
20 ▪ C H A P T E R 2
Figure 2.4 Vegetation of the Northern Selous. The map was prepared by noting the dominant vegetation at 1,300 points, localized with a geographic positioning system. These locations were plotted onto a digital base map in the geographic information system IDRISI, taken from an existing 1:125,000 map. Boundaries of vegetation zones were digitized around clusters of points by hand. Vegetation classes are described in the text. Deciduous woodland includes miombo, chipya, and Combretum woodland.
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 21 Table 2.1 Estimates of wildlife populationsa in the Selous Game Reserve, 1989–1994b
Area Surveyed Species Buffalo Elephant Hartebeest Impala Wildebeest Zebra
Sept. 1989
June 1991
Sept. 1994
42,822
41,991
43,626
61,846 Ⳳ 14,359 24,548 Ⳳ 3,564 14,840 Ⳳ 2,432 21,698 Ⳳ 2,419 43,189 Ⳳ 2,296 18,729 Ⳳ 2,397
30,134 Ⳳ 5,147 12,262 Ⳳ 1,625 9,706 Ⳳ 1,338 53,632 Ⳳ 6,674 39,088 Ⳳ 3,590 23,699 Ⳳ 7,112
138,102 Ⳳ 11,271 31,735 Ⳳ 4,536 11,788 Ⳳ 1,752 29,507 Ⳳ 3,040 46,347 Ⳳ 5,923 22,454 Ⳳ 2,262
Mean Ⳳ SE. Data from aerial surveys conducted by the Tanzania Wildlife Conservation Monitoring program of the Tanzania Department of Wildlife and the Frankfurt Zoological Society.
a
b
Small areas of northern Selous are dominated by short grass with scattered trees. These areas usually fall on alkaline hardpan soils neighboring Terminalia spinosa–Acacia drepanolobium woodland. They are maintained by heavy grazing, with a grass community similar to those in the neighboring Terminalia woodland. Because the grass is continuously cropped to a few centimeters, short grass areas rarely burn. Narrow bands of riverine thicket are found along small seasonal rivers. Larger riverine forest areas occur along the Rufiji River, grading into areas of palm swamp (Figure 2.4). Wildlife is distributed throughout the ecosystem, including areas outside the reserve itself, especially in the rainy season. Total wildlife densities greater than 100 individuals/km2 are tallied by aerial survey in many areas outside the reserve, with a maximum of 600 individuals/km2 in the Gonabisi Open Area just north of the reserve boundary (TWCM 1994). Inside the reserve, wildlife densities are higher, and distributions are less patchy. Wildlife density has consistently been higher in the north of the reserve (our study area) than in other areas. Tables 2.1 and 2.2 give population estimates from recent aerial surveys for major wildlife species, for the reserve, and for the entire ecosystem. Though densities of most ungulates are moderate, the reserve’s huge size makes it unique. Notably, Selous holds the largest elephant and buffalo populations in Africa, and is thought to hold the largest lion and leopard populations as well (Tanzania Department of Wildlife 1995). The aerial census data in Tables 2.1 and 2.2 are not adjusted for undercounting, as no ground truthing exercises were conducted. Particularly in woodland, small species (e.g., impala) are counted less effectively than large species (e.g., elephant and buffalo). On the ground, the most commonly encountered species are impala and wildebeest (Chapter 5). Rodgers (1979) conducted ground censuses each year between 1969 and 1977 in
22 ▪ C H A P T E R 2 Table 2.2 Estimates of wildlife populationsa and human activities in the greater Selous Ecosystem, 1989–1994b Sept. 1989
June 1991
Sept. 1994
Area Surveyed
73,947
72,658
75,443
Species Buffalo Elephant Hartebeest Impala Wildebeest Zebra
74,222 Ⳳ 13,194 29,597 Ⳳ 4,277 19,760 Ⳳ 2,995 25,065 Ⳳ 3,287 68,141 Ⳳ 6,892 28,420 Ⳳ 4,130
48,721 Ⳳ 13,202 20,290 Ⳳ 2,319 11,988 Ⳳ 1,373 61,248 Ⳳ 6,617 50,886 Ⳳ 2,975 32,095 Ⳳ 7,062
166,464 Ⳳ 15,180 47,860 Ⳳ 4,958 18,829 Ⳳ 2,176 34,398 Ⳳ 3,008 72,785 Ⳳ 7,752 36,418 Ⳳ 1,325
Human Activity Saw Pits Tree Felling Cattle Poacher’s Camp
136 Ⳳ 51 423 Ⳳ 150 11,622 Ⳳ 6,052 185 Ⳳ 66
378 Ⳳ 102 511 Ⳳ 144 5,893 Ⳳ 5,087 59 Ⳳ 30
2,373 Ⳳ 118 1,260 Ⳳ 406 8,166 Ⳳ 4,990 88 Ⳳ 35
Mean Ⳳ SE. Data from sources in Table 2.1.
a
b
eastern Selous, and found that that the most common large mammals were impala (28.6 ind/km2), wildebeest (26.1 ind/km2), zebra (10.7 ind/km2), and warthog (6.4 ind/km2). Human activities around the reserve include fishing the larger rivers, rice farming in floodplains, and pit-sawing hardwoods. Smaller areas are used to grow cassava, maize, millet, and sorghum. Because tsetse flies are abundant and trypanosomiasis is endemic, livestock (cows, sheep, and goats) are rarely kept near the reserve, with the exception of small areas to the west and north. Settlement is heaviest to the north and west of the reserve. In general, communities near the reserve are poorly developed, because they are at dead ends on poor dirt roads. The road from Dar es Salaam to Mloka, on the eastern border of the reserve along the Rufiji River, is a good example. The road is impassible in the rains, and 30km/h is the upper limit under the best of conditions. In most areas only 20–30% of households have access to piped water, in the form of a communal hand-pumped well. Access to health care and schools is below the national standard (Tanzania Department of Wildlife 1995). Population growth rates range from 1.5% to 3.4% per year in districts bordering the reserve (TDW 1995). It is our impression that pit-sawing was intensifying to the east of the reserve between 1991 and 1996, and livestock density was increasing to the north, though we have no systematic data on these activities.
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 23
2.2 The Study Area and Population Our study area covered 2,600 km2 in the north of the reserve (Figure 2.5). The study area was bounded on the south by the Rufiji River and five large lakes (from west to east: Tagalala, Manze, Nzerakera, Siwando, and Mzizimia). On the west, the study site ended at broken, rocky country just west of a line of large hills (from south to north: Kidai, Kipalala, Katumbulwa, Hatumbulwa, and Vianze). To the north, the study area extended north of the reserve boundary to include a floodplain on the south side of the Mgeta river. In the northeast, the site extended just into the western edge of the Nzasa plain, then diagonaled southeast to the eastern edge of the reserve. On the east, the site extended 15 km outside the reserve boundary. In practice, the area in which we worked varied slightly from year to year, depending on the areas that were used by radiocollared packs. We attempted to keep radiocollars on all the packs within the site, and we are confident that all packs in the study area shown in Figure 2.5 were identified, though some packs were not collared for parts of the study.
Population Size and Density The density of wild dogs in northern Selous was high and stable between 1991 and 1996. Total density (including pups) ranged from 1 dog/20.5 km2 to 1 dog/15.8 km2, with a mean of 1 dog/17.4 km2. Excluding pups, density ranged from 1 adult/28.6 km2 to 1 adult/21.8 km2, with a mean of 1 adult/26.0 km2. This is an unusually dense population, though Moremi National Park in Botswana holds a similar density (McNutt 1995). Densities are typically 1 adult per 50–60 km2 in wooded ecosystems, such as Hwange National Park in Zimbabwe or Kruger National Park in South Africa (Fuller et al. 1992a; Maddock & Mills 1994; Wilkinson 1996). Open grasslands in Serengeti National Park in Tanzania held an average of 1 adult/208 km2 over a period of 20 years (Frame et al. 1979; Malcolm 1979, Burrows 1995). Prey density is higher in northern Selous than in the rest of the reserve, so an extrapolation of wild dog density from our study area to the entire 43,600 km2 reserve might overestimate the population’s size. To estimate density in areas outside our study site, we collected sightings and photographs from reserve staff, professional hunters, and photo-tourists, for a period of two years. For most of the reserve, this gave little information, but coverage was good in three areas that ranged from 1,583 to 3,890 km2. In these areas, density ranged from 1 adult/41 km2 to 1 adult/62 km2. Combining data from our study site (2,600 km2) and estimates from the sighting program (7,408 km2), we estimate that the Selous holds 880 adult wild dogs (Creel & Creel 1998). This estimate is for the reserve itself, excluding Mikumi National Park (ca. 90 individuals) or adjacent Game Controlled Areas and Open Hunting Areas (totaling 25,000 km2)
24 ▪ C H A P T E R 2
Figure 2.5 (a) The northern Selous Game Reserve, showing the reserve boundary and major landmarks. Crosshatch ⳱ hill, solid ⳱ lakes, heavy line ⳱ reserve boundary. (b) The study area covered 2,600 km2, roughly bounded on the south by the chain of lakes, on the west by the line of hills, and on the north and east by the reserve boundary. The approximate boundary is shown as a shaded line. The actual boundary shifted slightly each year.
in which wild dogs are often seen. The Selous ecosystem holds a substantial fraction of the continental population of wild dogs—estimated at 5,000 or less—and will be a critical reservoir for the species (Fanshawe et al. 1991). The next largest known population is in Kruger, with 357 to 434 individuals (Maddock & Mills 1994; Wilkinson 1996), though there remain large protected areas where wild dogs are often seen, but population size is unknown (e.g., the 25,000 km2 Ruaha/Rungwa/Kisigo complex in southern Tanzania).
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 25
2.3 General Methods Observations This section describes our general methods. To keep the other chapters reasonably self-contained, the details of some methods are given in those chapters, alongside the results. The study began in June 1991, and ended in October 1996, providing demographic data from six field-years. For demographic analyses, the fieldyear ran from June 1 to May 31, so that age classes changed on June 1 for all dogs. This date falls near the time that the first litters were born each year. We were in the field for 1,200 days, including days spent working on logistic tasks or species other than wild dogs (e.g., lion and hyena tracking,
26 ▪ C H A P T E R 2
Figure 2.6 (a) Individual wild dogs can easily be identified by variation in their black, white, and tan coats. In general, white markings alone distinguish between any pair of individuals, though a few individuals have little or no white. Tail markings can be broken into a few classes that can rule out many individuals when making an identification.
prey censuses). In this time, we were in contact with the dogs on 840 days. “Contact” days include those on which we made counts (e.g., counting new litters), updated life-history information, mapped locations, attached radiocollars, collected fecal samples, or collected anecdotal behavioral data. All of the dogs in the study (366 individuals) could be identified by variation in coat patterns, using identification cards with photographs and drawings (Figure 2.6). On 518 days, we made systematic behavioral observations. These systematic observations were generally made in the morning and evening, when the dogs were most active (chapter 4). We always observed the dogs from a closed vehicle to avoid habituating them to the sight of humans. We typically watched from distances of 20–200 meters, depending on the thickness of the vegetation. At night, we followed the dogs only to maintain contact and to record if kills were made, but generally we could not follow closely enough to make detailed observations. In wooded areas, night vision goggles did not allow us to see well enough to identify individuals, though we could usually see what was happening. We made systematic observations for a total of 3,383 hours, in which the dogs were on the move for 1,450 hours and at rest for 1,933 hours. Most behavioral data were collected in “follows” that lasted several days. All data
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 27
on hunting behavior were collected on follows that were planned to last at least three days. Some follows ended early because we could not follow the dogs or lost the radio signal. We included data from all follows that lasted at least a full day (mean duration ⳱ 4.1 Ⳳ 0.5 days; range ⳱ 1–25 days). Behavioral observations focused on sexual, aggressive, and hunting behavior. For aggressive interactions, we recorded the identity of the initiator, the recipient, the winner, and the loser. For matings we noted the initiator, the recipient, and the duration of the mount. Estrus is synchronized among the females of a pack, but not perfectly. We defined mating periods to include any day on which there was mating within the pack, though specific individuals may or may not have mated. Wild dogs rest at midday and rates of all behaviors drop nearly to zero, so we calculated behavioral rates based on active hours (mean ⳱ 2.8 h/day) when dogs were socially active but not traveling. Most pack members are almost continuously in view under these conditions, so corrections for observation biases were not necessary (Drickamer, 1974). We determined dominance ranks from aggressive interactions by minimizing the number of reversals in a win-loss matrix for each sex in each pack. All aggressive behaviors showed similar patterns, so they were pooled in a single matrix to minimize the number of empty cells. A new matrix was tabulated each time the dominant individual died or was deposed, so that a given individual’s rank could change through the study. Alpha males were replaced as often as three times per year, usually by escalated fights that led to reversals of dominance. Alpha females were replaced less often, usually by being evicted from the pack by immigrants. During follows, we recorded the location of each resting site and kill using a geographic positioning system. We recorded the habitat type at each fix, and kept a running record of changes in habitat type as the dogs moved. We recorded all encounters with prey herds, though we could not always count the number of individuals in a herd, so some analyses are presented on a per-herd basis. Other data on the size and composition of wildebeest and impala herds came from systematic counts that we conducted twice annually when we were not following the dogs. Anesthesia and Radiocollaring We radiocollared two dogs in each pack to let us relocate packs regularly and maintain contact during follows. We used Telonics MOD-400 transmitters with a motion sensor, so that a change in the pulse frequency from 0.8 beat/sec to 1.0 beat/sec indicated that the dog was moving. Motion sensors were particularly helpful when monitoring the signal at night. Transmitters (172–174 KHz) were housed in a stainless-steel casing coated with epoxy, and mounted on a machine belting (CLM) collar. A woven steel antenna with a soldered tip was stitched between two layers of belting. The antenna
28 ▪ C H A P T E R 2
protruded 3–4 inches after surplus belting was cut away. The protruding length of antenna usually fell off after a few months. Surprisingly, this had little effect on the signal’s range, which was usually 1 to 1.5 km. We tracked with Telonics TR-2 and TR-4 receivers and David Clark headsets, using a flexible rubberized H-antenna (Telonics RA-2A) attached to a set of nesting aluminum poles that formed a mast 0.4–3 m high. This could be shortened and pulled in through a window to avoid breaking the antenna on branches. To begin a follow, we located a pack by driving transects with an omnidirectional antenna. Three packs could often be located from hilltops (170–600 m above the surrounding area), using a 5-element Yagi antenna that could detect signals 10–15 km away. We also located packs 4 to 9 times a year by conventional aerial tracking in a light plane. To attach radiocollars, we anesthetized dogs with a dart gun. We used 3 ml plastic Telinject darts fired from a Telinject Vario 3V airgun. This gun is equipped with a pressure gauge scaled in units of pressure (mPa) and distance (m). We sighted the gun so that the pressure was 67% of that suggested by the gauge. We used open sights to provide a view of movements by individuals near the target animal, to minimize the risk of hitting an animal behind the target dog in case of a miss. We preferred to dart males, which are larger and more heavily muscled, when they were standing 10– 15 m away, perpendicular to the flight of the dart. (Darts were more likely to bounce out before injecting their entire contents when the dart struck at an oblique angle.) Adjusting the pressure just before shooting, we fired the dart at the muscle mass of the upper hindleg. For reasons that were not obvious, the femur of one female was broken by a dart. The bone healed well in three months. We initially used 1100–1300 mg of Telazol (Tiletamine-Zolazepam) for anesthesia, but recovery took 3 to 4 hours. To shorten the recovery period, we switched to a combination of 50–55 mg Ketaset (Ketamine HCl) and 58–65 mg of Rompun (Xylazine), reversing the effects of Rompun with 560 µg of Yohimbine injected intravenously. We injected 1 ml of a broadspectrum antibiotic (Liquamycin) intramuscularly, and applied tetracycline ointment to the dart entry wound (which was always small, because wild dogs’ skin is very tough). If the weather was hot, we sometimes gave 500 ml of lactated Ringer’s solution intravenously, and/or poured water on the dog’s flanks and feet. With the ketamine-xylazine combination, darted dogs were apparently unconscious 5–8 minutes after darting, and could reliably be handled after 12–14 minutes. Because the darted dog was usually resting among its packmates in the shade, we drove very slowly toward the pack until the rest of the dogs moved away, then positioned the vehicle so that they could not see the anesthetized dog. If the remainder of the pack moved to another patch of shade, we collared the dog where it lay. If the pack did not move away, one of us held the dog in the passenger seat while we moved to a patch of shade 50–100 m away. The anesthesia lasted a minimum of 45 minutes, sometimes slightly over an hour. Dogs could stand 0.5–5 minutes
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 29 Table 2.3 Measurements of body size for male and female wild dogs Males
Females
Difference
Skeletal Size Back Length Skull Length Toe–Shoulder
81.4 Ⳳ 4.1a 25.6 Ⳳ 1.3 72.0 Ⳳ 3.3
78.1 Ⳳ 2.9 24.4 Ⳳ 0.8 69.9 Ⳳ 2.1
4%, P ⳱ 0.042b 5%, P ⳱ 0.008 3%, P ⳱ 0.045
Musculature Chest Girth Neck Girth
61.7 Ⳳ 3.4 39.2 Ⳳ 2.0
57.6 Ⳳ 1.7 36.8 Ⳳ 1.5
7%, P ⳱ 0.002 7%, P ⳱ 0.008
All measurements are mean Ⳳ SE, in centimeters. P-values are from one-tailed z-tests.
a
b
after reversal, and typically rejoined their packmates within 15 minutes of reversal (1–1.5 hours after being darted). Dogs were capable of moving with the pack at this point, if necessary. Sometimes, a recovering dog would stand and look for its packmates, lay back down after seeing them, and rejoin them later. We avoided darting in the evening or early morning, to minimize the chances of the darted dog getting separated from its pack. Nonetheless, the remainder of the pack moved away while we collared a dog on several occasions. These dogs scent-tracked and hoo-called until the pack answered, and rejoined their packs with little apparent trouble. We drew 5–10 cc of blood from the lateral tarsal vein for genetic analysis and serological screening. These samples were heparinized and allowed to settle, separated, and frozen in liquid nitrogen. We noted general body condition; tooth color, breakage, and wear; and measured the canine teeth with a dial caliper. For males, we measured the testes along three perpendicular axes with dial calipers. For females, we noted the condition of the nipples and vulva. For both sexes, we used a cloth tape to make five measurements of body size. These included toe-shoulder height (from the tip of the middle pair of toes to the dorsal edge of the scapula, with the leg held straight), chest girth (just behind the forelegs), neck girth (midway along the neck from anterior to posterior), head length (from the tip of nose to the pronounced lateral ridge at posterior edge of skull), back length (from the lateral ridge at the posterior edge of skull to the pronounced dorsal projection of the first sacral vertebra), and tail length (from the first sacral vertebra to the end of the flesh). Males were significantly larger than females in every measure (Table 2.3). Wild dogs have previously been described as sexually monomorphic (Malcolm 1979; Girman et al. 1993), though we could find no published data on sexual dimorphism. The sexual dimorphism in body size ranges from 3– 7%, and is easy to see once one is aware of it. We could usually guess a dog’s sex without seeing the genitalia, though we did not rely on body size
30 ▪ C H A P T E R 2
Figure 2.7 Baseline fecal corticosteroid levels in collared and uncollared African wild dogs in Selous. Error bars show one S. E. P values are for one-tailed tests.
to determine sex. Wild dogs are very easy sexed by the large tuft of hair on the penis. tests for effects of handling In 1991, the last few wild dogs on the Serengeti plains went through a disease-driven decline to local extinction. Burrows (1992; Burrows et al. 1994, 1995) hypothesized that this decline may have resulted from immune suppression induced by the stress of handling by researchers. This hypothesis has been extensively debated, in part because no data were collected in Serengeti to determine the effects of handling on stress hormones or immune function, and in part because all of the dogs under study died—handled and unhandled alike. General anesthesia entails risks, and field biologists should take care when darting and radiocollaring. However, darting and radiocollaring had no detectable effect on mortality rates in five wild dog populations, including the Selous (Ginsberg et al. 1995a,b). In the Selous, the glucocorticoid stress hormone levels of collared and uncollared dogs did not differ detectably (Creel et al. 1996; Creel 1997; Figure 2.7). For dogs that had been noninvasively sampled for stress hormone prior to radiocollaring, a matchedpairs t-test detected no change in basal glucocorticoid levels. Rates of aggression received and annual mortality were not detectably affected by radiocollaring (Creel et al. 1996; Creel 1997). In short, our data suggest that anesthesia and radiocollaring do not provoke an immunosuppressive stress response in wild dogs (also see de Villiers et al. 1995; van Heerden et al. 1995). We concur with Burrows that field biologists should carefully consider whether their methods are benign, but, in this case, radiocollaring did not entail unusual risks.
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 31
Endocrine Methods Endocrine data came from radioimmunoassay (RIA) of fecal estrogen, progestin, testosterone, and corticosterone concentrations. We collected 216 samples from 22 females and 34 males over two years. Gus Mills also provided 16 samples from 15 males in Kruger National Park, collected over one month. To validate methods (Monfort et al. 1997) we used longitudinal fecal samples from three males (271 samples) and one female (188 samples) at the Brookfield Zoo, Chicago, provided by Bruce Brewer and Mary Burke. In the wild, we collected fecal samples by watching a known individual defecate and collecting the dropping when the pack moved away. Each dropping was mixed using a wooden stick, then four 1-ml subsamples were combined and frozen in liquid nitrogen in the field. Samples were transported on dry ice and stored at ⳮ80⬚C until assay. We recorded time of defecation, lag until collection, and lag until freezing, but these variables did not affect the concentration of any hormone, so they were not included as covariates in final analyses. Wild dogs normally defecate once a day, and this 24-hour pooling apparently removed diurnal fluctuations in hormone levels. We extracted steroids using published methods (Brown et al. 1994; Wasser et al. 1994; Creel et al. 1996). Samples were dried in a rotary evaporator, pulverized, and 0.1–0.2 g of the resulting powder was boiled 20 minutes in 10 ml of 100% ethanol. After centrifuging at 500 G for 15 minutes, supernatant was recovered, dried under nitrogen, rinsed with 1–2 ml of ethanol, redried, and reconstituted in 1 ml absolute methanol. This extract was vortexed for 1 minute, placed in an ultrasonic glass cleaner to free adhering particles, then vortexed for 15 seconds. Extraction efficiency for corticosteroids was determined by measuring recovery of 3000–5000 decays/minute of 3H-cortisol added to each sample prior to extraction. For female feces, we added 3H-estradiol-17b and 14C-progesterone (3000–5000 decays/min each) to measure extraction efficiency. For males we added 14C-testosterone (3000–5000 decays/min). Hormone concentrations were corrected for extraction loss and expressed as ng or mg of hormone per mg of dry feces. We validated radioimmunoassays for fecal corticosterone, progestins, estrogens and testosterone using standard criteria (Cekan 1975; Reimers et al. 1981). Progestin Assay: We injected a captive female wild dog (who was to be euthanized for medical reasons) with 15mCi 14C-progesterone and collected all urine and feces for 80 h. This trial showed that 60.3% of progesterone was excreted in feces (rather than urine), with peak excretion 18 h after injection. High performance liquid chromatography (HPLC) revealed five progesterone metabolites. HPLC combined with gas chromatography/mass spectrophotometry (GC/MS) suggested that the sole immunoreactive progestin (⬎ 90% of total immunoreactivity) in wild dog feces was prenanolone (the immediate precursor of pregnanediol; Feder 1981). We assayed fecal progestins using a previously described 125I RIA that
32 ▪ C H A P T E R 2
cross-reacts with several progestins (Brown et al. 1994; Wasser et al. 1994). Fecal extracts were diluted 1:8,000 and assayed in duplicate 100ml aliquots. Serially diluted extract gave an antibody displacement curve parallel to serially diluted progesterone standard. We tested accuracy by adding known quantities of progesterone (3.75–240 pg/tube) to fecal extracts and measuring the amount recovered, which averaged 95.0% Ⳳ 13.4% (r2 ⳱ 0.94). Assay sensitivity was 3.75 pg/tube. The interassay coefficient of variation was 6.4% for a high-progestin control, and 15.5% for a low-progestin control. Estrogen Assay: As described for progestins, we injected an adult female wild dog with 90mCi 3H-estradiol to determine what estrogen metabolites were excreted in feces (Monfort et al. 1997). The majority of estradiol (58.9%) was excreted in feces, with peak excretion 18 h after injection. HPLC and GC/MS showed that the two major immunoreactive fecal metabolites were unconjugated estrone and estradiol-17, which combined to account for 60% of total immunoreactivity. Three other metabolites were immunoreactive, so we consider the assay a measure of total estrogens. We used a double-antibody 125I RIA (ICN Biomedicals) for total estrogens following the protocol provided (with the exception of halving all volumes). We diluted fecal extracts 1:500 and assayed duplicate 250 ml aliquots. The antibody displacement curves of serially diluted fecal extract and serially diluted estradiol-17 were parallel. Mean recovery of estradiol-17 (2.5–100 pg/ml) added to fecal extracts was 121.5% Ⳳ 11.1% (r2 ⳱ 0.99). Sensitivity was 2.5 pg/ml. The interassay coefficient of variation was 6.7%. Testosterone Assay: We used a double-antibody 125I RIA (Diagnostic Systems Laboratories), according to the protocol provided (with the exception of halving all volumes). Fecal extracts were diluted 1:10 and assayed in duplicate 25ml aliquots. Serially diluted fecal extracts showed an antibody displacement curve parallel to that of serially diluted testosterone standard. Recovery of testosterone (0.05–12.5 ng/ml) added to fecal extract was 99.2% Ⳳ 9.1% (r2 ⳱ 0.99). Sensitivity was 0.1 ng/ml. The interassay coefficient of variation was 14.2% for a low-testosterone control and 11.1% for a high-testosterone control. HPLC showed that testosterone was the only immunoreactive androgen in wild dog feces. Corticosterone Assay: We validated a double-antibody 125I-corticosterone radioimmunoassay (ICN Biomedicals) for wild dog feces. This antiserum cross-reacted little with steroids other than corticosterone (deoxycorticosterone 0.34%; testosterone 0.10%; cortisol 0.05%; progesterone 0.02%). Serial dilution of wild dog feces (1:32–1:512) yielded an antibody displacement curve parallel to a serial dilution of corticosterone standards, demonstrating specificity. Accuracy was tested by measuring the recovery of known amounts of corticosterone added to wild dog feces (25–500 ng corticosterone/ml). Recovery was 102.6% Ⳳ 4.5% (ng measured ⳱ ⳮ1.78 Ⳮ 1.02 * ng added, r2 ⳱ 0.99). Assay sensitivity was 25 ng corticosterone/
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 33
ml extract. The interassay coefficient of variation was 8.0% for a highcorticosterone control and 13.7% for a low-corticosterone control; the intraassay coefficient of variation was less than 5%. Radioimmunoassay of fractions from high performance liquid chromatography (Monfort et al. 1990) showed a single peak of anticorticosterone immunoreactivity in wild dog feces with polarity intermediate between cortisol and corticosterone. Finally, we tested the response of fecal corticosteroids to a challenge with adrenocorticotrophic hormone (ACTH) in three female and two male wild dogs at the Brookfield Zoo (Monfort et al. 1997). Fecal samples were collected from 72 hours before to 144 hours after intramuscular injection of 800 µg of Acthar gel, a long-acting ACTH preparation. For all five dogs, fecal corticosterone concentrations increased significantly during the first 24 hours after injection, with peaks 10-fold to 25-fold above baseline. Corticosterone levels returned to baseline by 48 hours post-injection. Spotted Hyenas and Lions Hyenas. We determined spotted hyena density for our 2,600 km2 study site using highly amplified playbacks (N ⳱ 37) of a tape of noises known to attract spotted hyenas, following Smuts et al. (1977), Whateley & Brooks (1978), and Mills (1985). The tape (provided by Gus Mills) included whooping, the sounds of hyenas squabbling on a kill, dying wildebeest, a fight between two hyena clans, and hyenas mobbing lions. For playbacks we used a Sony TCM-5000EV recorder wired in series to two Anchor Audio PB-25 20-watt speakers with 15-watt amplifiers. These speakers produced sound pressure levels of 103 dB at 1 meter. Playbacks were clearly audible to human ears at 1.5 km. We conducted playbacks on moonlit nights or in the predawn at temperatures below 31⬚C. Playback sites were distributed across the study area using a geographic positioning system, locally sited in areas with good visibility. The five-minute tape was played twice, followed by a five-minute pause, then played twice again. Approaching hyenas were spotted by two to four observers standing on a Land Rover using binoculars, image-intensifying goggles, and high-intensity flashlights. The approaching hyenas were clearly motivated to locate this disturbance, usually arriving at a run. Thus, we consider it unlikely that they were dissuaded from approaching by the sight of people, particularly because hyenas are not timid after dark. Observations were made for 30 minutes after the playback began. Most hyenas remained at the playback site until the playback ended, so we recorded the maximum number of hyenas simultaneously in view; in some cases, additional individuals could be tallied due to nearly simultaneous arrivals/departures from different directions with good visibility. The area censused by playback was determined by response trials, in which we located a hyena before a playback. One observer stayed with the
34 ▪ C H A P T E R 2
hyena and noted its response to a playback at a known distance (measured by odometer or GPS). We recorded responses for 25 individuals in 10 trials. Comparisons with response distances in other hyena studies were used to corroborate these data (Kruuk 1972; Mills 1985; Sillero-Zubiri & Gottelli 1992). We determined spotted hyenas’ diet in Selous by opportunistic observations of feeding. This method does not distinguish between killed and scavenged food, and is likely to overestimate the importance of large prey relative to small prey (Kruuk 1972; Mills 1992). Some of these observations were made by tracking three radiocollared hyenas. We made repeated counts at two communal dens to estimate minimum clan size. Lions. We estimated the density of lions on our study area in two ways. A direct estimate came from counting individually recognized individuals from nine prides, for which we maintained a photographic identification file. We identified individuals on the basis of nose color, whisker spots, broken or worn teeth, and permanent scars. These prides (46 adults, 21 cubs in August 1996) occupied contiguous home ranges covering a total of 350 km2 at the south of the study area, a density of 0.13 adults/km2. We radiocollared one lioness in five of the prides to determine habitat use and diet. As with our data on foraging by hyenas, these opportunistic observations probably overestimate the importance of large prey species. A second lion density estimate came from the ratio of adult lions to adult spotted hyenas. We recorded all lions and spotted hyenas that we saw from June 1991 to March 1993, and from June 1994 to September 1995 (1,035 sightings). The density of adult hyenas was determined by the playback experiments described above. We then estimated the density of adult lions by multiplying the ratio of lions to hyenas by the density of hyenas (Caughley 1977). In these observations, we tallied 273 lions and 762 hyenas, giving a ratio of 0.36 lion:1 hyena. Multiplying this proportion by the density of hyenas from audiotape playbacks (0.3 adult hyenas/km2) gave an estimate of 0.11 adult lions/km2 (Creel & Creel 1996). This “sightings ratio” method assumes that lions and hyenas are equally visible. This assumption seems reasonable for northern Selous, based on our radiotracking of both species. The assumption was further supported by data from our intensive study area of 350 km2 in which all lions were individually recognizable. This area had a density of 0.13 adult lions/km2 and a density of 0.31 adult hyenas/km2 (as estimated by eight hyena-census playbacks within the intensive lion study area). Within this area, the ratio of the two species’ densities (0.42) was similar to the ratio of opportunistic sightings (0.36), suggesting that the two species are similarly visible in northern Selous. Finally, we compared our lion population density estimates with a previous estimate for eastern Selous (0.08 adult lions/km2) (Rodgers 1974) for a total of three estimates that ranged from 0.08 to 0.13 adults/km2. For lions of unknown age, we estimated age on the basis of body size
T H E S E L O U S , S T U D Y P O P U L AT I O N , M E T H O D S ▪ 35
until two years, and afterwards using nose color. The nose of young lions is entirely pink and becomes increasingly black until it is entirely black at approximately 8–10 years of age and beyond. Age and nose color were related using information from known-age lions in Serengeti, as follows: black speckling ⳱ 2–4 yr; 25% black (mottled) ⳱ 4–5 yr; 50% black (splotched) ⳱ 5–8 yr; 75% black (livered) ⳱ 8–10 yr (Craig Packer, personal communication 1991). We determined the lion sex ratio in two ways. One estimate used all sightings of lions that could be sexed, whether they were individually known or not. This provided a large random sample, but some individuals were undoubtedly tallied more than once. This method assumes that males and females are equally likely to be seen. The second measure used only known individuals, which provided a smaller sample, but did not rely on the assumption that males and females are equally likely to be seen. Our observations suggest that hunted males are more secretive than unhunted females.
3
Home Ranges and Habitat Selection
Wild dogs live at low population densities, even under the best of conditions, and this is a fundamental reason that they are endangered. Habitat loss is clearly a problem for wild dogs, but this problem should also affect other large carnivores. In general, ecosystems that hold thousands of lions and spotted hyenas hold a few hundred wild dogs at best, and sometimes hold no wild dogs at all (Creel & Creel 1996). Parallel to their low population density, wild dogs have large home ranges (for example, home ranges have been estimated at 1,500–2,000 km2 in Serengeti; Frame et al. 1979). All else equal, large home ranges will be associated with low population density (though variation in group size or range overlap could offset the association). Consequently, identifying factors that affect home range size is a step toward understanding why wild dogs always live at low density and thus are endangered. In this chapter, we present descriptive data on home ranges and habitat use of wild dogs in Selous. We then test whether home range sizes, shapes, or overlaps are related to pack size, prey availability, or habitat type. We compare home range sizes in Selous to those of other wild dog populations, testing for an association between home range size and mean pack size across populations. We compare the home ranges of wild dogs to those of other medium-sized and large canids, asking whether wild dog home ranges are unusually large for their body mass and group size. Finally, we compare wild dog ranges to those of sympatric large African carnivores, looking for ecological and social correlates of home range size. We use these analyses to consider why wild dog ranges are large, relating our considerations to variation among ecosystems in wild dog density.
3.1 Specific Methods Home Range Mapping We used a satellite geographic positioning system (GPS) to obtain latitude– longitude fixes while following wild dog packs. The GPS fixes were accurate to within a radius of 100 meters without base-station correction, and this is a small error relative to the daily movements of wild dogs (12.3 Ⳳ SE 0.5 kilometers per day; Creel & Creel 1995b; see Chapter 5). Rather than using a fixed-time interval between fixes, we recorded one fix at each kill site, day resting site, or night resting site, so that the distribution of points within a
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 37
home range had a straightforward biological meaning—consecutive points are separated by one complete period of movement. For 11 packs monitored over periods ranging from one month to five years, we tallied 1,540 GPS fixes. We used a biweight kernel density estimator to produce a space-use distribution (utilization distribution, or UD) for each pack. Using the adaptive kernel module of the program CALHOME (Kie et al. 1994), optimal bandwidth was set for each distribution, using least squares cross validation (Seaman & Powell, 1996). From CALHOME, we exported contours that defined the smallest area with 50%, 75%, 85%, and 95% of the space-use distribution for each pack (Figure 3.1). We operationally defined a home range as the area enclosed by the 95% contour. Wild dogs occasionally made brief forays into areas that they rarely, if ever, visited on other occasions. By excluding 5% of the utilization distribution, we dropped these forays from the home ranges. We defined the core of a home range as the area enclosed by the 50% contour. Space-use contours were imported into a geographic information system (Idrisi; Eastman 1997). Kernel density estimators determine contours based only on the distribution of points, so a set of fixes next to an uncrossable body of water is likely to produce a contour that extends over the water. Consequently we modified the shapes of several home ranges to exclude areas that could not be used (lakes and areas south of the Rufiji river). We did this in Idrisi, overlaying the home range contours onto a base map of the study area and cutting out the unusable areas by manual onscreen digitizing. We examined the relationship between home range size and number of fixes, and found that a minimum of 40 fixes were required for a stable home range estimate. For 4 of the 11 packs, we collected more than 40 fixes per year in some years. We determined eight annual home ranges for these 4 packs, based on a mean of 124 fixes per pack per year. For the remaining 7 packs, partitioning the fixes by year gave some years with fewer than 40 fixes, so we determined a single home range. These multiyear home ranges were based on periods averaging two years (range 0.5–3 years). Using Idrisi, we determined the geographic center of each range, the area within each contour, the proportion of each contour that was not overlapped by any other pack (exclusive area), and the proportion of each contour that overlapped each contour of each neighboring pack (with four contours determined per home range, this yields 16 overlaps per pair of ranges). We also determined the compactness ratio for each contour. This is a measure of the area of a polygon relative to the length of its perimeter, defined as: Compactness Ratio of polygon x ⳱
√
Area of polygon x Area of circle with same perimeter as polygon x
38 ▪ C H A P T E R 3
Figure 3.1 A representative annual home range, for the Wengi pack in the 1992 census year (136 fixes). The contour lines come from an adaptive kernel space-use distribution. The outer contour holds 95% of the space-use distribution. The inner contour holds 50%, and the intermediate contours hold 75% and 85%.
The compactness ratio varies from 0 to 1. A small value indicates that a home range has a complex, irregular shape, with a long border for its area. Habitat Availability, Use, and Preference We recorded the vegetation type at 1,540 locations throughout the study site, fixed by GPS. For mapping, we grouped habitat types into six classes: deciduous woodland (miombo, chipya and Combretum), thorn woodland (Terminalia spinosa, whistling thorn, Acacia nigrescens), short grass (⬍ 20 cm, but typically ⬍ 10 cm), long grass (⬎ 20 cm, but typically 1–4 meters),
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 39
riverine thickets, and palm swamp. Habitat types are described in Chapter 2. Using Idrisi, we plotted the fixes for each vegetation type on a base map of the study area, and hand-digitized boundaries for each vegetation type to create a GIS layer. Because the true boundaries between habitats have highly irregular shapes, dependent on factors such as topography and soil type, we found that hand-digitized boundaries gave a better match to reality than the interpolation routines available in Idrisi. With this vegetation map (Figure 3.2), we used Idrisi’s cross-tabulation module to determine habitat availability, defined as the proportion of each home range covered by each vegetation type. We then calculated habitat use as the proportion of fixes for each pack that fell in each habitat. Finally, we calculated habitat preference as the ratio of habitat use to habitat availability. A preference ratio of 1 indicates that the wild dogs use a habitat in proportion to its availability. A preference ratio greater than 1 indicates that a habitat is favored. We did not calculate preference ratios for palm swamp and riverine thicket because they covered very small areas, so the preference ratio was very sensitive to the addition or deletion of a small number of fixes. Prey Density and Encounter Rates We recorded encounters between wild dogs and ungulate prey while we followed wild dogs that were traveling or hunting. These data were tabulated as herds encountered per kilometer traveled. We tallied herd encounter rates for wildebeest, impala, and all prey species combined (wildebeest and impala made up 78% of all kills; see Chapter 4). Habitat type was noted for all herds, so herd encounter rates could be tested for differences among habitats. Because herd sizes and compositions were difficult to record when we were following wild dogs, we also made systematic herd counts for wildebeest and impala. Twice annually (once in the wet season and once in the dry), we drove census transects to record the number of adults, subadults, and juveniles of both sexes in wildebeest and impala herds (N ⳱ 660 counts). We used these censuses to describe variation in herd size across habitats and seasons.
3.2 Description of Home Ranges Area For eleven packs, multiyear home ranges (95% contour) averaged 433 km2 (Ⳳ SE 66), with a minimum of 156 km2 and a maximum of 846 km2. Core areas (50% contour) for these ranges averaged 64 km2 (Ⳳ SE 10), with a minimum of 10 km2 and a maximum of 124 km2. Eight annual home ranges were slightly smaller than the multiyear ranges, though the difference was
40 ▪ C H A P T E R 3
Figure 3.2 Vegetation map for Northern Selous.
not significant (t17 ⳱ 0.54, P ⳱ 0.60, β ⳱ 0.70 for α ⳱ 0.10). Annual home ranges averaged 379 km2 (Ⳳ SE 74), with a minimum of 206 km2 and a maximum of 851 km2. The cores of annual ranges averaged 52 km2 (Ⳳ SE 14), with a minimum of 4 km2 and a maximum of 117 km2. Wild dog packs did not visit all parts of their range equally often. Packs
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 41
spent more time near the center of their range than at the edges. On average, the contour holding 50% of the fixes covered only 14% Ⳳ 1% of the home range area, so the core of a range received four times more usage than would be expected if locations were uniformly distributed within a range. Similar patterns, though less pronounced, held for the 75% and 85% contours of the utilization distribution. The contour holding 75% of the fixes covered 36% Ⳳ 4% of the home range area, approximately double the usage expected with a uniform distribution. The contour holding 85% of the fixes covered 55% Ⳳ 4% of the home range area. Thus, the intensity of use decreased substantially from the core of a range to its periphery. This has important implications for understanding the substantial overlap that can be seen in the ranges of wild dogs if measured by 95% minimum convex polygons (Reich 1981; McNutt 1995; Mills & Gorman 1997). Much of the overlap zone is rarely used by either pack, so the likelihood of direct encounters is rather low. We discuss overlaps in more detail below. The data used in the analyses just described include fixes from denning and nondenning periods, to give a view of space use over an entire year, responding to seasonal changes in prey distribution, vegetation height, and the availability of water. With this approach, the core of a home range will tend to be associated with the den site, but the two are not synonymous. Wild dogs occupy their dens for a period of 2–3 months while the pups cannot travel with the pack. For 80% of the year, the pack is not tied to the den site, so 80% of the fixes used in a typical home range estimate are independent of the den’s location. Table 3.1 summarizes data describing home ranges. Because multiyear and annual home ranges did not differ significantly in size, they were pooled in subsequent analyses. Shape Home ranges had a wide variety of shapes, examples of which are shown in Figures 3.1, 3.3, and 3.5. Inner contours were more compact than outer contours. Recalling that low compactness ratios indicate large deviations from a circular shape, the compactness ratio (CR) of contours were as follows: 95% contour CR ⳱ 0.24 Ⳳ 0.02, 85% contour CR ⳱ 0.24 Ⳳ 0.02, 75% contour CR ⳱ 0.31 Ⳳ 0.03, 50% contour CR ⳱ 0.62 Ⳳ 0.03.
3.3 Exclusive Areas, Overlaps, and Territorial Defense Reich (1981) noted that the home ranges of wild dog packs in Kruger National Park generally overlapped by 30% to 35%. Mills & Gorman (1997) plotted the ranges of eight packs in Kruger, revealing substantial overlap for some neighboring packs, and neatly disjunct borders for others. Both studies
Year
All All All All All All All All All All All 92 93 92 93 93 95 92 93
Pack
WA WE MP BO MA WI NY ND KU JA KI WA WA WE WE MP MP BO BO
606 293 183 146 48 16 23 48 26 83 64 433 88 127 136 79 38 48 44
Fixes
846 531 298 419 393 267 402 796 156 222 437 851 206 496 249 322 291 348 271
95% 484 317 239 253 245 92 274 259 43 149 298 508 99 391 134 151 126 227 79
85% 289 212 178 168 108 26 179 182 33 99 159 328 46 172 51 131 95 170 16
75%
Area
104 86 55 124 51 10 63 80 25 44 63 117 24 105 23 49 45 52 4
50% 0.21 0.17 0.16 0.22 0.20 0.23 0.19 0.32 0.52 0.18 0.22 0.21 0.24 0.11 0.25 0.30 0.31 0.21 0.30
95% 0.54 0.58 0.64 0.55 0.71 0.74 0.66 0.77 0.56 0.45 0.71 0.54 0.76 0.43 0.59 0.55 0.55 0.47 0.99
50%
Compactness Ratio
0.71 0.19 0.23 0.16 0.23 0.33 0.28 0.64 0.72 0.03 0.40 0.59 0.56 0.20 0.35 0.38 0.46 0.09 0.64
95% 0.75 0.23 0.23 0.19 0.33 0.74 0.37 0.50 0.87 0.02 0.43 0.58 0.49 0.24 0.46 0.50 0.70 0.11 0.46
85% 0.71 0.28 0.20 0.17 0.50 1.00 0.45 0.40 0.89 0.02 0.40 0.46 0.30 0.28 0.34 0.51 0.67 0.12 0.07
75%
Proportion Exclusive
Table 3.1 Properties of wild dog home ranges in Selous
0.86 0.38 0.29 0.04 0.59 1.00 0.81 0.00 0.93 0.00 0.37 0.61 0.20 0.37 0.13 0.63 0.59 0.15 0.00
50% 0.57 0.60 0.80 0.60 0.62 0.35 0.68 0.33 0.28 0.67 0.68 0.60 0.48 0.79 0.54 0.48 0.43 0.65 0.29
85% 0.34 0.40 0.60 0.40 0.27 0.10 0.45 0.23 0.21 0.45 0.36 0.39 0.22 0.65 0.21 0.41 0.33 0.49 0.06
75% 0.12 0.16 0.19 0.30 0.13 0.04 0.16 0.10 0.16 0.20 0.15 0.14 0.11 0.21 0.09 0.16 0.15 0.15 0.02
50%
Proportion of 95% Contour
8 9 12 5 12 7 8 17 5 8 9 5 7 10 12 10 16 5 5
Adult 19 33 24 7 21 10 11 38 11 21 14 13 17 44 43 19 32 5 14
Total
Pack Size
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 43
Figure 3.3 Overlap between the home range of Border pack and neighboring packs. Almost the entire range is overlapped by other packs. 1992 ranges are plotted for Border, Wachunga, and Wengi packs. Multiyear range is plotted for Kilima packs.
noted that home ranges generally had a central area that excluded other packs (usually along a watercourse, according to Reich). This pattern is predicted if wild dogs exclude other packs from their ranges with strong defense at the core, but tolerate intrusion or defend less effectively at the periphery. In Selous, 38% of each home range excluded all other packs. Intriguingly and surprisingly, inner contours did not have a higher proportion of exclusive area than outer contours. The core contour had an almost identical proportion exclusive, 39% Ⳳ 7% (range: 0%–93%) as the outer contour, 38% Ⳳ 5% (range: 3%–72%). The 85% and 75% contours had similar values (43% Ⳳ 5% and 38% Ⳳ 5% exclusive). None of these percentages
44 ▪ C H A P T E R 3 Table 3.2 Overlaps between pairs of home range contours for neighboring packs. Values in the table body give the mean proportion of the focal range overlapped by the neighboring range, with the standard error in parentheses Neighboring Range Focal Range 95% 85% 75% 50%
95% 0.22 0.16 0.12 0.09
(0.03) (0.02) (0.01) (0.01)
85% 0.16 0.10 0.08 0.05
(0.02) (0.02) (0.01) (0.005)
75% 0.12 0.08 0.05 0.02
(0.01) (0.01) (0.01) (0.02)
50% 0.09 0.05 0.02 0.005
(0.01) (0.005) (0.02) (0.003)
differ significantly (95% vs. 50%: t ⳱ 0.09, df ⳱ 28, P ⳱ 0.93, β ⳱ 0.05 for α ⳱ 0.05 and 10% effect size). At first glance, these data suggest that the core of a range is no more exclusive than its periphery, but that conclusion is not correct. Some areas of overlap are heavily used by both packs, while other overlap zones are rarely used by either pack. In other words, being overlapped by the 95% contour of an adjacent pack indicates that the adjacent pack is rarely in the overlap zone, but being overlapped by the 50% contour indicates a greater impact. For a detailed look at patterns of overlap, we used data for 55 pairs of adjacent ranges to calculate the proportion of each contour that was overlapped by each contour of the neighboring range. This analysis (Table 3.2) shows that overlap between areas of intense use was rare, while overlap between areas of infrequent use was common. (Figure 3.4; Friedman’s test: χ2 ⳱ 11.45, P ⳱ 0.009). For example, the mean overlap between neighboring packs’ inner contours was less than 1%, while the mean overlap of outer contours was 22% (Table 3.2). Territorial behavior generally produces an efficient separation of neighboring packs of wild dogs (Mills & Gorman 1997). Nonetheless, overlap can be substantial in some cases, particularly for new packs. It should be noted that our measures of overlap between ranges are underestimates. Although we excluded ranges based on fewer than 40 fixes, it is virtually certain that some ranges included areas that we did not detect. On the other hand, spatial overlap can be offset by temporal avoidance. In Kruger, direct encounters between packs are rare because neighboring packs rarely used an overlap zone simultaneously (Mills & Gorman 1997). In Selous, Border pack illustrates an extreme case of overlap (Figure 3.3). Previously unknown immigrants formed this pack by squeezing between existing packs. Border pack persisted for at least five years (through the end of the study) though only 3% of their home range was exclusive. The range’s core was overlapped completely by other ranges, but none of the core overlapped with other cores. The maximum overlap between cores of ranges was
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 45
Figure 3.4 Overlaps between pairs of neighboring home ranges. The x and y axes give contour values for the focal and neighboring packs, and the z-axis gives the proportion of the focal pack’s contour, overlapped by the neighboring pack’s contour. Overlap between pairs of outer contours is common (at the back of the figure). Overlap between pairs of inner contours is rare (at the front).
for Jangili pack (Figure 3.5), which shared 12% of its core with Wengi pack’s core. Jangili pack was formed when females from Wengi pack joined males from Kilima pack, which bordered Wengi’s home range to the west. Jangili pack established a home range between the two packs from which its members emigrated (Figure 3.5). A portion of Jangili pack’s range was subsequently used by six packs, more than any other area on the study site (Figure 3.6). Large areas of the study site were used by two or more packs (Figure 3.6). Of the area used by at least one pack, 57% was used by two packs, 26% was used by three packs, 10% was used by four packs, 5% was used by five packs, and 0.2% was used by six packs. Formal analyses of habitat preference are given below, but comparing this distribution with the vegetation map (Figure 3.2) shows that areas with zero or one pack were likely to be long grass, and areas of heavy overlap were likely to be woodland. Behavior during Territorial Encounters Direct encounters between packs were rare, but they were more common than Mills & Gorman (1997) observed in Kruger National Park, where wild
46 ▪ C H A P T E R 3
Figure 3.5 Overlap between Jangili pack and the packs from which Jangili’s founders emigrated. Founding females came from Wengi pack; founding males came from Kilima pack. The core areas of Jangili and Wengi packs’ home ranges had the greatest overlap (12%) observed.
dog density was lower. Wild dogs show considerable interest in the scentmarks of other packs, sniffing them repeatedly, defecating and urinating over them, and sometimes tracking the scent trail for kilometers. Small packs occasionally heard or smelled a neighboring pack nearby and quickly moved off to avoid a direct encounter. Large packs generally followed the scents or sounds of other packs, apparently seeking a direct encounter. We observed 13 direct encounters between packs during 518 observation days, or 1 encounter per 40 days (nine encounters per year). If two packs clearly saw one another, a chase or fight always ensued. In 11 of 13 cases, the larger pack pursued or attacked the smaller pack, which fled. The re-
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 47
Figure 3.6 Areas used by more than one pack, showing the number of overlapping home ranges in each area.
maining two cases involved a pair of equal-sized packs that met twice in four days; each pack fled once. Behavior during interpack encounters was aggressive, with most aggression directed at dogs of the same sex. For example, two attacking females once ran directly past five males of a neighboring pack (to whom they were related and familiar) in pursuit of the pack’s alpha female, and the males did not interfere. During clashes, both packs (particularly the retreating pack) made a warbling alarm bark that we never heard in other circumstances. Fights causing injury occurred in 5 of the 13 clashes. We saw serious bite wounds from encounters with other packs on seven females and no males, suggesting that interpack aggression among females may be more serious than aggression among males (χ2 ⳱ 7.0, P ⳱ 0.008). Two fights led to fatalities. An adult female was deeply bitten in the hindlegs and soon died because she could
48 ▪ C H A P T E R 3
not stay with her pack. In the second case, two packs of nine adults clashed. One pack had eight pups (six months old), the other pack had no pups. Five pups and their mother disappeared during the chase. The mother and three of the missing pups were never seen again, but two of the missing pups rejoined the pack two days after the clash. two case studies of interpack encounters Case one: Border pack (3 adult males, 1 adult female) was feeding at an impala carcass when they detected Wachunga pack (4 adult and 2 yearling males, 1 adult and 2 yearling females) within 200 meters, apparently by smell. Border pack fled 4.8 km, abandoning the carcass. These packs encountered each other again 10 days later, and we simultaneously followed both packs. On Day 1 of this encounter, the two packs were resting 1.4 kilometers apart. At 18:58, Wachunga pack encountered the scent of Border pack (which was traveling) and tracked them 2.2 km in 10 minutes. The packs were within 100 meters (and aware of this) at 19:08, and fighting broke out at 19:16. By 19:18, the signals of the two radiocollared Border males began moving away, and were 1–1.5 kilometers away by 19:22. From 19:32 to 20:30, Wachunga pack moved 2.1 kilometers along the line taken by Border pack, hoo-calling often. Both packs were scattered, and hoo calls came from several directions. By 20:30, both packs had apparently regrouped and were separated by 4.5 km, but we could not identify individuals. At 05:45 on Day 2, the two yearling females from Wachunga pack were with the Border males, and this new pack composition was stable over the next few years. The single adult female in Wachunga pack (Clara) was with her pack, badly injured in the hindlegs. One of her hindlegs was torn at the knee fascia and could not bear weight. She was cut to the bone in two places (6 to 8 inches long) on the other hindleg (Figure 3.7). The single adult female from Border pack (Milo) was missing. Both packs moved little during Day 2 until the evening. At 18:27, Border pack heard hoo calls from Wachunga pack, 1.5 km away, and moved off in the opposite direction. At 20:00 they came to rest 5.2 km away. At 02:30 on Day 3 Border pack moved again, following the same direction as on Day 2 and coming to rest 14.3 km away at 07:42. Wachunga pack moved little during this time. The hoo calls at 18:27 of Day 2 came from the Wachunga males calling to their alpha female (Clara), who had fallen behind during a hunt. Her injured knee did not allow her to stay with the group as it hunted, and she did not arrive at any kills in time to eat. She did not beg food from the males. On Days 3 and 4, Border pack moved more than 20 kilometers to the southwest, out of the Wachunga pack range, while Wachunga pack moved little. On Day 4 at 19:28, Milo (the former alpha female of Border pack) began shadowing Wachunga pack, hoo-calling often. She initially stayed about 200 meters from the others, but by Day 5 was lying with the males and avoiding Clara, who tried to chase her but could not. By Day 7, the Wachunga males were
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 49
Figure 3.7 Injuries sustained in an interpack clash. This female could not remain with her pack during travel and hunts, and soon died.
interacting often with Milo and hooing less often for Clara, who still had not eaten and was often far from the pack. She was not seen again, and Milo became the alpha female of Wachunga pack. Case two: The same two packs encountered each other the following year, and the attacking pack (Wachunga) showed little aggression. The dens of Wachunga pack (4 adult and 2 yearling males, 1 adult female) and Border pack (3 adult males, 2 adult females) were separated by 6.6 kilometers. Both packs had pups less than 2 months old, still in the den. While hunting, the six males of Wachunga pack encountered fresh scent from Border pack, and tracked the scent 3.7 kilometers in 40 minutes, directly to the Border pack den. They rallied with loud yittering hoo calls, 150 meters from the Border den. The three Border males ran out to investigate, and both Border females went to the den hole. The alpha female entered the den and the beta female stood on the apron. The Wachunga males charged and the Border males fled, leading to several short pursuits that lasted 10 minutes. No dogs were injured. After 10 minutes, the Border pack males fled 2.4 kilometers and did not return for 1.5 hours. When the Border males fled, the Wachunga males went to the Border den hole, where the two Border females remained. These females were familiar to all the Wachunga males (the daughters of one), having dispersed the previous year (in the encounter described just above). The alpha female stayed underground, growling when approached. The beta
50 ▪ C H A P T E R 3
female could not enter the den, but remained in the area. The Wachunga males chased her half-heartedly several times over the next 29 minutes, but she remained within 50 meters of the den, ruff-barking often, and returned to the den hole several times. The Wachunga males patrolled the den area, rolling, defecating, and urinating. They left after 39 minutes and went directly back to their den. The beta female left at the same time, to join the Border males 2.4 km away. The alpha female never left the den. During this interaction, the Wachunga males made no serious effort to attack the females or pups. Perhaps the normal aggression of interpack clashes was inhibited by the ties of familiarity and relatedness between these individuals. Perhaps aggression is muted when the dogs from each pack are of the opposite sex. Perhaps aggression is inhibited when young pups are involved. McNutt (1995) reported one clear-cut case in which pups were adopted by males from another pack, even though the pups were not accompanied by adult females that could be immediate mates.
3.4 Den Locations and Characteristics Wild dogs rarely spend two consecutive nights in the same location, except when they are denning. In Selous, pups are born early in the dry season, between June and August. The pups are confined to a den for about three months, and the adults return after each hunt (Kuhme 1965; Malcolm & Marten 1982). There are occasional exceptions—a pack that has moved far (ca. 10 km) or failed to kill during a morning hunt may not return until after the evening hunt. The dominant female usually remains at the den during hunts and obtains her food by begging from returning hunters. Again there are exceptions—alpha females sometimes join the hunt and leave the den unattended, particularly in small packs. Reich (1981) noted that wild dog home ranges in Kruger National Park usually had a core area that did not overlap with any other ranges, and suggested that these areas were used for denning. In Selous, we located 22 dens. This number includes only one den site per pack-year, disregarding small den moves. Most dens were moved once or twice during a denning period, from a few meters to a few hundred meters. Two dens were moved more than a kilometer (maximum of 1.2 kilometers). The reasons for den moves were not obvious, though Kuhme (1965) observed a pack of wild dogs in Serengeti move their den after lions investigated the burrow. Dens might also be moved if some of the pups die, but we have no direct data to support this idea. Fifteen of 22 dens (68%) were located within the core contour of their range, though cores occupied only 14% of the average home range. This is a highly nonrandom pattern (z ⳱ 2.59, P ⳱ 0.010). Nineteen dens (86%) were within the 75% contour (36% of the average home range), and all 22
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 51
dens fell within the 85% contour (55% of the average home range). Overall, the proportion of dens located within inner contours was greater than expected if dens had been located randomly (χ2 ⳱ 56.2, df ⳱ 1, P ⬍ 0.001). Reich (1981) noted that dens in Kruger were usually located along river courses. In Selous, dens were rarely along river courses. Most dens were located in dense Combretum woodland or in thickets associated with water holes. All the dens were in sandy areas, using holes originally excavated by other species and re-excavated by the alpha female (wild dogs are powerful diggers in loose soil). A few dens were in open miombo woodland, but even these were locally sited in areas of thick vegetation. Many dens were in sites thick enough that it took days to locate the den hole, though we knew it was nearby. Dens might be located in thickets to avoid encounters with other carnivores, because most animals circumvent thickets in their travels. In 161 days of den observations, we saw lions locate only one den. The pups were scattered, and three pups were killed. As described above, one den was located by a neighboring pack of wild dogs, but the pups were not attacked. We only saw one instance of a spotted hyena approaching a wild dog den, and this hyena was attacked and fled while still 50 meters from the den itself. Collectively, the daily probability that a den would be approached by lions, hyenas, or wild dogs from another pack was 0.019. It seems surprising that carnivores did not approach dens more often, because they develop a strong smell from droppings, cached meat, and bones. Rallies at dens (as elsewhere) often include loud yittering. On the other hand, wild dogs rarely use their loudest call (the “hoo”) when they are denning. If dogs are separated during a hunt in the denning period, they return to the den independently, rather than hoocalling to one another. Perhaps this is to avoid attracting hyenas and lions by hoo-calling. Perhaps they do not hoo-call simply because they have a fixed base (the den) to which they can return and relocate their packmates, rather than calling for them. A monitor lizard entered one den, and apparently ate one of eight pups while underground. A porcupine also entered a den, but did no harm to the pups (despite remaining in the den for hours). Adults watched the monitor and porcupine approach with apparent curiosity, but did not block their entrance, something the dogs could easily have done. Subjectively, it seemed as though they did not recognize a threat until the animal was in the den. The monitor was ignored completely. The dogs tried to enter the den occupied by the porcupine, but were blocked by its quills. They then tried to call the pups out by hooing, but the pups did not emerge until the porcupine left.
3.5 Pack Size and Range Size Larger packs might defend larger home ranges, if wild dogs are “expanionists” in the terminology of Kruuk & Macdonald (1985), using strength in
52 ▪ C H A P T E R 3
numbers to defend more resources. On the other hand, Macdonald’s (1983) resource dispersion hypothesis suggests that range size and group size can be decoupled, if range size depends on the distribution of food patches and group size depends on the richness of patches. In Selous, home range size and pack size showed no significant correlation (r ⳱ 0.14, df ⳱ 17, P ⳱ 0.58). For annual ranges, the correlation was negative and nonsignificant (r ⳱ ⳮ0.33, df ⳱ 7, P ⳱ 0.42). For multiyear ranges, the correlation was nonsignificantly positive (r ⳱ 0.49, df ⳱ 9, P ⳱ 0.13). Using multiyear ranges for eight packs in Kruger National Park, Mills and Gorman (1997) also found a nonsignificant positive relationship between range size and pack size. In short, there is no evidence that range size is related to pack size in wild dogs. If a relationship exists, it is weak.
3.6 Habitat Selection Habitat Availability, Habitat Use, and Habitat Preference within Home Ranges The study area had six major vegetation types. Deciduous woodland was predominant, covering 59% of the study site. Deciduous woodland included miombo, chipya, and Combretum-Terminalia woodlands. Long grass covered 26% of the area. This included areas with grass over 20 centimeters high, but “long grass” was usually over a meter, often 2–3 meters. Long grass areas generally had few trees, or none. Thorn woodland (Terminalia spinosa—Acacia drepanolobium) covered 10% of the area. Short grass (less than 20 cm) areas covered 5% of the area, and generally supported scattered trees (of varying species). Riverine thickets and palm swamps each covered less than 1% of the study area. Figure 3.2 maps the study area vegetation. To test if dogs preferred certain habitats, we compared habitat use to habitat availability within each home range. For each pack, we measured habitat availability by determining the proportion of the home range covered by each vegetation type. We then measured habitat use by determining the proportion of fixes that fell in each vegetation type (recall that fixes were taken only at kills and resting sites). Habitat preference was measured as the ratio of use to availability; values greater than 1 indicate preference and values less than 1 indicate avoidance. For this analysis we used a single multiyear range for each pack (N ⳱ 11), so that all packs had equal weight. We did not calculate preference ratios for palm swamp and riverine thicket, which covered very small proportions of most home ranges (Table 3.3). For habitats covering small areas, small measurement errors can alter the preference ratio substantially. Wild dogs showed a significant preference for deciduous woodland (Table 3.3: preference ratio ⳱ 1.3, single-point t10 ⳱ 2.23, P ⳱ 0.05). This preference has one clear cause. All dens were located in
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 53 Table 3.3 Habitat availability, habitat use, and habitat preference within wild dog home ranges. Entries in the body of the table give mean Ⳳ standard error, with the range in parentheses Availability (% of area)
Use (% of fixes)
Preference Ratio (use/availability)
Deciduous Woodland
53.4 Ⳳ 7.2 (11.7, 95)
45.4 Ⳳ 6.4 (8, 75)
1.30 Ⳳ 0.13 (0.90, 2.14)
Thorn woodland
23.0 Ⳳ 5.8 (0, 62)
23.6 ⳮ 5.5 (0, 57)
1.20 ⳮ 0.27 (0.03, 3.00)
Short Grass
8.9 Ⳳ 2.4 (0, 21)
11.5 Ⳳ 2.9 (0, 27)
0.94 Ⳳ 0.28 (0.08, 2.93)
Long Grass
14.5 3.7 (0, 31)
14.4 Ⳳ 3.4 (0, 37)
1.07 Ⳳ 0.20 (0.10, 1.92)
Palm Swamp
0.9 Ⳳ 0.5 (0, 4)
0.3 Ⳳ 0.3 (0, 3)
N/A
Riverine Thicket
0.2 Ⳳ 0.1 (0, 1)
4.7 Ⳳ 2.1 (0, 25)
N/A
Habitat Type
sandy soil, and sandy soil usually supports deciduous woodland in Selous. Hence, for three months of each year while the dogs were denning, habitat use was biased toward deciduous woodland. Herd sizes are also large in deciduous woodland, and large herds provide a greater likelihood of making a kill (Chapter 5), but the density of prey in deciduous woodland was low (see “Effect of Prey Distribution on Habitat Selection and Home Range Properties,” below). habitat availability and home range size A home range with a high proportion of preferred habitat might be smaller than a home range with little preferred habitat. To test this hypothesis, we regressed home range size on the proportion of the range covered by each habitat type. We ran four regressions, one for each major habitat type (deciduous woodland, thorn woodland, short grass, and long grass). Home range size did not significantly correlate with the proportion of the range covered by any of the four habitat types (with Bonferroni adjusted α ⳱ 0.0125). The strongest (though nonsignificant) association was for ranges with high proportion of long grass to be large (r ⳱ 0.28, t14 ⳱ 1.24, P ⳱ 0.3). This tendency is consistent with data on prey encounter rates, which suggest that long grass is a poor habitat for wild dogs (see below). We used the same procedure to test whether the area of home range’s core was associated with habitat type. None of the regressions were significant
54 ▪ C H A P T E R 3 Table 3.4 Habitat availability and habitat selection for the population as a whole. Entries in the body of the table give the proportion of the area for each level of overlap covered by each habitat type Number of Overlapping Packs Habitat Type
1
2
3
4
5
6
Proportion of Study Site
Deciduous Woodland Thorn Woodland Short Grass Long Grass
0.68 0.07 0.01 0.24
0.52 0.21 0.08 0.19
0.28 0.43 0.17 0.17
0.36 0.39 0.16 0.09
0.83 0.11 0.03 0.03
1.00 0 0 0
0.59 0.10 0.05 0.26
(Bonferroni adjusted α ⳱ 0.0125). The strongest association was for cores to be small if they held a high proportion of deciduous woodland (r ⳱ ⳮ0.39, t14 ⳱ 1.78, P ⳱ 0.1), consistent with other results (above) showing that deciduous woodland is preferred. Habitat Preference for the Population as a Whole In addition to the pack-by-pack method of testing habitat preferences described above, we calculated a second index of habitat preference for the population as a whole, as follows. Using Idrisi, we cross-tabulated the map of home-range overlap zones (Figure 3.6) with vegetation types (Figure 3.2), to determine the proportion of the area for each level of overlap (1 pack, 2 packs, . . . 6 packs) covered by each habitat type (call this quantity A). These proportions are given in Table 3.4. We then determined the proportion of the entire study area covered by each habitat type (call this quantity B; Table 3.4). We calculated the preference ratio as (A* 噛 of overlapping packs)/B. Habitat preference ratios are given in Table 3.5. Ideally, we would use factorial ANOVA to test the interaction between habitat type and level of Table 3.5 Habitat preference values for the wild dog population as a whole. High preference values at the right of the table indicate a preferred habitat. High preference values at the left indicate an avoided habitat. (See text for details of calculation) Number of Overlapping Packs Habitat Type
1
2
3
4
5
6
Deciduous Woodland Thorn Woodland Short Grass Long Grass
1.15 0.70 0.20 0.92
1.76 4.20 4.80 1.46
1.41 8.60 10.20 1.96
2.44 15.60 12.80 1.73
7.05 5.50 3.00 0.58
10.14 0 0 0
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 55
overlap, but our data provide no degrees of freedom for this test. Informally, a habitat is preferred if values at the right of Table 3.5 are larger than values on the left. This logic suggests that deciduous woodland is most preferred, thorn woodland is less strongly preferred, short grass is neutral, and long grass is avoided. In principle, areas not used by any pack (Figure 3.6) could be compared with the vegetation map (Figure 3.2) to identify habitats that are avoided. This comparison confirms that long grass areas are avoided, but rests on the assumption that these areas were not used by packs that we failed to detect. It is possible that undetected packs occupied the apparently unused areas, but we think it is unlikely that any packs permanently occupied the large long grass areas at the northwest and northeast of the study site (Figure 3.2), for three reasons. First, ungulate density in these plains is very low during the dry season, because the grass is either long (⬎2 meters) and fibrous, or burnt. Second, the plains are permanently flooded and the grass is dense in the wet season, making movement difficult for wild dogs. Third, the collared packs that bordered these plains rarely entered them.
3.7 Effect of Prey Distribution on Habitat Selection and Home Range Properties Encounter Rates with Herds Obviously, the density of prey might vary among habitats, and this might be a cause of wild dogs’ preference for deciduous woodland. To test this hypothesis, we recorded the rate at which packs encountered prey (herds per kilometer traveled) for 11 pack-years, with an average of 44 daily data points per pack-year (Table 3.6). We tallied encounters with all prey species combined, as well as separate tallies for encounters with wildebeest and impala, the two most important prey. We regressed prey encounter rates on the proportion of the home range covered by each major habitat type (Table 3.7). These regressions show that prey herds were encountered significantly more often in home ranges with a high proportion of short grass (F ⳱ 10.58, P ⳱ 0.0087, r2 ⳱ 0.51). In particular, encounter rates with wildebeest herds were high for home ranges rich in short grass (F ⳱ 22.27, P ⳱ 0.0063, r2 ⳱ 0.67). Encounter rates with impala herds did not vary significantly across home ranges of varying composition. Impala are found in all parts of the study area except long grass, which was little-used by the dogs. When one considers that short grass is highly nutritious for grazing wildebeest, it is not surprising that encounters with prey are highest in short grass areas. However, it is surprising that the wild dogs’ preferred habitat (deciduous woodland) provides a lower rate of encounter with prey than
1.30 Ⳳ 0.13* 1.20 Ⳳ 0.27 1.07 Ⳳ 0.20 0.94 Ⳳ 0.28 not measured4
Preference Ratio2,3 422 370 235 297 28 1,352
(31%) (27%) (17%) (22%) (2%)
Prey Herds Encountered 5818 5908 5183 6462 199 23,570
(25%) (25%) (22%) (27%) (1%)
Prey Individuals Encountered 1552 825 458 393 43 3271
(47%) (25%) (14%) (12%) (1%)
Kilometers Traveled in Habitat
2
Entries give raw numbers, then percentage in brackets. Preference ratio ⳱ % of radiolocations within a habitat type ⳰ % of homerange covered by that habitat type. 3 Mean Ⳳ SE. 4 Riverine thicket covered ⬍1% of the wild dog study area, making its preference ratio highly sensitive to measurement error. *P ⬍ 0.05 for single-point t-test comparing preference ratio with 1.
1
Deciduous Woodland Thorn Woodland Long Grass Short Grass Riverine Thicket Totals
Habitat Type
Wild Dog
Table 3.6 Encounters between wild dogs and prey as a function of habitat type1
3.75 7.16 11.31 16.4 4.63
Prey Individuals per km Traveled 4 2 5 3 1
Lion Preference Rank
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 57 Table 3.7 P-values from regressions of herd encounter rates on the proportion of the home range covered by each major habitat type Herds Encountered per Kilometer
Proportion of Range Covered by:
Wildebeest
Impala
All Prey
Deciduous Woodland Thorn woodland Short Grass Long Grass
0.28 0.73 0.00063* 0.65
0.82 0.52 0.89 0.51
0.40 0.61 0.0087* 0.46
*Asterisks denote a significant correlation between encounter rate and proportion of range covered by that habitat type. The Bonferroni-adjusted critical value for each test is 0.0041, to give an overall (experiment-wise) error rate of 0.05.
could be obtained in an unpreferred habitat (short grass). As mentioned above, one reason that wild dogs prefer deciduous woodland is because it is suitable for denning. It is also possible that wild dogs would prefer short grass if it were not for lions. Across the study site, there was a negative correlation between the spatial distributions of wild dogs and lions (Figure 3.8). Wild dogs’ only significant habitat preference was for deciduous woodland, which held a low density of prey, but was little used by lions (Table 3.6; Figure 3.8). Data from 1,352 encounters with prey during 3,271 kilometers of wild dog follows showed that only 3.75 prey were encountered per kilometer, significantly lower than prey encounter rates in other habitat types, which ranged from 4.6 to 16.4 prey/km (Table 3.6). Similar to Selous, wild dogs in Kruger National Park show no significant preference for habitats with high densities of prey (Mills & Gorman 1997). Mills & Gorman (1997) compared the habitat use of wild dogs and lions, and showed that wild dogs avoid habitats that are heavily used by lions. Although the habitat types in Selous and Kruger differ, the conclusion that wild dogs avoid lion habitats is the same (Mills & Gorman 1997), and this avoidance may constrain the dogs’ ability to use areas of high prey density. By contrast, space use by wild dogs and spotted hyenas in Selous is similar, as shown by a significant correlation between the spatial distributions of the two species (Chapter 11). Herd encounter rates had no significant association with home range area (regression, F1,10 ⳱ 0.05, P ⳱ 0.82) or the proportion of the range that was exclusive (F1,10 ⳱ 0.17, P ⳱ 0.69). Herd Sizes The preceding analysis of rates of encounter with herds considered all herds equivalent. However, wild dogs are more likely to make a kill when they
58 ▪ C H A P T E R 3
Figure 3.8 Space-use distributions of lions in the northern Selous Game Reserve. Dark indicates high usage. Plot frame shows 37⬚ 45⬘ E to 38⬚ 30⬘ E and 7⬚ 15⬘ S to 7⬚ 55⬘ Space use was estimated for lions by projecting habitat preference ranks onto a map of habitat types (see Figure 3.2). There is a significantly negative correlation between the spatial distributions of lions and wild dogs, when the distributions are compared after thinning (resampling) to account for spatial autocorrelation (Cramer’s V ⳱ 0.36, df ⳱ 35, P ⬍ 0.01). See Figure 3.6, which has the same plot frame coordinates.
hunt a large herd (see Chapter 6), so if herd sizes vary among habitats, this variation might affect the wild dogs’ habitat preferences. To test this hypothesis, we recorded the size and composition of 660 herds of wildebeest and impala, encountered randomly in transect drives. Factorial ANOVA showed that herd sizes do vary among habitats (F ⳱ 2.99, P ⳱ 0.019) and seasons (F ⳱ 5.28, P ⳱ 0.021), with a significant interaction between the two factors (F ⳱ 3.01, P ⳱ 0.018). There was also a significant three-way interaction between prey species, habitat, and season (F ⳱ 3.0, P ⳱ 0.017), although the difference between species was marginally nonsignificant
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 59 Table 3.8 Local habitat thickness and the likelihood of killing prey Outcome of Hunt Habitat change at end of hunt
Kill
No Kill
Thicker Thinner or no change
98 282
17 476
(F ⳱ 3.58, P ⳱ 0.059; impala herds ⳱ 13.3 Ⳳ 1.47; wildebeest herds ⳱ 9.3 Ⳳ 1.51). Does this variation predict a preference for deciduous woodland? The answer is a qualified yes. Prey herds were larger in deciduous woodland (least square mean ⳱ 11.7 individuals) than in all other habitats except long grass (least square mean ⳱ 18.2 individuals), and a multiple range test (Fisher’s protected 95% LSD) includes deciduous woodland in the set of three habitats with the largest herds. This set also includes long grass and riverine thickets, which are difficult habitats in which to hunt. Because wild dogs are shorter than most of their prey, they cannot see well in long grass and consequently cannot make as direct a pursuit as they can in more open environments. Because of the height differences, dogs are also physically impeded by thick long grass to a greater extent than their prey are. This is easily seen in hunts of reedbuck, which outpace the dogs easily in long grass, but are quickly caught in the open. The dogs also cannot make fast pursuits in riverine thickets, but prey cannot flee quickly either, and in this case the difference in body size favors the dogs. Wild dogs do not abandon a hunt when prey flee through thickets. The opposite is more often true—prey often hesitate when they encounter a thicket, making a decision about the route to take, and this allows the dogs to gain ground. In particular, impala are often grabbed while hesitating near a barrier of some sort. Consequently, kills were significantly more likely at sites thicker than the habitat through which the prey was chased (Table 3.8, χ2 ⳱ 91.7, P ⬍ 0.001). In summary, herd sizes (and hunting behavior) favor both deciduous woodland and riverine thickets for wild dogs. By contrast, herd encounter rates clearly favor short grass, but lions may prevent wild dogs from using short grass areas heavily.
3.8 Comparison with Other Wild Dog Populations The annual home ranges of Selous wild dogs averaged 379 km2, the smallest on record (Table 3.9). Average home range sizes in five other populations ranged from a minimum of 423 km2 in Hwange (Fuller et al. 1992a) to a
60 ▪ C H A P T E R 3 Table 3.9 Annual home range size, pack size, and population density in six wild dog populations
Home Range (km2)
Pack Size (Adults)
Population Density (ind/km2)
Selous Kruger
379 533
8.9 4.8
0.040 0.017
Serengeti
665
5.7
0.011
Moremi
617
9.0
0.040
Hwange
423
7.0
0.015
Aitong
650
4.2
—
545 Ⳳ 49
6.6 Ⳳ 0.8
0.025 Ⳳ 0.006
Population
Mean Ⳳ SE
Source This study Reich 1981, Gorman et al. 1992, Mills & Gorman 1997 Schaller 1972; Frame et al. 1979; Fanshawe & Fitzgibbon 1993 McNutt 1996; J. Bulger unpublished report Fuller et al. 1992a; Woodroffe et al. 1997 Fuller & Kat 1990; Fuller et al. 1992a
maximum of 665 km2 in Serengeti (Schaller 1972). Including Selous, the mean home range size was 545 Ⳳ 49 km2. Excluding Selous, the average home range size across five populations was 577 Ⳳ 45 km2, not significantly larger than in Selous (single-point t ⳱ 1.79, df ⳱ 4, P ⳱ 0.15, β ⳱ 0.9), but this comparison has little statistical power. There is great variation in the quality of the home range estimates from different ecosystems (Table 3.9). The value from Kruger is based on 21 ranges from two studies (Reich 1981; Mills & Gorman 1997). The value from Moremi is based on nine ranges (McNutt 1996). The value for Hwange is based on four ranges (Woodroffe et al. 1997). We did not use Childes’s (1988) estimate of 1,500 km2 for Hwange, because it was not based on systematic data. The value from Serengeti comes from two packs (Schaller 1972). We elected not to use a value of 1,500–2,000 km2 (N ⳱ 6) from Frame et al. (1979), because these ranges included all observations over a period of 4.5 years. This extraordinarily large value is widely cited (e.g., Moehlman 1989; Fuller et al. 1992a), but Schaller’s (1972) estimates are probably more comparable to data from other populations. We included the value for Aitong (Kenya), though it comes from a single pack observed for six months (Fuller & Kat 1990). Home range size and group size might be associated (Kruuk & Macdonald 1985). Within the Selous population, range size and adult pack size were not associated, but one can also ask this question at the interpopulation level. Here again, the correlation was not significant, and tended to be negative
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 61 Table 3.10 Home range size, group size, and body size for canids greater than 7.5 kilograms
Species
Home Range (km2)
Group Size (adults)
Female Body Mass (kg)
Black-backed Jackal Golden Jackal Coyote
2–22A (12)B 0.1–2 (1) 4.5–107 (50)
2.9 2.7 4.4
7.5 9.0 10.9
Ethiopian Wolf Maned Wolf Dhole
5.5–13.4 (9.5) 22–30 (26) 20–83 (51)
5.1 2.0 7.7
12.8 22.7 13.8
African Wild Dog
150–1,500 (545)
6.6
23.0
Wolf
94–13,000 (1,200)
6.7
31.1
Source Moehlman 1986 Moehlman 1986 Springer 1982; Andelt 1985; Thurber et al. 1992 Sillero-Zubiri 1994 Dietz 1984 Johnsingh 1982; A. Venkataraman, pers. comm. This study; Reich 1981; Frame et al. 1979; Mech 1970
A ⳱ range; B ⳱ mean.
(r ⳱ ⳮ0.47, t5 ⳱ 1.07, P ⳱ 0.34). Overall, the data do not support the hypothesis that larger packs defend larger ranges. Implications for Population Density and Conservation Wild dog density is unusually high on our study site (Chapter 1). Comparing with other populations, high density is the result of two trends. First, home ranges are the smallest of six populations. Second, adult pack size is the second-largest of six populations (Table 3.9; single-point t ⳱ 1.10, df ⳱ 4, P ⳱ 0.38). Large packs on small home ranges will inevitably produce high population density, if the degree of home range overlap does not counteract the pattern. As noted above, home range overlap in Selous was substantial, and was similar to overlap in other populations (averaging 25–35%; Schaller 1972; Reich 1981). Comparison with Other Canids Home range size, pack size, and female body mass for canids greater than 7.5 kilograms are summarized in Table 3.10. For these species, published home range sizes span an astounding five orders of magnitude, from 0.1 km2 in golden jackals to 13,000 km2 for wolves. Across species, there was a positive exponential relationship between home range size and body mass (F1,6 ⳱ 10.9, P ⳱ 0.016, r2 ⳱ 0.64). Wild dogs were an outlier from this
62 ▪ C H A P T E R 3
Figure 3.9 The regression of home range size on female body mass for canids ⬎ 7 kg. Each data point represents one species, using mean values for mass and range size. Wild dog ranges are considerably larger than predicted by this regression.
relationship, with ranges five times larger than predicted (Figure 3.9). There was also a positive exponential relationship between range size and adult group size (Figure 3.10; F1,6 ⳱ 4.99, P ⳱ 0.067, r2 ⳱ 0.45) of marginal significance. Here again, wild dog ranges were five times larger than expected. Comparison with Ecologically Similar African Carnivores In two ecosystems, Serengeti and Kalahari, home range sizes are known for the entire suite of predators that rely on ungulate prey (Table 3.11). The comparison of these ecosystems shows that home range sizes are affected by ungulate densities. Prey density is much lower in Kalahari than in Serengeti, and ranges are considerably larger. Migrations of prey in Serengeti also com-
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 63
Figure 3.10 The regression of home range size on adult group size for canids ⬎ 7 kg. Each data point represents one species, using mean values for mass and group size. Wild dog ranges are considerably larger than predicted by this regression.
plicate comparison among species, but the problem of migratory prey is faced by all the large carnivores in Serengeti (Hofer & East 1995; Sinclair 1995). Within Serengeti, wild dogs had the largest home ranges among the large carnivores, with the possible exception of female cheetahs (which are not territorial). The territory size tabulated for Serengeti spotted hyenas excludes extraterritorial forays, which were common and produced total ranges of several hundred square kilometers (Hofer & East 1993a,b). However, data from Kalahari show that, with low prey density, the home ranges of other carnivores rival those reported for wild dogs. Interestingly, a single pack of wild dogs was seen in Kalahari, but its range was unknown (Mills 1990). It appears that Kalahari barely supports wild dogs (or does not support them at
64 ▪ C H A P T E R 3 Table 3.11 Home range sizes of large (⬎20 kg) African carnivores in two ecosystems Ecosystem
Home Range (km2)
Leopard
Serengeti
Spotted hyena Lion
Serengeti Serengeti
female: 14–50 male: 18–58 56 30–400
Cheetah
Serengeti
Wild dog
Serengeti
male: 37 female: 833 610–2000
Leopard Brown hyena Cheetah Spotted hyena Lion
Kalahari Kalahari Kalahari Kalahari Kalahari
200–400 215–461 ⬎300 553–1776 187–3900
Wild dog
Kalahari
1 pack/entire park
Species
Source Schaller 1972; Bertram 1982 Hofer & East 1993b; 1995 Schaller 1972; van Orsdal et al. 1985 Caro 1994 Schaller 1972; Frame et al. 1979 Bothma & LeRiche 1984 Mills 1990 Mills 1984 Mills 1990 Eloff 1973 Owens & Owens 1984 Mills 1990
all). Overall, wild dog ranges are larger than predicted by comparisons with sympatric carnivores, or by comparisons with other canids. We can suggest two hypotheses why wild dog ranges are large. First, hunting success in a small area might decrease as na¨ıve prey become more aware, or better at avoiding predation. If so, the dogs would benefit from having a home range large enough to shift among areas in which they have not hunted recently. We can test this hypothesis with data on hunting success during denning periods. Denning dogs return to the same spot every day, which restricts the area in which they hunt. Contrary to our hypothesis, hunting success did not decrease as the denning period progressed (Figure 3.11; F1,50 ⳱ 0.34, P ⳱ 0.56, r2 ⳱ 0.007), and hunting success during the denning period (51% Ⳳ 5%) was greater than hunting success outside the denning period (42% Ⳳ 2%, t172 ⳱ 2.59, P ⬍ 0.01). Moreover, the daily distance traveled did not increase over the denning period (F1,64 ⳱ 0.002, P ⳱ 0.96, r2 ⬍ 0.005). Overall, there is no evidence that restricted movements decrease hunting success. Second, wild dogs might need a range large enough to avoid areas that are in heavy use by larger carnivores, to minimize interspecific competition. Data from Kruger National Park (Mills & Gorman 1997) and from Selous (Creel et al. 2001) suggest that wild dogs avoid lions, consistent with this hypothesis.
H O M E R A N G E S A N D H A B I TAT S E L E C T I O N ▪ 65
Figure 3.11 The relationship between hunting success (kills/hunt) and the number of days that a pack had denned, and thus had hunted in a confined area. Hunting success does not change as the denning period progresses.
3.9 Summary Wild dogs are endangered largely because they always live at low population density. Large home ranges underlie wild dogs’ low population density. The analyses in this chapter suggest that wild dog home ranges are unusually large, in comparison to those of other canids and in comparison to those of ecologically similar African carnivores. Across wild dog populations, density tends to be higher where home ranges are smaller (and where packs are larger). Selous exemplifies this pattern, with the smallest home ranges and the highest population density reported. Large ranges do not serve to keep hunting success high by shifting among areas with na¨ıve prey. Wild dogs
66 ▪ C H A P T E R 3
may be a fugitive species, with large ranges allowing avoidance of areas that are heavily used by larger carnivores, particularly lions. Within Selous, and across wild dog populations, home range size and pack size were not significantly correlated. The dogs showed a significant preference for deciduous woodland. No other habitats were significantly preferred or avoided. Prey encounter rates were highest in short grass, but wild dogs may be prevented from using short grass areas heavily because lion density is high there. Herds were large in deciduous woodland, and wild dogs hunt large herds more effectively. Deciduous woodland also provided suitable sites for denning.
4
Cooperative Hunting and the Evolution of Sociality
Cooperative hunting is one of the most conspicuous aspects of the behavior of large social carnivores, one that draws attention from anyone who sees it. In a hunt, the power and grace that predators and prey have bred into one another are on display with life and death in the balance. Prior to quantitative research, it was widely but uncritically accepted that group living in large carnivores arose through the benefits of cooperative hunting. In the 1960s, influential field studies confirmed that communal hunting could favor life in groups, either by improving the odds of making a kill or by increasing the size of prey that could be killed (Schaller 1972; Kruuk 1975). For example, Ngorongoro spotted hyenas, Crocuta crocuta, typically hunted alone when pursuing Thomson’s gazelles, Gazella thomsoni, but formed groups averaging 10.8 hyenas when hunting zebras, Equus burchelli (Kruuk 1972). Wolves, Canis lupus, typically hunt Dall sheep, Ovis dalli, alone, but tackle moose, Alces alces, in groups (Murie 1944; Mech 1970). Associations between hunting group size and prey size are common among carnivores (Gittleman 1989), but are not universal. For example, hunting group size had no effect on species hunted or captured by lions, Panthera leo, in Etosha National Park (Stander 1992). An association between hunting group size and prey size could result from a beneficial increase in the vulnerability of large prey to large hunting groups. However, an association between group size and prey size does not resolve whether hunting large prey is a benefit or a necessity. Large hunting groups might take large prey simply because small prey do not meet their needs (Caro 1994). Large prey might be disproportionately costly to hunt, but more profitable than killing several small prey. If so, then large groups would be selected to take large prey, but would still be less efficient than small hunting groups. Beyond effects on prey size, communal hunting usually improves hunting success (defined as the percentage of hunts that end in kills), but sometimes does not. Ngorongoro spotted hyenas succeeded in 15% of solitary hunts of wildebeest calves, but in 74% of group hunts. This benefit arose mainly because groups of hyenas never failed to kill a calf when its mother chose to stop and fight, while solitary hyenas never succeeded in this situation (Kruuk 1972). Averaged across all prey types, the hunting success of spotted hyenas in the Masai Mara increased significantly as the number of hunters increased from one to three, but leveled off thereafter (Holekamp et al. 1997). In contrast, spotted hyenas in Kalahari showed no increase in hunting success
68 ▪ C H A P T E R 4
across group sizes one to seven when hunting gemsbok or wildebeest, which comprise 68% of their prey (Mills 1985, 1990). In Serengeti lions, the hunting success of groups was double that of singletons (groups: 82 kills in 273 hunts ⳱ 30%; solitaries: 37 kills in 249 hunts ⳱ 15%; Schaller 1972). In Etosha, female lions’ hunting success significantly increased across group sizes one to seven, for each of the five most common prey (Stander & Albon 1993). Male Serengeti cheetahs showed no clear effects of group size on hunting success for large or small prey (Caro 1994). All in all, the relationship between group size and hunting success is variable, but usually positive. Aspects of the predator’s behavior, defenses of the prey, and the structure of the habitat can alter the costs and benefits of hunting with others. In summary, several decades of field studies have shown that cooperative hunting provides some benefits that could favor the evolution of sociality, or could provide selection to maintain sociality once it arose for other reasons. However, none of these studies directly demonstrated that communal hunting has a net benefit (Packer 1986). Even if large groups benefit from higher hunting success, larger prey, or a higher probability of killing multiple prey, there still might be a decline in per-capita food intake because the prey is divided among a large number of hunters (Kruuk 1972, 1975). This is fundamentally an empirical issue—many of the variables affecting food intake change with group size, and they must be quantified simultaneously to determine how costs and benefits vary with group size. In the 1990s, several studies measured daily per-capita food intake as a function of group size, and tested whether individuals typically adopted the group size that maximized intake (see Table 4.3 and associated text). This approach was taken in studies of African wild dogs (Fanshawe & Fitzgibbon 1993), cheetahs (Caro 1994) and two lion populations (Packer et al. 1990; Stander 1992). Grouping improved foraging success in several tests, but only lions in Etosha National Park showed grouping patterns that closely matched predictions based on the benefits of communal hunting (Stander 1992). Using the currency of daily per-capita food intake, only Etosha lions have been shown to hunt preferentially in the group size predicted if communal hunting is the prime determinant of group size (Stander 1992; Stander & Albon 1993), and there is considerable debate about relationships between foraging success and group size in social carnivores. While acknowledging that data are limited, recent summaries of communal hunting have argued that communal hunting has little power to explain grouping patterns in felids (Packer 1986; Packer et al. 1990; cf. Stander 1992) or social carnivores in general (Caro 1994). However, all these studies have addressed only the effects of group size on the payoff to hunting, disregarding costs. This is an important limitation of virtually all analyses of cooperative hunting by social carnivores. If individuals in groups of all sizes hunted sufficiently often to meet their needs and no more, daily per-capita food intake would not vary across group sizes.
C O O P E R AT I V E H U N T I N G ▪ 69
Measures of belly distention (Packer 1986) would not vary with group size. Although groups of some sizes might work harder to attain this food intake, in terms of time or distance traveled, this variation in hunting effort would not affect any of the results discussed above. Wild dogs are well suited for a test of the effects of communal hunting on pack size (Packer & Ruttan 1988), because common prey are large relative to the dogs (e.g., yearling wildebeest, Connochaetes taurinus, outweigh a wild dog by a ratio of 6:1; Figure 4.3), or are risky to attack (e.g., warthogs, Phacochoerus aethiopicus; Figure 4.3). In this chapter, we present data on hunting effort, hunting success, and food intake for African wild dogs in the Selous Game Reserve. We determine optimal hunting group size using the traditional evolutionary currency, daily per-capita food intake. We then show that inclusion of hunting costs substantially alters predictions for optimal hunting group size. We discuss the methodological hurdles of converting costs and benefits to a common currency (kilojoules), in order to calculate the net benefit of hunting in different pack sizes. Finally, we compare our results to those found in other wild dog populations, and present a quantitative meta-analysis of the relationship between food intake and group size in cooperative hunters.
4.1 Specific Methods Hunting Observations Systematic hunting data came from six packs, between November 1991 and March 1994 (11 pack-years of observation). We collected hunting data by direct observation during periods lasting from 1 to 14 days, during which we observed 404 kills in 905 complete hunts. Most observation periods lasted 3 or more days, but a few were cut short by losing the dogs, getting stuck, or breaking down (the vehicle, not us). After checking that the data from periods less than 3 days did not differ with respect to frequency of hunting or hunting success, we included it. Of 310 days on which we recorded hunting data (2210 h), the analyses of this chapter are restricted to 266 days on which we observed entire hunting periods (Mills 1992). Following Packer et al. (1990), our measures of hunting cost and benefit use daily means as data points. For wild dogs, this removed a bias toward zeros that existed for some variables when expressed per hunt, allowing the use of parametric statistics. Most of the hunting observations were made at a distance of 50–150 m, but occasionally we were as near as 20 m or as far away as 400 m. Because the wild dogs were well habituated to the vehicle but prey were not, we took care not to startle prey herds, but we did occasionally cause prey to flee (about 5–10% of herds). Disturbance by the vehicle favors the prey in impala hunts (because they are alerted at a greater distance), but favors the
70 ▪ C H A P T E R 4
Figure 4.1 Time of day at which kills were made.
dogs in wildebeest hunts (because wildebeest can defend themselves at bay, but not while running from a vehicle). Most hunting occurred in two periods, 0500–0900 hours and 1700–2000 hours (Figure 4.1), as in other populations (Kuhme 1965; Estes & Goddard 1967; Fuller & Kat 1990; Mills & Biggs 1993). Perhaps because Selous is wooded, wild dogs rarely hunted at night,
C O O P E R AT I V E H U N T I N G ▪ 71
although they often traveled slowly on moonlit nights. Most nocturnal kills were made by a single pack that hunted on clear nights with a half-moon or more, during denning periods. Nocturnal hunts may be easier in the denning period because the den provides a fixed point to rejoin dogs that are separated in the chase. During each hunt, we attempted to record the habitat type, prey species, herd size and composition, distance chased (estimated using the vehicle odometer), individuals leading the chase, individuals initiating the kill, characteristics of the kill site, estimated mass of prey killed, estimated mass of remains not eaten, time on kill, and interactions with other carnivores. This chapter focuses on relationships between group size and the costs and benefits of hunting. Chapter 5 addresses prey selection, and Chapter 6 addresses the vulnerability of prey as a function of herd size. Definition of Terms Due to variation in hunting techniques (e.g., stalking versus coursing) no single definition of a hunt applies to all large carnivores (Kruuk 1972; Schaller 1972; Mills 1990; Stander 1992). We defined a hunt (or “chase”) as a pursuit of prey that either exceeded 50 m at a full run, ended with the dogs testing prey at bay, or ended in a kill. In Selous, wild dogs traveled 12.3 km Ⳳ 0.5 (mean Ⳳ SE throughout) daily, and often passed near groups of potential prey that were ignored or tested briefly and with low effort. Our definition excludes low cost and apparently casual interactions with prey. Different studies have defined hunts in different ways. Some authors consider an entire morning or evening hunting period to be a single hunt (e.g., Schaller 1972; Fuller et al. 1995). In our analyses, this would be considered a “hunting period.” In a typical hunting period, a wild dog pack would locate and chase several herds of prey, with chases interspersed with periods of slow travel and rest. In our main analyses, each distinct chase is considered a hunt, and hunting success is calculated as kills per hunt. (As the analyses below show, the number of hunts per hunting period is an important aspect of group-size-specific hunting success and hunting effort.) However, we also report hunting success calculated as kills per hunting period, to allow easy comparison with other studies that define hunting success in this way. We defined hunting group size as the number of adults in the pack. A priori, yearlings could plausibly be considered either dependents or hunters. Yearlings (4.3 Ⳳ 0.3 per pack in the 266 days analyzed) usually participate in hunts and sometimes provide obvious help, but sometimes cause obvious hindrance. The number of yearlings was not associated with hunting success (t ⳱ 0.66, P ⳱ 0.51) or with mean kill mass (t ⳱ 0.22, P ⳱ 0.83), but had a significant association with increased chase distance (t ⳱ 2.55, P ⳱ 0.012). Together, these relationships suggest that, on average, yearlings slightly impede hunting efforts and should be considered dependents. When
72 ▪ C H A P T E R 4
yearlings are considered part of the hunting force, noise is added to the analyses (i.e., R2 values drop) but the slopes and shapes of curves are not substantially altered. Chases were probably longer for packs with yearlings because they are sometimes impatient when approaching prey. It is not unusual for yearlings to rush a prey herd before the adult dogs, causing the herd to flee at a greater distance than normal. When this happens, the adults sometimes do not bother to pursue such herds, but when they do, chases are longer. No measures of hunting success or effort were affected by the adult sex ratio within a pack, which ranged from 20% to 80% male. This confirms our impression that male and female wild dogs were equally effective hunters in Selous. In Serengeti, data from four packs showed that males were more likely than females to lead chases and to make the first grab at prey (Malcolm & Marten 1982), but this result arose mainly because two of the four packs contained only one female. Alpha females are not effective hunters while they are pregnant or lactating. We had the impression that some alpha females did not hunt as intensively or effectively as other pack members even outside the breeding season, perhaps as a consequence of being “excused” from hunting during pregnancy and denning. Wild dog packs are highly cohesive, and all pack members normally moved together during morning and evening hunting periods. (An exception is the 2–3 month denning period, when the dominant female usually does not hunt; Malcolm & Marten 1982). Social rallies ordinarily preceded hunting, and appeared to excite and coordinate the pack in becoming active (Kuhme 1965; Estes & Goddard 1967; Malcolm 1979). When a chase began, all pack members normally pursued and harassed prey. We feel it would be misleading to use behavioral data to exclude some pack members from the hunting group, for three reasons: (1) We often did not have all dogs continuously in view through an entire hunt. In a wooded habitats, individual hunters can pass in and out of view repeatedly during a chase, even when all pack members are pursuing the same prey. (2) Multiple kills were common, but we often detected multiple kills only after the hunt was finished. Thus, dogs not in view at one kill were often pursuing another prey animal from the same herd. (3) Participation in a hunt is difficult to define operationally because the simple presence of an additional hunter may affect the prey’s behavior or escape options (Reich 1981; Stander 1992; see data below). Fanshawe & Fitzgibbon (1993) classified wild dogs as nonparticipants if they neither led the chase nor pulled the prey to the ground. By this definition, a wild dog would be scored as a nonparticipant if it pursued one member of a wildebeest herd, broke off its pursuit when it heard the bellowing of another wildebeest as it was grabbed, then ran to join the kill after the wildebeest was off its feet. In this case (which is common in Selous), the unsuccessful hunter may have exerted a greater effort than other dogs that were scored as hunters. Failure should not be equated with nonparticipation. The drawback
C O O P E R AT I V E H U N T I N G ▪ 73
to our definition is that genuine nonparticipation is ignored, but in wild dogs true nonparticipation seems rare (cf. Scheel & Packer 1991, for lions). Although we measured hunting group size as the number of adults, daily mean food intake rates were estimated by dividing the mass of prey killed by total pack size, including yearlings and pups. Pups were weighted by a factor of 0.5 (following Mills & Biggs 1993; also see Packer et al. 1990). This yields a measure of the food intake that a pack’s hunting force provided for themselves and their dependents. We estimated mass of kills using published figures (Sachs 1967; Blumenschine & Caro 1986). Our basic analyses are based on prey mass killed (except where noted), which could be converted to edible mass using weighting factors, but similar percentages of the mass (60–80%) of most carcasses was edible. We re-analyze some of the data after converting prey mass killed to energy obtained. Data points were calculated on a daily basis (e.g., daily mean mass of kills, daily mean chase distance) and variances were calculated using daily means (N ⳱ 266 days) as data points. We used temporal autocorrelation to determine that successive days provided statistically independent points.
4.2 Hunting and Foraging Success Selous wild dog packs made from 0 to 16 chases per day (mean ⳱ 4.2 Ⳳ 0.2; N ⳱ 266 days), and killed from 0 to 10 animals per day (mean ⳱ 1.8 Ⳳ 0.1, N ⳱ 266). Hunting success (kills/hunt) was 44% (range 0–100%) when calculated using data only from complete days, and 45% (range 0–100%) using all observations. Prey mass ranged from 0.5 to 208 kg (mean ⳱ 48.5 Ⳳ 2.15, N ⳱ 384). Feeding duration ranged from 1 to 312 minutes (mean 35.3 Ⳳ 2.1, N ⳱ 357). Chase distances ranged from 50 m to 4.6 km (mean 0.57 Ⳳ 0.03 km, N ⳱ 775). Successful chases also ranged from 50 m to 4.6 km, but were generally longer (mean ⳱ 0.84 Ⳳ 0.05 km, N ⳱ 304). Packs killed 4.0 Ⳳ 0.35 kg/dog/day (N ⳱ 216), with a range of 0–37.5 kg. Clearly, a wild dog cannot eat 37.5 kg in a day. Actual food consumption averaged between 2.0 and 2.5 kg/dog/day, based on two adjustments to the overall mass killed. First, mass of prey was devalued to reflect that from 20% to 40% is usually not eaten (e.g., large bones, stomach contents). Second, observations of feeding by wild dogs known not to have eaten for several days suggest that adult stomach capacity is roughly 9 kg, so edible biomass in excess of 9 kg/dog was excluded. This capacity is considerably higher than the 4.4 kg reported by Reich (1981) and used elsewhere (Carbone et al. 1997), but Reich simply reported the mass of the stomach contents of two dead dogs. Based on our observations, neither of these dogs had a fully distended stomach when it died.
74 ▪ C H A P T E R 4
Figure 4.2 Prey species ranked by importance to wild dogs, showing the cumulative number of hunts and kills (a), and kilograms of meat killed (b). Packs smaller than the median size (10 adults) relied heavily on impala (c), while median-sized or larger packs relied heavily on wildebeest (d).
4.3 Prey Selection and Hunting Success In 817 hunts, 17 species were hunted (Figure 4.2), namely, impala, Aepyceros melampus (N ⳱ 293 hunts), blue wildebeest (N ⳱ 266), warthog (N ⳱ 88), African hare, Lepus capensis (N ⳱ 32), zebra (N ⳱ 30), duiker, Sylvicapra grimmia (N ⳱ 27), Lichtenstein’s hartebeest, Alcelaphus lichtensteini (N ⳱ 17), eland Taurotragus oryx, common reedbuck, Redunca
C O O P E R AT I V E H U N T I N G ▪ 75 Table 4.1 Profitability of common prey for African wild dogs in Selous Profitability
Species
Hunts
Kills
Percent Success
Mass (kg)
Chase (km)
kg/hunt
kg/km chased
Impala Wildebeest Warthog Afr. Hare Zebra C. Duiker
293 266 88 32 30 27
188 100 31 10 2 16
64% 38% 35% 31% 7% 59%
31.9 92.7 33.8 2.0 157.5 17.6
1.19 0.69 0.31 0.13 1.70 0.53
20.4 35.2 11.8 0.6 11.0 10.4
17.1 51.0 38.1 4.8 — 19.6
Total
736
347
47%
48.8
0.88
22.9
29.8
Weighted Mean1 1
Means were weighted using number of kills or chases for each species.
arundinum, buffalo, Syncerus caffer, greater kudu, Tragelaphus strepsiceros, bushbuck, Tragelaphus scriptus, sable antelope, Hippotragus niger, bushpig, Potamochoerus porcus, waterbuck, Kobus ellipsiprymnus, banded mongoose, Mungos mungo, and yellow baboon, Papio cyanocephalus (N ⱕ 10 hunts each). In a sample of 370 identified kills, 10 prey species were killed (Figure 4.2). These were impala (N ⳱ 188 kills), blue wildebeest (N ⳱ 100), warthog (N ⳱ 31), common duiker (N ⳱ 16), Lichtenstein’s hartebeest (N ⳱ 15), African hare (N ⳱ 10), common reedbuck (N ⳱ 4), zebra (N ⳱ waterbuck (N ⳱ 1), bushbuck (N ⳱ 1), eland (N ⳱ 1 calf) and banded mongoose (N ⳱ 1). As Figure 4.2 shows, three species (impala, wildebeest, and warthog) were hunted and killed far more frequently than all other prey combined (see Chapter 5). The three ungulate species that were hunted but not killed were either much larger than the range of normal prey (buffalo), had unusually dangerous horns (greater kudu and especially sable), or were not common relative to other ungulates (greater kudu and especially sable). Yellow baboons were also not killed, but appeared to be hunted at least partly in play. Table 4.1 shows hunting success, chase distance, and two measures of profitability (mass killed per hunt, and per kilometer chased) for prey species hunted on more than 25 occasions. Impala were hunted most often (40% of the total), killed most often (54% of the total) and yielded the highest hunting success (64%). Zebra provided the most mass per kill, but were rarely killed, with a probability of killing (7%) far lower than other species (minimum of 31%). Excluding zebra, wildebeest were the heaviest prey killed (mean of 93 kg). African hares were killed with the shortest chases (mean of 130 m), but yielded little food (2 kg).
76 ▪ C H A P T E R 4
Combining these relationships shows that wildebeest yield the greatest mass per hunt and the greatest mass per kilometer chased (Table 4.1). Wildebeest accounted for the greatest mass killed by a wide margin (Figure 4.2: 9,270 kg of wildebeest vs. 5,997 kg of impala) and thus are the most important prey. Wildebeest are hunted 3 to 10 times more frequently than all prey other prey except impala (Figure 4.2). Impala are hunted most frequently of all (Figure 4.2), despite ranking second in mass/hunt and fourth in mass per kilometer chased. However, examining the preferences of large packs (equal to or greater than 10 adults, the median) and small packs independently shows that small packs hunt and kill impala more often than wildebeest (Figure 4.2c), while large packs hunt and kill wildebeest more often than impala (Figure 4.2d). The overall preference for impala also results from different population densities of prey species (impala are common, see Chapter 5). Finally, seasonal patterns of reproduction by prey create peaks in the availability of vulnerable young (which are highly preferred by wild dogs), and there is some variation among prey species in the timing of the birth pulse. Prey availability and prey choice are discussed in detail in the next chapter.
4.4 Cooperative Hunting Behavior Coordination between the members of an African wild dog pack is seen throughout a hunt (Figure 4.3), and effectiveness appears to depend on the number of cooperating hunters at several stages. The members of a pack almost invariably go through an intense greeting ceremony or rally just prior to a period of hunting (Figure 4.3a). The rally appears to ensure that all pack members are awake, alert and ready to hunt simultaneously, prior to trotting in search of prey (Estes & Goddard 1967; Malcolm 1979). Once on the move, pack members usually trot or canter together at about 10 km/h, spread over 10 to 100 m (Figure 4.3b). Upon sighting prey, a pack often does not hunt. If the pack hunts, small prey (e.g., impala or duiker) flee immediately, but large prey (e.g., wildebeest) often stand in a defensive “pinwheel,” facing outward, charging and using their horns to defend themselves (Figure 4.3c). Juveniles keep to the center of the pinwheel: We saw some cases in which young wildebeest tried to climb over one another to stay in the middle of the herd. Well-armed prey (e.g., warthog, greater kudu males) may also stand and defend themselves rather than fleeing, even when solitary. Prey taller than the dogs will move into water to stand at bay, though this usually isn’t an option (Figure 4.3d, h). When faced with a defensive formation, wild dogs simultaneously attack from several directions (Figure 4.3c). If the prey defend themselves well, the dogs often depart after testing the prey for 10 seconds to 5 minutes. The apparent goal of testing and attack is to force some or all of the herd to run,
C O O P E R AT I V E H U N T I N G ▪ 77
Figure 4.3 Sequence of hunting behavior in communal hunts by African wild dogs: (a) Social rally, which precedes morning and evening hunts. (b) Group travels slowly in search of prey (compare gait with e). (c) Wildebeest herd in “pinwheel” formation, attacked from several angles. (d) A yearling wildebeest stands at bay in a water hole, taking advantage of greater height than the attacking wild dogs. (e1) Gait in full speed chase at 60 km/h (compare with b). (e2) Full speed chase, closing in on an adult wildebeest. (f) The grab. (g) Dogs follow several lines of pursuit: Should the herd swing left, the wild dog in the foreground is positioned to intercept the wildebeest calf.
78 ▪ C H A P T E R 4
Figure 4.3 Continued (h) Several wild dogs distract an adult female wildebeest while another attacks her calf. (i) Well-armed prey such as warthogs can only be killed if the head is restrained. (j) Large prey such as yearling or adult wildebeest can only be killed if restrained by more than one dog. Nose holds and leg holds are common. (k) Multiple kills are common for large packs. Here, three wildebeest are being killed within 10 meters of one another, by a pack of 20 adults. (l) A single spotted hyena appropriates the remains of a wildebeest carcass from two feeding wild dogs.
C O O P E R AT I V E H U N T I N G ▪ 79
Figure 4.3 Continued
thus increasing their vulnerability. Simultaneous attacks are effective because one wild dog can incite a charge, then packmates rush behind the charging prey to separate it from the herd. Once one or a few prey begin running, the entire herd often bolts, and a full-speed chase ensues (at 40–60 km/h) (Figure 4.3e, f). Especially in woodland, prey do not run in a straight line. For example,
80 ▪ C H A P T E R 4
Figure 4.3 Continued
they follow lines of low resistance through trees, uneven ground, and water holes. Prey with territories that are small relative to the length of a chase (especially duiker) often attempt to circle. Individual wild dogs pursuing a prey animal do not all follow the same line (Figure 4.3g). Together, these patterns sometimes result in one or more wild dogs intercepting a prey animal after a shortcut, whether intentional or not (Estes & Goddard 1967; cf. Fanshawe & Fitzgibbon 1993).
C O O P E R AT I V E H U N T I N G ▪ 81
Figure 4.3 Continued
Once a prey animal has been caught, pack members cooperate in pulling it to a halt, or in occupying the animal’s attention by feinting from the front, while others attack from behind and begin disemboweling. Several dogs may attack and distract a female while packmates attack its dependent offspring (Figure 4.3h), as in spotted hyenas (Kruuk 1972) and cheetahs (Caro 1994). Cooperation is important in restraining the head of dangerous prey (e.g., warthog; Figure 4.3i), or large prey (e.g., wildebeest; Figure 4.3j). Because
82 ▪ C H A P T E R 4
Figure 4.3 Continued
killing can take several minutes and prey remain dangerous, restraint of the head is important to protect the dogs involved in disemboweling from being butted or hooked by horns or tusks. It is not unusual for hunting dogs to receive deep cuts, broken teeth, and injured limbs. Although simultaneous chases and kills are not “cooperative,” they are a
C O O P E R AT I V E H U N T I N G ▪ 83
Figure 4.3 Continued
benefit of hunting in larger groups. In Selous, large packs often chased several prey animals from a single herd. Simultaneous chases resulted in simultaneous kills of up to six wildebeest or seven impala (Figure 4.3k and see below). Multiple kills usually included juveniles—either a mother and her offspring or several juveniles from one herd. Finally, wild dogs cooperate in defense of their kills from other carnivores. In Selous, competition at kills was not intense, and came primarily from spotted hyenas (Figure 4.3l). Spotted hyenas were present at 76 wild dog kills (18% of all kills), but appropriated only 14 kills (2%). These percentages form a sharp contrast to Serengeti National Park, where hyena group sizes are larger (Hofer & East 1993a,b). There, hyenas were present at 86% of wild dog kills (excluding gazelle fawns), and the duration that wild dogs retained their kills depended on the number of each species present (Fanshawe & Fitzgibbon 1993). In Selous, competition at wild dogs’ kills also came occasionally from lions (4 interactions, 4 kills lost; ⬍ 1% of all kills), and other packs of wild dogs (2 interactions, both kills lost to larger pack; ⬍ 1% of all kills). Wild dogs rarely scavenged. Three times, wild dog packs attacked adult leopards (two females and one male) until they fled into a tree, appropriating one impala carcass and one wildebeest carcass. One adult wildebeest was taken from a lion, and four kills were taken from spotted hyenas (two impala, two of unknown species). Two additional kills were scavenged from unidentified carnivores, and three were scavenged from poachers’ snares.
84 ▪ C H A P T E R 4
4.5 Characteristics of Kill Sites Habitats are patchy in northern Selous, on the scale relevant to a wild dog hunt. Consequently, a chase that begins in a patch of open short grass may end in a riverine thicket, or vice versa. For 873 chases, we scored the site at which the chase ended as thicker, thinner, or similar to the habitat through which the prey was chased. Of 115 chases that ended at a thicker site, 85% end in kills. Of 758 chases that ended at thinner or “no-change” sites, a much lower percentage (37%) ended in kills (see Table 3.7). These data suggest that pursuing prey into thick habitats increased the likelihood of making a kill (χ2 ⳱ 91.7, P ⬍ 0.001). Encountering thicker habitat favored the dogs for two reasons. First, the dogs are more agile than most of their prey, so they gain ground when dodging and hurtling are required. Second, prey often hesitate when they encounter an obstacle. A fleeing antelope must make a decision as to the best route around an obstacle, and sometimes slows down while making the decision. On the heels of prey in a wooded habitat, a wild dog usually follows the exact line taken by the prey, rather than making independent decisions. This helped the dogs to catch prey, particularly impala. Impala (especially juveniles) often hesitated or even stopped when they encountered fallen trees, dense thickets of whistling thorn, a wall of long grass, or broken ground associated with Sporobolus tussock grass. On several occasions, we saw wild dogs catch stationary impala fawns at such barriers (sometimes slowing to a walk as they reached the impala), under circumstances that made it unlikely that the fawn was simply exhausted. It is likely that decisions about the direction of pursuit are largely in the hands of the prey, because they lead the chase. When hunts split into multiple chases, one subgroup of dogs sometimes intercepted prey from the front, but this was not common. The fact that the direction of flight is determined by the prey may explain why relatively few chases (13%) ended in thick habitats, which favored the hunters.
4.6 Quantitative Effects of Pack Size on Hunting Benefits and Costs Hunting success significantly increased as adult pack size increased (Figure 4.4), ranging from 42% in packs of three adults to 67% in packs of 20 adults (b ⳱ 1.64 Ⳳ 0.61, r2 ⳱ 0.16, t ⳱ 2.70, P ⳱ 0.007). The mean mass of prey killed significantly increased with adult pack size (Figure 4.5), from 21 kg in packs of three to 68 kg in packs of 20 (b ⳱ 2.79 Ⳳ 0.14, r2 ⳱ 0.45, t ⳱ 6.96, P ⬍ 0.001). The distance chased in a successful hunt significantly decreased as group
C O O P E R AT I V E H U N T I N G ▪ 85
Figure 4.4 Hunting success (measured as kills/hunt) significantly increased as the number of adult wild dogs increased. Small packs ⬍ median of 10 adults; Large packs ⱖ median of 10 adults.
Figure 4.5 Mean mass of prey killed significantly increased as the number of adult wild dogs increased. Points are group-size means, with sample size for each point and OLS regression.
86 ▪ C H A P T E R 4
Figure 4.6 Mean distance of successful chases significantly decreased as the number of adult wild dogs increased. Small packs ⬍ median of 10 adults; Large packs ⱖ median of 10 adults.
Figure 4.7 Mean number of prey killed simultaneously increased significantly as the number of adult wild dogs increased. Point size is proportional to the number of observations.
C O O P E R AT I V E H U N T I N G ▪ 87
Figure 4.8 The number of hunts per day significantly increased as the number of adult wild dogs increased. Points are group-size means. Dashed lines are 95% confidence limits.
size increased (Figure 4.6), from 1.1 km in packs of three adults to 0.5 km in packs of 20 adults (b ⳱ ⳮ33.7 Ⳳ 9.37, r2 ⳱ 0.20, t ⳱ 3.60, P ⬍ 0.001). The number of animals simultaneously killed increased with group size (Figure 4.7: b ⳱ 0.02 Ⳳ 0.005, r2 ⳱ 0.24, t ⳱ 4.80, P ⬍ 0.001). In summary, larger packs were more likely to kill in a given hunt, killed heavier prey with shorter chases, and killed more members of the herds they chased. Nonetheless, larger packs made more chases per day (Figure 4.8: b ⳱ 0.23 Ⳳ 0.05, r2 ⳱ 0.28, t ⳱ 4.66, P ⬍ 0.001), with the number of chases/day doubling over the observed range of adult pack sizes. Might larger groups be required to work harder to meet their greater absolute food needs, despite the advantages shown above? This question can be resolved only with data on per-capita food intake.
88 ▪ C H A P T E R 4
4.7 Optimal Hunting Pack Size The standard measure of foraging success among large carnivores has been kilograms eaten or killed per animal per day (Packer et al. 1990; Stander 1992; Fanshawe & Fitzgibbon 1993; Stander & Albon 1993). Figure 4.9a shows the relationship of this measure of foraging success to pack size (y ⳱ 6.0–0.6x Ⳮ 0,04x2, F ⳱ 54.4, df ⳱ 3,246, P ⬍ 0.001). As pack size increases, kg killed/dog/day initially decreases, reaches a minimum at 8 to 9 adults, and subsequently increases. To maximize kg killed/dog/day, selection would favor wild dogs that avoided adult pack sizes 7 to 11, particularly by forming packs larger than this range. But Selous wild dogs were most often found in packs of 10 adults (mean Ⳳ SD of normal approximation for frequency distribution ⳱ 9.8 Ⳳ 3.9; Figure 4.11), diametrically opposing the predictions shown in Figure 4.9a. Using this measure of foraging success, wild dog pack size appears unrelated to cooperative hunting. Measuring foraging success as kg/dog/day does not incorporate variation in hunting effort. Hunting effort can be crudely incorporated by measuring foraging success as kg/dog/hunt. The relationship between kg/dog/hunt and pack size is shown in Figure 4.9b (y ⳱ 2.1–0.13x Ⳮ 0.005x2, F ⳱ 34.2, df ⳱ 3,246, P ⬍ 0.001). Similar to kg/dog/day, this measure of hunting success initially decreases as pack size increases. However, kg/dog/hunt does not reach its minimum until a pack size of 14, and maximal foraging success was obtained by dogs in the smallest packs observed. To maximize kg/dog/ hunt, wild dogs would be selected to live in the smallest packs possible. Hunting costs can be incorporated into foraging success in a more precise way by measuring kg/dog/km chased per day, or kg/dog/km traveled per day. Both of these measures improve on kg/dog/hunt by incorporating variation in effort expended in a hunting bout. “Chasing” refers only to focused, highspeed pursuit of prey (Figure 4.3e, f). “Traveling” refers to all movement, including both chases and slow-paced searching (Figure 4.3b). Kilograms/ dog/km chased has the advantage of focusing narrowly on effort that is unequivocally directed to hunting. Kilograms/dog/km traveled has the advantage of being a more inclusive measure of cost, but it might include costs of travel that was, in fact, directed to another purpose. In practice, wild dog packs rarely travel without hunting, and we consider all movement to be hunting-related, though travel undoubtedly serves other functions simultaneously (e.g., territorial defense, assessing dispersal opportunities). As adult pack size increases, kg killed/dog/km traveled also increases, throughout the observed range of pack sizes (Figure 4.9c: y ⳱ 0.15 Ⳮ 0.06x–0.001x2, F ⳱ 16.3, df ⳱ 3,244, P ⬍ 0.001). Using this measure of foraging success, selection would favor wild dogs that foraged in packs as large as possible, with other factors setting an upper limit on pack size.
C O O P E R AT I V E H U N T I N G ▪ 89
Figure 4.9 The relationship of foraging success to pack size, shown by nonlinear regressions. Four measures of foraging success are shown as the dependent variable: (a) kg killed/dog/day, or gross benefit, r 2 ⳱ 0.20. The other three dependant variables are measure of efficiency: (b) kg killed/dog/hunt, r 2 ⳱ 0.28; (c) kg killed/dog/km traveled, including search, r 2 ⳱ 0.21; (d) kg killed/dog/km chased in full-speed pursuits r 2 ⳱ 0.16.
As adult pack size increases, kg killed/dog/km chased increases until reaching an intermediate optimum at 12–14 adults and subsequently decreasing (Figure 4.9d: y ⳱ ⳮ1.05 Ⳮ 0.99x–0.04x2, F ⳱ 10.57, df ⳱ 3,194, P ⬍ 0.001). By this measure of foraging success, selection acting on hunting alone would favor life in intermediate pack sizes. Optimal pack size is slightly higher than the observed peak in the pack size distribution (12–14 versus 10 adults; Figures 4.9d and 4.10).
4.8 Net Rate of Food Intake versus Efficiency Kilograms killed/individual/day has been the typical currency for analyses of cooperative hunting and group size in social carnivores (Fuller & Kat 1990; Packer et al. 1990; Stander 1992; Fanshawe & Fitzgibbon 1993; Caro 1994). As discussed at the outset of this chapter, kg/individual/day is a measure of gross benefit, but a correct currency should incorporate the costs of hunting.
90 ▪ C H A P T E R 4
Figure 4.10 Foraging success of African wild dogs in the Selous Game Reserve as a function of adult pack size. Gross benefits are kJ obtained from prey eaten, and costs are kJ expended in travel, including search costs (analogous to Figure 4.9c). Net benefit/dog/day was maximized in the largest packs. Points are means, with point size proportional to the number of observations for each pack size. Dashed line shows linear OLS regression.
The analyses just presented incorporate costs by calculating the ratio of benefit to cost, or “efficiency” (Stephens & Krebs 1986). Packer & Caro (1997) pointed out that our use of efficiency could be affected by a well-known problem: The most efficient strategy (maximizing b/c) does not necessarily maximize the net rate of food intake, because efficiency does not distinguish between small gains at small cost and large gains at large cost (Stephens & Krebs 1986). Packer & Caro gave a hypothetical example in which hunters in an optimal group expend 1,000 calories to obtain 5,000 calories, netting 4,000 calories. Hunters in a putatively suboptimal group obtain 500 calories but expend only 5 calories (1⁄200 of effort in the alternative group size). Packer & Caro note that the suboptimal group attains a smaller net benefit, even though it is more efficient (500/5 ⳱ 100, while 5,000/1,000 ⳱ 5). While this example is algebraically correct, it depends on the implicit assumption that the “suboptimal” group is somehow constrained to make a much smaller hunting effort than the “optimal” group. This constraint on effort simply isn’t realistic when the “strategies” under comparison are actually just different group sizes. Nothing in the time budgets of wild dogs or other carnivores suggests that pack size limits the effort that individ-
C O O P E R AT I V E H U N T I N G ▪ 91 Table 4.2 Conversion of kilograms of prey killed to kilojoules of energy obtained for Selous wild dogs Prey Species Impala Mean mass killed 31.9 % of carcass (% eaten): flesh 53.7 (100) viscera 19.9 (68) skin 7.7 (50)a bone 18.8 (0)a Energy (kJ) per kg: flesh 11,640b viscera 6,190b, c skin 6,190b,c kJ/carcass 233,000 kJ/kg killed 7,304 Conversion factor (mean kJ/kg killed) ⳱ 6862 kJ/kg killed
Wildebeest 92.7 43.7 27.7 8.5 20.1
(100) (77) (0)a (0)a
11,640b 6,190b,c n/a 595,000 6,419
a
These percentages vary substantially from kill to kill. The energy obtained is probably not great, though bone and skin may be important for specific nutrients. b From beef carcasses of the lowest grade reported by the U.S. Department of Agriculture. c Mean value for ten organs (heart, kidney, lung, liver, intestine, brain, spleen, thymus, tongue, and pancreas) usually eaten by wild dogs.
uals can invest in hunting, and equal effort would yield a much larger net benefit for hunters in the more efficient but putatively suboptimal group size. Nonetheless, to avoid the assumptions implicit in the use of efficiency (Stephens & Krebs 1986), we converted the costs and benefits of wild dog hunts into units of energy (kJ), allowing us to test how net benefit varies across pack sizes. We determined kJ/kg of prey killed (Table 4.2) for the two most important prey (wildebeest and impala; Figure 4.2), using measurements from Blumenschine & Caro (1986) for the proportion of wildebeest and impala carcasses made up of flesh, viscera, bone, and skin. Mass of kills and the proportions of flesh, viscera, bone, and skin that were eaten came from observations in Selous. We used data from the USDA (1990) on kJ per kg of viscera and “separable meat and fat” from low-grade beef carcasses. The conversion yielded 6419 kJ/kg of wildebeest killed and 7304 kJ/kg of impala killed (Table 4.2). We used the mean of these values (6862 kJ/kg killed) to convert kg killed/dog/day into kJ obtained/dog/day. To convert the costs of hunting into units of energy, we used an allometric regression relating oxygen consumption (ml O2/s) to body mass and speed (Taylor et al. 1982). Oxygen consumption can be converted to energy expenditure (1 ml O2 ⳱ 20.1 J), if anaerobic contributions to energy production are negligible (Calder 1984). The allometric equation for carnivores is:
92 ▪ C H A P T E R 4
E/M ⳱ 6.47 Mⳮ0.289 Ⳮ 10.23 Mⳮ0.310 v
(Eq. 4.1)
where E ⳱ energy expended (watts), M ⳱ body mass (kg), and v ⳱ speed (m/s). This allometric regression is based on low speed (ⱕ 5 m/s) treadmill tests with captive animals. It can be used to convert time or distance of lowspeed travel (4 to 6 m/s for wild dogs) to units of energy, but nothing is known about the costs of high-speed chases (up to 17 m/s for wild dogs). Even slight nonlinearity in the allometric relationship would invalidate an extrapolation to high speeds, and high-speed energetics are complicated by anaerobic energy production. Thus, Equation 4.1 can be used to convert total travel costs (Figure 4.9c) to units of energy, but cannot be used to convert high-speed chase costs (Figure 4.9d). For wild dogs, Equation 4.1 yields a best estimate of 3042 kJ/h of travel. An alternative method is to use oxygen consumption data from wolves and interpolate a correction for body mass (using data from Taylor et al. 1982), which gave a similar estimate (3329 kJ/ h). The lower and upper 95% confidence limits for the allometric exponents yield 2162 kJ/h and 4190 kJ/h (71% and 138% of the best estimate, respectively). This variation is sufficient to alter conclusions. Also, energy expenditure during a treadmill run is less than expenditure over real terrain with turns, accelerations, and the effort of subduing prey. With these caveats, we used the allometric conversion (3042 kJ/h) to estimate minimum energy expenditure per hour of travel. Net kJ obtained/dog/day increased significantly as a function of adult pack size (Figure 4.10: linear regression, t10 ⳱ 3.16, P ⳱ 0.01, r2 ⳱ 0.53). Using this measure of foraging success, selection would favor wild dogs that lived in packs as large as possible, with other factors setting an upper limit on pack size, as we concluded from our original analysis of daily kg killed/ dog/km traveled (Creel & Creel 1995b). Although the peak, at pack size 20, is much higher than the other peaks, there is substantial variation in net benefits at smaller pack sizes. For example, the payoff at pack size 10 is several times greater than the payoff at pack size 8. A positive relationship between net benefit and pack size in not surprising, given that hunting success, prey mass, and the number of simultaneous kills increase with pack size, and chase distance decreases (Creel & Creel 1995b). Viewed in detail (Figure 4.10) net benefit shows a local optimum at pack size 3, declines through pack size 8, rises to a local optimum at pack size 10, declines through pack size 14, then rises to a maximum at pack size 20. Shifts in prey selection produce these patterns (Figure 4.11). Packs of 3 to 8 prey primarily on easy (small) prey, but divide the prey among an increasing number of stomachs, so that net benefit declines. Packs of 9 to 14 prey more heavily on prey of medium difficulty and size, taking fewer easy prey (Figure 4.11), again dividing the same foods among an increasing number of stomachs. Packs of 15 or more take the highest percentage of medium-sized prey, and the fewest easy prey, and also benefit from a high frequency of
C O O P E R AT I V E H U N T I N G ▪ 93
Figure 4.11 Patterns of prey selection for small, medium, and large wild dog packs. As pack size increases, easy prey form a smaller part of the diet, and prey of medium size form a larger part of the diet. The hardest class of prey is not often killed by packs of any size. In addition to larger prey, multiple kills are common for large packs (inset). Easy (small) prey includes impala fawns and yearlings, wildebeest calves, warthog piglets, African hares, duikers, bushbuck, and juveniles or yearlings of all other species. Medium prey includes adult female impala, yearling wildebeest, and yearling warthogs. Hard (large) prey includes adult male impala, and adult wildebeest, warthog, zebra, hartebeest, reedbuck and waterbuck. Reedbuck are classified as hard (despite being small) because wild dogs cannot see or run well in the long grass that reedbuck favor. Warthogs are classified as hard (despite small-medium size) because their defensive slashing is very effective compared to other prey.
multiple kills (Figure 4.11). Interestingly, pack size had little effect on the frequency of killing the hardest class of prey (adult male impala and adults of either sex of wildebeest, warthog, zebra, hartebeest, waterbuck, and reedbuck). These hard prey were rarely killed by any pack, and we suspect that they are killed only when a compromised individual (e.g., sick, weak, heavily pregnant) is encountered by chance. The odds of such an encounter are probably not affected by pack size. Figure 4.12 shows these patterns in greater detail, presenting the proportions of easy, medium, and hard age-sex
94 ▪ C H A P T E R 4
Figure 4.12 Age-sex composition of kills by different pack sizes for the three most common prey: wildebeest, impala, and warthog. Using the categories of Figure 4.11, Filled bars ⳱ easy prey; Open bars ⳱ medium prey; Hatched bars ⳱ hard prey.
classes of prey for the three most important prey (impala, wildebeest, and warthog). For other species, too few kills were tallied to allow for a meaningful breakdown. Overall, the regression of net rate of energy gained on pack size is positive and significant (P ⳱ 0.01, r2 ⳱ 0.53), supporting the hypothesis that cooperative hunting is a force in the evolutionary maintenance of group living in wild dogs. We could compare the observed distribution of pack sizes
C O O P E R AT I V E H U N T I N G ▪ 95
to that predicted by Figure 4.10. Most analyses of cooperative hunting (including ours, Creel 1997) have done this, but in retrospect we think it is not particularly useful. Foraging success is one variable that affects fitness, but it is virtually certain that processes unrelated to hunting also affect the fitness payoffs to living in different pack sizes. For example, larger packs are better able to defend their kills against kleptoparasitism by spotted hyenas (Fanshawe & Fitzgibbon 1993). Conceptually, it is clear that the frequency distribution of pack sizes will be shaped by selection acting simultaneously on the many components of fitness that are affected by pack size. Comparing the pack size distribution to the predictions of a single variable is a weak test of that variable’s importance (even if there is a nice match), unless the predictions for other variables are known. Implicitly, the null hypothesis for this comparison includes an assumption that no other components of fitness vary with group size, or that other effects cancel out. Because of this problem, we think the most reasonable test is to identify a variable that is likely to affect fitness (such as foraging success or predation risk), quantify the variable, test whether it varies significantly across group sizes, and stop there. This approach makes no assumptions about the relationship of other variables to pack size. An alternate (and extraordinarily hard) approach is to identify all the variables likely to affect fitness, quantify each as a function of group size, convert them to a common currency (or establish weighting factors for the impact of each variable on fitness), and make a joint prediction to which the frequency distribution of group sizes could be compared. The latter approach would be a life’s work for many species, certainly for wild dogs. Our analysis suggests that cooperative hunting favors living in packs as large as possible (up to 20 adults), with other factors setting an upper limit on pack size. Reproductive suppression of social subordinates is one factor that is likely to limit pack size. In a wild dog pack, only the socially dominant individuals maintain normal reproductive hormone levels (Chapter 9), and subordinates are unlikely to reproduce (Girman et al. 1997; Chapter 8). As pack size increases, the breeding queue gets longer, reducing the odds of surviving long enough to become a breeder. Pack size will eventually cross a threshold at which subordinates would do better to disperse in search of breeding opportunities. Consistent with this argument, young adults are the most common dispersers, and dispersal is most likely from large packs that have just recruited a large cohort of same-sexed young adults (McNutt 1996; Chapter 8).
4.9 Effects of Group Size Unrelated to Hunting Our analyses of optimal group size address only the effects of communal hunting. Although our results confirm that communal hunting favors sociality in wild dogs, factors completely unrelated to hunting are also likely to affect pack size. We do not suggest that other group-level activities (e.g.,
96 ▪ C H A P T E R 4
group defense of territories or offspring; Packer et al. 1990) or patterns of relatedness (Giraldeau & Gillis 1988; Giraldeau & Caraco 1993) are unimportant. Preliminary evidence suggests that there are other benefits of group living for Selous wild dogs. First, clashes between packs were won by the larger pack in 11 of 11 cases (Chapter 3). In two additional clashes between packs of equal size, each retreated once. Second, large groups produce large samesexed cohorts, which may confer advantages in dispersal (Chapter 8). Groups of transient females can take over existing packs by evicting resident females, and numbers are likely to affect the outcome of takeover attempts. Finally, large packs may be better at defending their pups from predation. For example, a pack of 16 attacked an adult male lion that was stalking its pups, and drove the lion away without casualties. Better quantitative data are needed to test whether these and other potential benefits vary with pack size, but it is clear that some combination of processes creates an association between pack size and reproductive success (Chapter 7). No single ecological or behavioral variable is likely to explain variation in fitness across pack sizes. In debating the effects of cooperative hunting in lions and cheetahs, Packer & Caro (1997) argued that “grouping patterns are better explained by alternative hypotheses than by the consequences of group foraging” (p. 1318). In our view, this is not the point, when asking if cooperative hunting itself is important. Our data address only the energetics of hunting; selection on the risks of injury during hunting might not act in parallel. Though life-threatening injuries from hunting are rare, selection on traits to avoid them might be strong. The difference between a 2% chance and a 6% chance of having your head kicked in by a wildebeest is probably of evolutionary significance, even though it is difficult to quantify.
4.10 Variance in Foraging Success Risk-sensitive foragers should hunt in group sizes that depend both on mean foraging success and its variance, to minimize the risk that food intake will dip below the starvation level (Pulliam & Caraco 1984; Houston et al. 1988; Mangel & Clark 1988). Most stochastic models of risk-sensitive foraging and group size in large carnivores depend on estimates that are poorly known for wild dogs (e.g., daily requirements, toleration of starvation, stomach capacity), so we have not applied them. However, risk sensitivity will modify the optimal group size only if variance in foraging success is affected by group size (Pulliam & Caraco 1984). For Selous wild dogs, variance in foraging success did not correlate with group size (NS for all measures of foraging success; e.g., for kg/dog/km traveled, rs ⳱ 0.20, P ⳱ 0.69), nor were nonlinear associations apparent. Thus the results of
C O O P E R AT I V E H U N T I N G ▪ 97
simple optimality and stochastic models are likely to coincide (as in Serengeti lions; Packer et al. 1990).
4.11 Other Wild Dog Populations Wild dogs are generally regarded as efficient hunters, and their high hunting success in Selous (44%) parallels that recorded in other populations, which ranges from 39% to 85% (Estes & Goddard 1967; Kruuk & Turner 1967; Schaller 1972; Malcolm & van Lawick 1975; Fanshawe & Fitzgibbon 1993; Fuller & Kat 1993). Some of the variation in hunting success among studies is probably due to small sample sizes (the highest and lowest values reported were based on ⱕ 30 hunts). Pooling data from four wild dog studies in the Serengeti ecosystem, hunting success was also 44% (N ⳱ 666 hunts: Schaller 1972; Malcolm & van Lawick 1975; Fanshawe & Fitzgibbon 1993; Fuller & Kat 1993). The match with Selous is surprising, given substantial differences in the prey set available and the physical environment. Despite similar hunting success, energetic returns might differ substantially between populations, due to variation in prey size or hunting effort. The range of prey species hunted and killed is broader in Selous than has been reported for most other populations (e.g., Malcolm & van Lawick 1975; Fuller & Kat 1990, 1993). This difference is probably partly due to sample size differences. All of the prey species killed in Selous have been recorded in at least one other study (see Pienaar 1969, who summarizes results for 4,406 carcasses fed on by wild dogs in Kruger National Park). Of the species hunted but not killed in Selous, two have not previously been reported (bushpig and yellow baboon). Two studies of wild dogs in the Serengeti ecosystem have related food intake to pack size. A two-month study of one pack in Aitong did not detect a significant change in kg killed/dog/day when pack size changed from 29 to 19 dogs (including yearlings and pups), although hunting success fell from 75% to 60% for Thomson’s gazelle, and from 40% to 18% for other prey (Fuller & Kat 1993). Fanshawe & Fitzgibbon (1993) also studied one pack, for two years, and found that a “meat yield index” (similar to kg eaten/dog/ day) was greater for groups than for singletons when hunting wildebeest, but greater for singletons when hunting Thomson’s gazelle. In Serengeti, wildebeest are the most important prey in terms of meat obtained, though Thomson’s gazelles are killed more often (Kruuk & Turner 1967; Schaller 1972; Fanshawe & Fitzgibbon 1993; see Chapter 5). Fanshawe & Fitzgibbon’s data are not directly comparable to ours, because they considered a dog to be hunting only if it was closest to the prey during the chase, or helped to pull the prey down. In Selous, this definition would exclude dogs that were indeed hunting (though perhaps with less effort) but did not succeed (see “Definition of Terms,” above). Though hunting group size has a
98 ▪ C H A P T E R 4
slightly different meaning in Fanshawe & Fitzgibbon’s study, the general pattern across three studies of wild dogs is that larger groups have higher per-capita food intake.
4.12 Communal Hunting and Group Size: Comparisons with Other Species In a recent review of cooperative hunting, Caro (1994), concluded that . . . few studies report per capita foraging returns, but in the majority of those that do, per capita foraging success did not increase with group size. In populations in which it did, grouping patterns did not reflect optimal foraging group sizes. Though limited, current evidence therefore suggests that cooperative hunting is not responsible for group living in any carnivore. (p. 342) For seven species that hunt communally, foraging success has been measured as a function of group size (Table 4.3). These studies provide 14 independent tests of the relationship between group size and foraging success, because some species have been studied in several locations (lions, wild dogs, chimpanzees), or the original data were partitioned by season or group composition (lions: Packer et al. 1990; cheetahs: Caro 1994). With these 14 tests, we ran a quantitative meta-analysis to test the relationship between foraging success and group size, using Fisher’s method for combining probabilities from independent tests (Fisher 1954; Sokal & Rohlf 1995). This meta-analysis is based primarily on data on gross daily intake, because few studies have estimated net rate of intake, and this is a serious limitation. Like all meta-analyses, there is concern that negative results can be difficult to publish, biasing the set of primary studies available for inclusion. With these caveats, the meta-analysis reveals a significant positive relationship between group size and foraging success (Table 4.3: Fisher’s statistic ⳱ 70.7, df ⳱ 28, P ⬍ 0.001). Though small, this set of species includes stalkers and coursers that hunt on land, in water, in the trees, and in the air, pursuing their prey by swimming, climbing, running, and flying. A positive result across such dissimilar species suggests that cooperative hunting is beneficial under a broad range of ecological conditions. There are too few studies to make strong arguments about differences among species, but these studies are broadly consistent with published hypotheses that the benefits of cooperation are more consistent for hunters that do not rely on stealth, and under ecological conditions where solitary individuals have low foraging success (Kruuk 1975; Packer & Ruttan, 1988; Boesch 1994). Studies of other carnivores generally suggest that per-capita food intake is higher in groups than in solitaries. Of seven studies that have measured per-
Spearman Spearman Kruskal-Wallis Kendall’s Tau Mann-Whitney
Mann-Whitney
OLS Regression Mann-Whitney Mann-Whitney
Kruskal-Wallis
Spearman Mann-Whitney Mann-Whitney OLS regression
Chimpanzee Chimpanzee Killer Whale Harris’s Hawk African Wild Dog
African Wild Dog
African Wild Dog Female Lion Female Lion
Female Lion
Female Lion Cheetah Cheetah Wolf
0.05 0.7 0.08 0.98
0.48
0.007 0.98 0.002
0.9
0.05 0.55 0.001 0.05 0.02
P
Fisher’s statistic ⳱ 70.7, df ⳱ 28, P ⬍ 0.001
Test
Species
Measure of Intake Net Net Gross Gross Gross
Gross
Net Gross Gross Gross Gross Gross Gross Gross
ln P ⳮ2.9957 ⳮ0.5978 ⳮ6.9078 ⳮ2.9957 ⳮ3.912
ⳮ0.1054
ⳮ4.9618 ⳮ0.0202 ⳮ6.2146 ⳮ0.734 ⳮ2.9957 ⳮ0.3567 ⳮ2.5257 ⳮ0.0202
Tai forest, low singleton hunting success, Boesch 1994 Gombe stream forest, high singleton hunting success, Boesch 1994 Transient killer whales near Vancouver Island, Baird & Dill 1996 Los Medanos; Data from Bednarz 1988 Comparing packs of 1–2 vs. 3Ⳮ, hunts of wildebeest, the most important prey, Serengeti short grass plains; Data from Fanshawe & Fitzgibbon 1993 Comparing packs of 1–2 vs. 3Ⳮ, hunts of Thomson’s gazelle, second most important prey Serengeti short grass plains; Data from Fanshawe & Fitzgibbon 1993 Selous, woodland and wooded grassland, Creel & Creel 1995b; Creel 1997 Serengeti, prey scarce, comparing prides of 1 vs. 2–4, Packer et al. 1990 Serengeti, prey scarce, comparing prides of 2–4 vs. 5Ⳮ, Packer et al. 1990 Serengeti, prey abundant, P value derived from chi-squared approximation for reported H statistic; Data from Packer et al. 1990 Etosha, Stander 1992; lower P value in Stander & Albon 1993 Adolescent females only, Serengeti short grass plains, Caro 1994 Adult males only, Serengeti short grass plains, Caro 1994 Schmidt & Mech 1997; combines data from several ecosystems
Notes & References
Table 4.3 A meta-analysis to test the relationship between foraging success and group size in cooperatively hunting species
100 ▪ C H A P T E R 4
capita food intake (Table 4.3), four have found that intake increases with group size (male cheetahs, Selous and Serengeti wild dogs, Etosha lions), and a fifth study found a tendency to increase that was not significant (wild dogs in Aitong). In Etosha National Park, hunting groups of 2 to 7 lionesses had higher food intake (6 to 12 times higher) than solitary females, when prey was scarce. Gross food intake was highest for pairs, and pairs were the most common hunting group size for Etosha lions (Stander 1992). Two packs of wild dogs in the Serengeti ecosystem obtained more food when hunting in larger groups (Fuller & Kat 1993; Fanshawe & Fitzgibbon 1993; see discussion above). Trios of male cheetahs in Serengeti obtained higher gross intake than pairs or singletons (Caro 1994). Three studies found no gross benefit of hunting in groups (wolves, Serengeti lions and adolescent cheetahs, see below). None of these studies report net benefits of hunting— all rely on gross per-capita food intake. While this limits the strength of the conclusion, the combined data favor the hypothesis that communal hunting favors group living in carnivores (Table 4.3). Packer et al. (1990) showed that in large prides, lionesses with cubs preferred to hunt in groups of five, which maximized foraging success, but lionesses without cubs showed no preference. Lionesses in small prides (ⱕ4)
Figure 4.13 Two male cheetahs cooperating to kill a yearling male wildebeest. Trios of cheetahs cooperated in this manner in 24% of cases in which prey was contacted. Such cooperation might explain why hunting success remains high for trios, though they attack larger prey than singletons or pairs. Redrawn from Figure 10.6 in Caro (1994), who notes “the lack of coordination. The male on the left of the picture is using his weight to twist the victim’s neck and so make it fall to the left, while the male on the right of the picture has bitten the wildebeest’s right hindleg and is pulling it backward in order to make it topple to the right” (p. 267). Compare with Figure 4.14.
C O O P E R AT I V E H U N T I N G ▪ 101
Figure 4.14 Two wild dogs cooperating to subdue a yearling male wildebeest. Wildebeest of this size often break free after being grabbed, and wild dogs cannot kill large prey without this type of cooperation. Note the similarity to cheetahs in Figure 4.13. The wildebeest’s options are more constrained by the presence of two hunters than they would be if one hunter were absent, for both the cheetahs and the wild dogs.
preferred to hunt in groups, though hunting alone would have maximized gross per-capita food intake. Overall, Packer et al. concluded that grouping patterns were not well explained by patterns of foraging success. This conclusion is fairly strongly influenced by the results for small prides, so it is worth noting that pairs made no kills in this study (in seven observation days for pairs), while other data from Serengeti (Schaller 1972) and elsewhere (Stander 1992) suggest that pairs hunt well. The data point for pairs had a strong influence on predictions for hunting group size in small prides (Packer et al. 1990, Figure 1). Perhaps differences in ecological conditions or pride compositions (Stander 1992) altered the payoff to hunting in pairs between the observations of Schaller and Packer. For Serengeti cheetahs, coalitions of three males had higher per-capita food intake than solitaries or pairs (Caro 1994). Cheetahs in large coalitions obtained more food because their hunting success remained high despite their preference for large, difficult prey (wildebeest). Two studies have measured gross per-capita food intake for cooperatively hunting species outside the order Carnivora. For Harris’s hawks (Bednarz 1988) and killer whales (Baird & Dill 1996), individuals that hunt in groups obtain more food than individuals that hunt alone. In both of these species, the most common group size matches the optimal hunting group size. Here again, the data favor the conclusion that communal hunting is an important force in the evolution of sociality and cooperation (Bednarz 1988).
102 ▪ C H A P T E R 4
Boesch (1994) showed that the net benefit of hunting was greater for groups than for singletons in Tai forest chimpanzees. This study is not directly comparable to the others discussed above, because costs and benefits were calculated per hunt, rather than per day (see Stephens & Krebs 1986). The data are also somewhat restricted because they were collected only in the “hunting season” (three months), from one group, intentionally focusing observations on experienced hunters that were judged to be highly motivated to hunt. Thus, Boesch showed that individuals gained a net benefit by hunting in groups, but under narrowly defined conditions. The role of cooperative hunting (and cooperation in general) as a force in carnivore social evolution is still debated (Creel 1997, 2001; Packer & Caro 1997). Before this question is laid to rest we will need data from more species, and multiple studies of single species under varying ecological conditions. For now, some authors (e.g., Caro 1994; Grinnell et al. 1995; Heinsohn & Packer 1995) place considerable emphasis on the details of behavior during group actions, generally concluding that the benefits of cooperation are weak or mixed (Figures 4.13 and 4.14). Others place more emphasis on the outcome of group actions (e.g., Boesch 1994; Baird & Dill 1996; Creel 2001), concluding that cooperation often, but not always, yields a benefit.
5
Prey Selection
During the course of our study, we were often puzzled by the wild dogs’ disinterest in some species that seemed to us to be suitable prey (waterbuck, for example). We were also struck by the differences between large and small packs in the strength of their preferences for common prey, such as impala and wildebeest (Figure 5.1). The Wachunga pack provides a good example. During the first year that we followed them, the Wachungas preyed mainly on impala and we never saw them kill a wildebeest. By the end of the second year, the pack had grown, and they eventually killed a wildebeest, with difficulty. During the third year, they killed wildebeest with great regularity, preferring them to impala. There was no turnover in the adult membership of the pack during this time—the change was due to the addition of new hunters as pups aged and joined the hunting force. Observations such as these made us curious about the effects of pack size and other variables on prey selection. Several processes might affect the set of species that a predator eats, and the proportion of its diet formed by each prey. Profitability (energy obtained per unit of foraging effort) clearly should affect decisions about including a species in the diet (Stephens & Krebs 1986). If a predator seeks to minimize the risk of starvation, then the variance of profitability can also affect prey selection (Stephens & Charnov 1982). The abundance of a prey species relative to alternative prey could affect whether they are hunted or not (Scheel 1993; Huggard 1993), as could weaponry or behavioral defenses that make predation dangerous. Theoretical models of prey selection are highly developed (Charnov & Orians 1973; Pulliam 1974; Charnov et al. 1976; Stephens & Krebs 1986), and a great deal of field research has focused on predator-prey relationships in carnivores. Surprisingly, this combination has produced only one quantitative application of prey choice models to data from large carnivores. Scheel (1993) examined several models with data from Serengeti lions, but we know of no other examples. In this chapter, we examine prey selection by wild dogs in three ways. First, we compare the abundance of prey species to the proportion of kills, hunts, and prey encounters for wild dogs. The proportion of the diet formed by each species is the final measure of prey selection, but a prey type can be important because it is encountered, hunted, or killed nonrandomly, or through a combination of these steps. By looking for nonrandom patterns at each step of the sequence, we gain a better picture of the mechanics of prey selection. For these analyses, we examine large and small packs separately. Most quantitative models of prey selection use the rate of energy intake as
104 ▪ C H A P T E R 5
Figure 5.1 Impala (a) and wildebeest (b) were the two most important prey for all packs in Selous, small packs preferring impala and large packs preferring wildebeest.
P R E Y S E L E C T I O N ▪ 105
their currency. Caro & Fitzgibbon (1992) reviewed interactions between large carnivores and their prey, and noted that “attempts have hardly begun to examine [prey] selection in terms of maximizing energy intake among large carnivores” (p. 121). In this chapter we apply two quantitative models of prey selection. One model assumes that wild dogs maximize food intake relative to foraging effort (Schoener 1971; Charnov & Orians 1973); the other model assumes that wild dogs minimize the risk of starvation (Stephens & Charnov 1982). We test both models by comparing predictions to the observed diet, and to observed probabilities of attacking prey upon encounter.
5.1 Prey Availability and Encounter Rates Many studies have compared the diet of a carnivore with the relative abundance of each prey species (e.g., Pienaar 1969; Viljoen 1993). Such data describe the end-point of prey selection, but do not reveal whether prey selection is affected by nonrandom prey encounter rates, nonrandom decisions to hunt once prey are encountered, or nonrandom success once a hunt has begun. It is widely appreciated that a given prey species might be encountered more (or less) often than would be predicted by its abundance (Greene 1986). Habitat selection, selective searching by predators, or active avoidance by prey could all produce encounter rates that are not proportional to the relative abundance of prey types (Mech 1977; Huggard 1993). Nonetheless, direct measures of species-specific prey encounter rates are rare. Scheel (1993) directly measured prey encounter rates by following Serengeti lions for periods of four days, and found that several important prey species were encountered more often than expected from transect censuses. Stander (1992) directly measured prey encounter rates for lions in Etosha National Park, but did not test whether encounter rates were nonrandom. From Stander’s observations, four species (springbok, zebra, wildebeest, and gemsbok) accounted for 97% of 856 hunts in Etosha. Comparing Stander’s data on encounter rates with estimated population densities for prey species from Gasaway et al. (1991), encounter rates were nonrandom for these four species combined (χ2 ⳱ 33.8, df ⳱ 3, P ⬍ 0.001) and the most commonly killed species was encountered considerably more often than expected by chance (springbok, which provided 59% of 137 kills). Huggard (1993) used less direct methods (intersecting track lines in snow) to conclude that wolves in Banff National Park encountered deer less often than expected from census data, while other prey were encountered randomly. Because deer formed only 5–10% of the wolves’ diet, nonrandom encounter rates probably had little effect on prey selection, overall. In summary, fairly little is
106 ▪ C H A P T E R 5
known about patterns of prey encounter in large carnivores, but limited data suggest that nonrandom encounter rates are not unusual. To test for nonrandom prey encounter rates in Selous, several measures of prey density were available, based on different methods (Table 5.1). Data from four aerial surveys between 1981 and 1991 (two wet season and two dry season) could be extracted for our study area. A potential problem with these data is that aerial surveys undercount different species by different factors, and correction factors have not been determined for Selous. To assess the reliability of the aerial data, we determined the ratio of impala to wildebeest for each census and compared the ratio across surveys. These are the two most abundant prey in the area, and ground counts suggested that the ratio did not change over the six years of our study. In the aerial survey data, impala-to-wildebeest ratios varied from 0.05 (in 1989) to 0.63 (in 1991). This fluctuation suggests that aerial surveys are unreliable when it comes to comparing the relative abundance of ungulates in Selous. A comparison of data from aerial surveys and ground transects suggests that impala are heavily undercounted from the air (Table 5.1) Based on these comparisons, we have used data from 82 ground transects to estimate the relative abundance of prey species. To test if encounter rates were nonrandom, we recorded all encounters with prey herds (N ⳱ 2015, with herds defined to include lone animals, here and elsewhere) for six packs of wild dogs during 243 complete observation days, over a two-year period. We recorded these data by following a pack continuously for 1–13 days, at a distance that did not alarm prey (although we sometimes did alarm prey unintentionally). Defining an encounter with prey was difficult. Visibility varied, so it was not appropriate to use a fixed distance between the dogs and potential prey to define an encounter. It was not always possible to know whether a specific herd was in view of the dogs, so we scored an encounter if the wild dogs reacted to a prey herd within our
Table 5.1 Relative abundance of ungulate prey species in northern Selous from ground transects and aerial surveysA Relative Abundance Species
Ground
81 wet Aerial
81 dry Aerial
89 wet Aerial
91 dry Aerial
Wildebeest Impala Zebra Warthog Others
0.51 0.32 0.09 0.04 0.04
0.59 0.13 0.15 0.04 0.10
0.47 0.18 0.19 0.04 0.11
0.82 0.04 0.09 0.03 0.03
0.30 0.19 0.43 0.02 0.06
A
Aerial survey data come from an unpublished 1981 survey of northern Selous by M. Borner, and TWCM (1990, 1991).
P R E Y S E L E C T I O N ▪ 107
Figure 5.2 Diet of wild dogs in Selous, shown as cumulative percentage of hunts (squares) and kills (circles) with prey ranked in order of importance. (a) Small packs (⬍9 adults). (b) Large packs (ⱖ 9 adults).
view (e.g., alert posture, watching, or walking in direction of herd) or if the prey reacted to the dogs (e.g., alarm calls, watching dogs, moving away). Small Packs Figure 5.2 shows that impala are the most important prey for small packs (less than nine adults). For small packs, the proportion of encounters with impala is not significantly different than the proportion of the ungulate community formed by impala, suggesting that small packs do not seek impala preferentially (Figure 5.3, Table 5.2). If small packs do seek impala herds actively, they do not succeed in raising the encounter rate, because the proportion of encounters tended to be lower than expected. Although the second most important prey for small packs was wildebeest, the proportion of encounters with wildebeest was significantly lower than the proportion of the ungulate community formed by wildebeest. By contrast, warthogs (the third most important prey) were encountered more often than expected. Zebra were encountered in proportion to their abundance. In Table 5.2, other prey is a class that combines species that were killed infrequently (hartebeest, duiker, hare, eland, reedbuck, and waterbuck). Collectively, these species were encountered more often than expected. Overall, patterns of encounter suggest that small packs are generalists in terms of search strategies—their search patterns lead to nonrandomly frequent encounters with uncommon prey.
108 ▪ C H A P T E R 5 Table 5.2 Patterns of prey selectionA,B Proportion of Prey Species
Population
Encounters
Hunts
Kills
0.32a (0.23–0.42) 0.51a (0.41–0.61) 0.04a (0.01–0.10) 0.09a (0.05–0.16) 0.04a (0.01–0.10) N ⳱ 81
0.25b (0.22–0.28) 0.40ab (0.37–0.43) 0.13b (0.11–0.15) 0.07a (0.05–0.09) 0.16b (0.14–0.18) N ⳱ 1627
0.36ab (0.33–0.39) 0.32bc (0.29–0.35) 0.11ab (0.09–0.13) 0.04ab (0.03–0.05) 0.17b (0.15–0.20) N ⳱ 817
0.51c (0.47–0.55) 0.27c (0.23–0.31) 0.08a (0.06–0.11) 0.01b (0.004–0.03) 0.13b (0.10–0.16) N ⳱ 370
Packs Smaller than Nine Adult Dogs Impala 0.32a (0.23–0.42) Wildebeest 0.51a (0.41–0.61) Warthog 0.04a (0.01–0.10) Zebra 0.09a (0.05–0.16) Others 0.04a (0.01–0.10) N ⳱ 81
0.29a (0.26–0.32) 0.29b (0.26–0.32) 0.13b (0.11–0.15) 0.07a (0.06–0.09) 0.22b (0.20–0.25) N ⳱ 971
0.43b (0.39–0.47) 0.21c (0.18–0.25) 0.11ab (0.09–0.14) 0.04a (0.03–0.06) 0.21b (0.18–0.25) N ⳱ 494
0.61c (0.54–0.68) 0.12d (0.08–0.17) 0.09ab (0.06–0.14) 0.005b (0.001–0.02) 0.17b (0.12–0.23) N ⳱ 212
Packs of Nine or More Adult Dogs Impala 0.32a (0.23–0.42) Wildebeest 0.51a (0.41–0.61) Warthog 0.04a (0.01–0.10) Zebra 0.09a (0.05–0.16) Others 0.04a (0.01–0.10) N ⳱ 81
0.17b (0.14–0.20) 0.56a (0.52–0.60) 0.12a (0.10–0.15) 0.07ab (0.05–0.09) 0.08a (0.06–0.10) N ⳱ 656
0.26a (0.21–0.30) 0.50a (0.45–0.54) 0.11a (0.08–0.14) 0.03b (0.02–0.05) 0.10a (0.07–0.13) N ⳱ 323
0.37a (0.29–0.45) 0.47a (0.38–0.55) 0.07a (0.03–0.12) 0.01b (0.001–0.04) 0.08a (0.04–0.14) N ⳱ 158
All Packs Impala Wildebeest Warthog Zebra Others
A
In each table entry, the upper number is the point estimate for the proportion, and the pair of numbers below (in parentheses) are unbiased lower and upper 95% confidence limits. Different superscript letters (a, b, c) indicate a significant difference at ␣ ⳱ 0.05.
B
P R E Y S E L E C T I O N ▪ 109
Large Packs Large packs (ⱖ 9 adults) showed little variability across prey species in the relationship between encounter rates and abundance. For all species except impala, encounter rates were not different from expectations based on relative abundance. Impala, which are the second most important prey for large packs, were encountered less often than expected. Overview It is interesting that large packs encounter impala less often than expected, while small packs encounter wildebeest less often than expected. The shift in encounter rates parallels a shift in the importance of these two prey, so that packs have a low encounter rate with the prey species that is of secondary importance. Although this does not produce a nonrandomly high encounter rate with the prey species of primary importance (as might be expected), it does produce a marked difference between the encounter rates of large and small packs with their two most important prey (compare Figures 5.3a and 5.3b). Of the encounters of large packs with prey, 56% are with wildebeest, while only 29% of the encounters for small packs are with wildebeest (Fisher’s LSD, P ⬍ 0.05). For large packs, only 17% of prey encounters are with impala, compared with 29% for small packs (Fisher’s LSD, P ⬍ 0.05). Viewed in this way, packs of different sizes have patterns of habitat use that bias prey encounters in favor of the species that they kill most often (Figure 5.3).
5.2 Encounters and Hunts In general, predators do not hunt the majority of the herds that they encounter. In Selous, wild dogs hunted only 37% of 2,015 herds they encountered. Small and large packs showed similar selectivity, hunting 36% and 39% of the herds they encountered, respectively (normal approximation test for binomial rates, z ⳱ 1.30, P ⳱ 0.19). Because the proportion of herds that are hunted upon encounter is low, there is considerable scope for prey selection. Small Packs In their decisions whether or not to hunt the prey they encounter, small packs show a clear preference for impala and against wildebeest. For small packs, impala account for 43% of hunts, significantly greater than the percentage of prey encounters contributed by impala (29%; Fisher’s LSD, P ⬍ 0.05). By contrast, wildebeest were hunted at a frequency (21%) less
110 ▪ C H A P T E R 5
Figure 5.3 The proportion of prey encounters, hunts, and kills contributed by each prey species, compared with the relative abundance of each species. (a) Small packs (⬍9 adults). (b) Large packs (ⱖ 9 adults). Other prey combines hartebeest, duikers, hares, eland, waterbuck, and reedbuck.
than proportional to their encounter rate (29%; Fisher’s LSD, P ⬍ 0.05). Although small packs encountered wildebeest and impala equally often, they hunted impala twice as often as wildebeest (Table 5.2, Figure 5.3). Small packs hunted warthogs, zebra, and uncommon prey (Table 5.2: other prey) in proportion to encounter rates.
P R E Y S E L E C T I O N ▪ 111
Large Packs Patterns of prey selection through hunting decisions were less clear for large packs than for small packs (Table 5.2). For wildebeest (their most important prey), hunts were proportional to encounters. Impala were hunted more often than expected (26% vs. 17% of encounters; Fisher’s LSD, P ⬍ 0.05). Zebra (3% of hunts) were hunted less often than expected (7% of encounters; Fisher’s LSD, P ⬍ 0.05). Other species were hunted in proportion to their rates of encounter. Overview Both large and small packs hunt impala more frequently than expected, based on encounter rates, perhaps because the likelihood of making a kill is high for impala (64%) relative to other prey (44% overall; Chapter 4). For large packs, the high probability of hunting impala upon encounter offsets a low probability of encounter (relative to abundance). Large and small packs were disinclined to hunt zebra, though the pattern was significant only for large packs. This negative preference can also be explained by the likelihood of killing, which is lower for zebra than for any other prey species (7%; Creel & Creel 1995b). Small packs were disinclined to hunt wildebeest, which large packs hunted in proportion to their encounter rate. Here again, hunting decisions reflect the vulnerability of the prey, because small packs have significantly lower success than large packs when hunting wildebeest (Figure 5.4). For example, packs of 3 to 4 adult dogs make a kill in 21% of wildebeest hunts, compared with 56% for packs of 15 or more (see below and Chapter 6).
5.3 Hunts and Kills It is debatable whether the likelihood of making a kill once a hunt has begun should be considered as part of the process of prey selection. Perhaps the prey has been “selected” once a chase begins, and one can assume that a predator will exert a maximal effort to make a kill once it commits to a chase. On the other hand, the probability of killing is clearly one component of the overall probability that an item will enter the diet, and there may be a final element of prey selection after the chase begins. Informal observations suggest that wild dogs do vary the effort they put forth in a hunt. It is common to see wild dogs begin a chase, but break off after assessing the prey’s vulnerability (we defined a chase as any pursuit of prey that exceeded 50 m at a full run or ended with prey at bay). The decision not to push to maximal effort is a final stage of prey selection. Unfortunately, this type of prey selection is entangled with simple variation in the vulnerability of prey,
112 ▪ C H A P T E R 5
Figure 5.4 Hunting success (kills per hunt) as a function of pack size, for hunts of wildebeest. Error bars show 95% confidence limits. The low success of small packs in wildebeest hunts leads them to specialize in hunting impala.
and the following discussion does not disentangle variation in effort from variation in vulnerability. Small Packs For small packs, variation in hunting success across prey species reinforces patterns of prey selection at earlier stages. Impala form a larger proportion of kills than hunts (61% vs. 43%), and wildebeest form a smaller proportion of kills than hunts (12% vs. 21%; P ⬍ 0.05 for both species, Fisher’s LSD; Table 5.2 and Figure 5.2). Zebra also comprise a smaller proportion of kills than hunts. Other species are killed in the same proportions that they are hunted. Large Packs Variation in hunting success does not create any clear pattern of selection among prey species for large packs. The proportion of kills does not differ from the proportion of hunts for wildebeest, impala, warthogs, zebra, or other prey (Figure 5.3; Table 5.2).
5.4 Combined Effects of Encounter, Hunting, and Killing Probabilities on Prey Selection Patterns of encounter, decisions to hunt, and hunting success combine to determine the diet. Bias in favor of or against a prey species at one step can
P R E Y S E L E C T I O N ▪ 113
be reinforced or counterbalanced by selection at later steps. We have examined the overall pattern of prey selection by large and small packs by comparing the proportion of kills with the relative abundance for each prey. Small Packs Small packs killed impala more often than expected based on their abundance (61% of kills vs. 32% of prey community; Fisher’s LSD, P ⬍ 0.05), and wildebeest less often than expected (12% of kills vs. 51% of prey community). Small packs’ preference against wildebeest was reinforced at every stage from encounter to killing—they were encountered less often than expected, the proportion of hunts was less than the proportion of encounters, and the proportion of kills was less than the proportion of hunts. Small packs’ preference for impala was not driven by encounter rates, but was reinforced at both subsequent steps. Small packs encountered wildebeest and impala equally often, but at subsequent stages they showed almost symmetrical preferences in opposite directions for these two prey (Figure 5.3a). Small packs killed zebras significantly less often than expected based on relative abundance (Table 5.2), showed no bias for or against warthogs, and took less common prey (other prey in Figure 5.3) more often than expected. Their preference against zebras arose through small, reinforcing biases against zebras at every stage. By contrast, their preference for less common prey was entirely due to nonrandomly high encounter rates (Figure 5.3; Table 5.2). Large Packs Unlike small packs, large packs killed most species in proportion to their abundance. The only significant preference overall was against zebra. Impala were killed more often than expected based on encounters, but this was offset by a low encounter rate, so that the proportion of kills did not differ from that expected on the basis of abundance. Overview Patterns of prey selection differ dramatically for large and small packs. Given the broad effects of pack size on hunting success (Chapter 4), this result is not surprising. Small packs are specialist hunters of impala, but also take uncommon prey types (hartebeest, duikers, hares, eland, reedbuck, and waterbuck) significantly more often than expected, based on abundance. These preferences are related to small packs’ strong preference against wildebeest, which they do not hunt effectively (Figure 5.4). By contrast, large packs are good hunters of most prey types, including wildebeest (Figure 5.4). Because large packs are effective in hunting a wider range of prey, they
114 ▪ C H A P T E R 5
are less selective. Small packs are poor hunters of wildebeest, for which they compensate by focusing on impala (Figure 5.3). Small packs also compensate by hunting uncommon prey (this can be seen by comparing Figures 5.2a and 5.2b). For large packs, the cumulative hunt curve is quite flat beyond the first two prey types, but it retains an appreciable slope for small packs. Both large and small packs avoid zebra, probably because they are dangerous (Malcolm & van Lawick 1975). Zebra can kick forward and backward and, unlike other prey, they have a dangerous bite. Stallions are very vigorous in defending their herds against wild dogs, and wild dogs are often chased away if they investigate a zebra herd.
5.5 Quantitative Models of Prey Selection Maximization of Food Intake A widely used model of prey selection (Charnov & Orians 1973; Stephens & Krebs 1986) assumes that animals’ decisions to hunt a given prey type (or pass it by) serve to maximize the rate of energy intake. This model examines the decision to attack a prey animal upon encounter or ignore it to search for prey elsewhere. The model depends on three variables: λi ⳱ the rate of encounter between the predator and prey of type i, during periods in which the predator is searching ei ⳱ the energy obtained by hunting prey of type i hi ⳱ the handling time for prey of type i, generally defined as the time required to catch, kill, and consume the prey once a hunt begins For a predator making decisions among more than two prey types, the diet that maximizes the average energy intake rate is determined by ranking prey types by profitability (ei/hi), then adding types to the diet in order of rank as long as Equation 5.1 is satisfied (Stephens & Krebs 1986): j
兺 λiei
i⳱1
j
1Ⳮ
兺 λihi
⬍
ejⳭ1 hjⳭ1
(Eq. 5.1)
i⳱1
In words, this means that animals include a prey type in their diet as long as its profitability exceeds the expected payoff from ignoring the prey and searching for other opportunities. Essentially, the model just formalizes the idea that some prey are sufficiently unprofitable that they should not be attacked even when they are the only prey immediately available. For example, most predators would have a very difficult time killing a giraffe, and consequently do not attack them. In the model’s terms, hi is too large for giraffe to satisfy Equation 5.1. For other prey, these decisions are less ob-
P R E Y S E L E C T I O N ▪ 115
vious, and the model provides a quantitative means of testing whether or not energetic profitability is a good predictor of prey selection. Stephens & Krebs (1986) give a clear description of the model’s assumptions and limitations. Applying the model is straightforward once the variables have been estimated, but the model was developed with solitary foragers in mind. To retain the logic of the original model with data from group hunters, care is needed in the operational definitions of the variables. The original model also does not address variation in hunting techniques (sit-and-wait, stalking, coursing) that might affect the logic behind operational definitions of variables. Most of the species killed by wild dogs live in herds, so the decision to hunt is essentially a decision to hunt an entire herd. It is logically possible for wild dogs to select a particular herd member prior to the chase, but our observations suggest that this is not common. For example, when an impala chase begins, the herd usually flees in several directions, and individual dogs pursue different impala. As one or more dogs close in on their target, other dogs switch their focus to that impala. The same is true for hunts of wildebeest. For these species (and other prey), there is probably some selection of individuals prior to the chase, based on age, sex, or apparent condition. For example, only 9% of wildebeest kills are adults, although adults form 69% of the wildebeest population, and they are rarely selected when a chase focuses on a specific individual. However, within age-sex classes that are commonly killed, the individual that is pursued depends primarily on the geometry of the herd relative to the approaching dogs, and this is largely a matter of chance. If herds are the unit that is hunted, then λi should be measured as the rate of encountering herds, rather than individuals (Scheel 1993; Huggard 1993). For lions, Scheel (1993) measured encounter rates in herds/hour, with the assumption that lions were hunting 24 hours a day. Wild dogs are less opportunistic hunters than lions, so this would not be a reasonable definition for them. Herds encountered per hour of searching would be a better measure for wild dogs, but this still does not incorporate variation in effort while searching. A pack might move slowly and stop often, or they might move quickly and without pause. By measuring search effort in distance, rather than time, variation in effort is incorporated into λi, so we measured λi as herds/km/day. We recorded herd encounter rates for six packs on 243 days by following the dogs around the clock. As noted above, it would not be useful to define an encounter using a fixed distance between the dogs and their prey, because visibility varied among habitat types. We scored an encounter if the wild dogs reacted to a prey herd within our view (e.g., alert posture, watching, deviating to move toward herd) or if the prey reacted to the dogs (e.g., alarm calls, watching dogs, moving away). If a herd held more than one prey species, we scored an encounter with each species present. To estimate the energy obtained by hunting a specific prey type (ei), we
116 ▪ C H A P T E R 5
used published figures on ungulate body mass to estimate the mass of each animal killed by eye (Table 4.1; Sachs 1967; Dorst & Dandelot 1986; Estes 1991). We estimated edible mass on the basis of data in Blumenschine & Caro (1986) on the proportions of wildebeest and impala carcasses made up of flesh, viscera, bone, and skin. The proportions of these tissues eaten came from observations in Selous. The energy content (kJ/kg) of each tissue came from data on low-grade beef carcasses (USDA 1990). Combining these data, an impala provided wild dogs with 7304 kJ per kilogram killed, and a wildebeest yielded 6419 kJ per kilogram killed (see Table 4.2 for calculations). We used the average of these two values (6862 kJ/kg killed) for other prey species, multiplying the estimated body mass for each species by this conversion factor. We included zeros for failed hunts, to account for variation in hunting success, so that ei was measured as energy obtained per hunting attempt, as the model’s logic requires. Profitability is normally measured as ei/hi, where hi is handling time. Handling time is often defined as the amount of time required to catch, kill, and consume the prey. For wild dogs, this definition poses two problems. First, recall that the rate-maximization model normally deals with solitary hunters. For communal hunters (including wild dogs), prey size is correlated with hunting group size (Kruuk 1972; Gittleman 1989; Creel & Creel 1995), and this affects handling times in a way that is not consistent with the model’s logic. If large prey are killed by large groups, then handling times will tend to decrease as prey size increases, simply because more individuals eat simultaneously. From the individual’s point of view, the handling time is decreased because meat is going into other individuals’ stomachs—this does not retain the logic of the original model. By skewing handling times downwards, profitability is inflated for large prey (see Table 4.1 in Scheel [1993] for an example of this problem with lions). For communal hunters, the logic of the original model is better retained by restricting handling time to the effort of capture and killing. Second, for wild dogs it is better to define encounter rates as herds encountered per kilometer of search (rather than herds encountered per hour of search) because this accounts for variation in search effort (as discussed above). By the same logic, the effort of killing (hi) is best measured by the distance chased. By substituting chase distance for handling time, we incorporate variation in the speed of the chase. This definition also avoids the inflated estimates of profitability for large prey discussed above. However, this definition does not incorporate variation in the effort of killing an animal once it has been grabbed. For coursing predators like wild dogs, chasing is the main effort involved in killing, particularly for prey like impala, young wildebeest, warthogs, duikers, and hares, which are quickly dispatched. However, larger ungulates (e.g., adult or subadult wildebeest) can be difficult to kill once they are caught, especially for smaller packs. Ideally, an index of hunting effort should combine the energetic costs of catching and killing,
P R E Y S E L E C T I O N ▪ 117
and the risks involved, but we do not have one (see Chapter 4). Chase distance is the best approximation that our data provide. Using these definitions for ei, hi, and λi with Equation 5.1, the rate-maximization model predicts that Selous wild dogs should include five species in their diet (Table 5.3). Ranked in order of profitability, the predicted diet includes hartebeest, warthog, duikers, impala, and wildebeest. Hares and zebra should be excluded. Several species were hunted too rarely to estimate the model’s parameters accurately (e.g., eland, waterbuck, and reedbuck), but hunting success was low for these prey, so it is likely that the model would exclude them from the diet. In Table 5.3, the mean profitability (ei/hi) for each species is not equal to the mean of ei divided by the mean of hi. The quantity desired here is the mean of a ratio (mean of [ei/hi]), which is not equal to the ratio of means (mean of ei/mean of hi). Dividing the mean of ei by the mean of hi gives profitabilities that are systematically low, and can affect the profitability ranks of prey species. Unfortunately, the only other study that has applied quantitative prey selection models to a large carnivore (Scheel’s [1993] analysis for Serengeti lions) made this error in estimating profitabilities, and did not present the data in a manner that allows for recalculation. How well does the rate-maximizing diet match the actual diet of wild dogs in Selous? The five prey species in the predicted diet account for 88% of all hunts (95% confidence interval ⳱ 85.5–91.5% of hunts), 95% of all kills (95% CI ⳱ 94.0–97.5%), and 96% of the prey mass killed (95% CI ⳱ 94.5–97.6%). Although this is not a powerful test, the fact that the predicted and actual diets are similar suggests that hunting decisions by wild dogs take profitability and encounter rates into account (as in Serengeti lions: Scheel [1993]). A stronger test is to ask whether the order of preference from the model predicts the likelihood of attacking each species, when encountered (Figure 5.5). The correlation between the probability of attack and preference ranks from the model is low, and the slope tends in the wrong direction (Figure 5.6: rs ⳱ 0.07, P ⳱ 0.86). If African hares are considered outliers (because they are the smallest prey by a large margin, and are unusually likely to be attacked; Figure 5.5), the model does better at predicting attack probabilities, but the correlation is still far from significant (rs ⳱ ⳮ0.38, P ⳱ 0.40). The rate maximization model does a reasonable job of predicting the wild dogs’ diet mainly because it predicts that wildebeest and impala should be included. Any model that includes wildebeest and impala in the optimal diet will give a reasonable match to the observed diet of Selous wild dogs, because these two prey account for 71% of hunts (N ⳱ 788), 78% of kills (N ⳱ 369), and 84% of the mass killed. Given this, a wide range of models or hypotheses (e.g., “eat the two most frequently encountered prey types”) would predict the dogs’ diet reasonably well. Using a more stringent criterion (the correlation between predicted preference ranks and observed attack
118 ▪ C H A P T E R 5
Figure 5.5 Probabilities of hunting each prey species when encountered.
Figure 5.6 A test of the rate-maximization model of prey selection, using the Spearman rank correlation between the probability of attacking a species upon encounter (y) and predicted preference ranks from the model (x). The model predicts a negative slope for the points in this plot. AH ⳱ African hare, ZB ⳱ Zebra, DK ⳱ duiker, WB ⳱ wildebeest, IM ⳱ impala, WH ⳱ Warthog, KG ⳱ kongoni (hartebeest).
P R E Y S E L E C T I O N ▪ 119
probabilities), the model does not make good predictions about hunting decisions. There are at least three plausible reasons for the lack of fit between predicted and observed patterns of prey selection. First, the model is based entirely on the energetics of hunting. If wild dogs’ hunting decisions are sensitive to the risk of injury, dangerous prey will be taken less often than predicted by the model. The probabilities of attack for each species in Figure 5.5 would probably correlate well with a variable that combined energetic cost and risk. The highest probabilities of attack upon encounter are for African hares and impala, which are easy to subdue and present little risk of injury. The lowest probabilities of attack are for warthog and zebra, both of which are difficult and risky to subdue. Warthogs use their tusks effectively against wild dogs (Figure 4.3i), and are responsible for most of the serious injuries inflicted by prey, including broken legs, deep punctures, and large cuts. We saw no injuries inflicted by zebra, perhaps because the dogs very rarely persisted in hunting zebra once the stallion began defending the herd. Unlike most ungulates, zebras kick forward and backward and bite effectively. Although wild dogs in Serengeti killed a substantial number of zebras in one study, most of these kills were made by one pack during a single denning period (Malcolm & van Lawick 1975). Most packs did not hunt zebras (Malcolm & van Lawick 1975), and some packs actively avoided zebra herds to avoid harassment by stallions (Kuhme 1965). Wild dogs were occasionally harassed by zebras in Selous, including instances in which zebras approached resting wild dogs and chased them away, even though there was no obvious resource (e.g., a water hole) nearby. Models of prey selection by large carnivores would probably perform better if data on energetic costs and benefits and risk of injury could be combined. This would be very difficult, because energy and risk would have to be related directly to fitness. Converting either measure into an effect on fitness would be difficult by itself. For risk, in particular, calculating an effect on fitness is a thorny empirical issue, because serious injuries are rare events with strong effects. If a wild dog has its leg broken by a warthog, its fitness is likely to fall to zero as a result. However, a huge volume of observational data would be needed to obtain representative data on the average loss of fitness due to injuries from each prey species. Second, the model is constructed so that inclusion of a prey species does not depend on its own encounter rate—the decision depends only on rates of encounter with other prey. Stephens & Krebs (1986, p. 23) call this “encounter-contingent policy making,” and note that “no opportunity can be lost by attacking an item of the highest possible rank, since the best alternative is immediately to encounter another top-ranking item.” This argument is logical, but it assumes that profitabilities are measured without error. We believe that profitability might be overestimated for uncommon prey. It is plausible that rarely killed prey are hunted only when the conditions make a kill likely
120 ▪ C H A P T E R 5 Table 5.3 Parameters of profitability for wild dogs hunting common prey in Selous Species
(herds/km)
e (kJ)a
h (km)b
e/h (kJ/km)c
Include?
Hartebeest Warthog Duiker Impala Wildebeest African Hare Zebra
0.01 0.09 0.02 0.11 0.21 0.01 0.03
19588 3987 7253 12063 12719 144 5284
0.46 0.20 0.47 0.94 0.51 0.11 0.44
60436 46263 28835 22876 22717 2773 2616
yes yes yes yes yes no no
a
Calculations of average energy obtained include zeros for unsuccessful hunts, so that hunting success is incorporated. Average chase distances include unsuccessful hunts. c Mean profitability is calculated as mean of (eij/hij) where i subscripts prey species and j subscripts kills. Mean profitability is not equal to (mean of ei)/(mean of hi). b
(e.g., visible injury or illness, prey encountered near a barrier such as a riverine thicket, or an undetected close approach). If so, the data from a few opportunistic hunts will overestimate the profitability that could be obtained in the long run, hunting under a wider range of conditions. We suspect that this effect underlies the very high profitability of hartebeest in our data (Table 5.3). Third, successful hunting behavior depends on learning, and it is likely that prey selection is among the learned components. For example, the tendency of wild dogs to hunt zebra is clearly transmitted by learning (Malcolm & van Lawick 1975). More generally, a prey species that is important in one ecosystem is sometimes ignored in another ecosystem. Greater kudu are the second most important prey of wild dogs in Kruger (Pienaar 1969; Mills & Biggs 1993), but are very rarely hunted in Selous (0.2% of hunts). The opposite is true for wildebeest, which are critical prey in Selous (Figure 5.2), but are not killed in Kruger (Pienaar 1969; Mills & Biggs 1993). Clearly, predators sometimes ignore potential prey for reasons that are not well understood, and these unexplained patterns might relate to learned prey preferences. Perhaps wild dogs in Selous ignore greater kudu simply because they were not exposed to kudu hunts as puppies—learning would perpetuate local anomalies in prey choice, reinforcing low preferences for uncommon species, even profitable ones. Minimization of Starvation Risk Instead of maximizing the rate of food intake, a predator might try to minimize the risk of failing to meet its food requirements (Stephens 1981). In this case, the optimal diet depends on the variance of profitability as well as the mean. Using the “Z-scores model,” the risk-minimizing prey type can be
P R E Y S E L E C T I O N ▪ 121
Figure 5.7 Results of a risk-sensitive model of prey selection. The two Y-intercepts labeled R are upper and lower estimates of daily energy requirements for wild dogs. (a) Overall, the model suggests that hartebeest should be the most preferred prey, followed by warthogs and duikers. (b) The model predicts similar preferences for wildebeest and impala. AH ⳱ African hare, ZB ⳱ Zebra, DK ⳱ duiker, WB ⳱ wildebeest, IM ⳱ impala, WH ⳱ Warthog, KG ⳱ kongoni (hartebeest).
determined from a plot of mean-variance pairs for the profitability of each prey (Figure 5.7). In a Z-scores plot, the steepest line connecting a point denoting energy requirements on the mean profitability axis to any point in the set of prey choices identifies the prey type that minimizes the risk of energy shortfall (Stephens & Charnov 1982; Stephens & Krebs 1986). To apply this model, we used the data described above for the rate-maximization model. We estimated that a wild dog requires between 2.0 and 2.5 kg of meat per day, based on long-term intake rates for our population (Creel & Creel 1995b). We converted this requirement to units of kJ/km chased using the mean chase distance for all species (0.8 km, including failed hunts) and the conversion for kg of meat to kilojoules of energy described above (6862 kJ/kg killed) from Table 4.2. As shown in Figure 5.7, the risk-minimization model predicts that hartebeest will be most preferred, followed by warthogs and duikers. These species are not common prey (though warthogs rank third), so the model does a poor job of predicting the diet. The model performs poorly, at least in part, because it does not take encounter rates into account. However, the model also does a poor job of predicting encountercontingent preferences (Figure 5.8). The correlation between attack probabilities and preference ranks from the model tends toward a positive slope
122 ▪ C H A P T E R 5
Figure 5.8 A test of the starvation risk-minimization model of prey selection, using the Spearman rank correlation between the likelihood of hunting a species upon encounter (y) and predicted preference ranks from the model (x). The model predicts a negative slope for the points in this plot. AH ⳱ African hare, ZB ⳱ Zebra, DK ⳱ duiker, WB ⳱ wildebeest, IM ⳱ impala, WH ⳱ Warthog, KG ⳱ kongoni (hartebeest).
(Figure 5.8: rs ⳱ 0.25, P ⳱ 0.54), while a negative slope would support the model. If one discounts the predictions for hartebeest and duikers (because they are uncommon in many habitats) and warthogs (because they are dangerous), the model does make the interesting prediction that preferences between wildebeest and impala should be weak (Figure 5.7b). This matches the observation that small packs prefer impala, while large packs prefer wildebeest (Chapter 4; and “Combined Effects of Encounter, Hunting, and Killing Probabilities on Prey Selection” above).
5.6 Summary Small packs of wild dogs are more selective than large packs in their hunting decisions, perhaps because small packs are not effective hunters of some prey types that are vulnerable to large packs. Small packs specialize in hunting impala, but also hunt rare prey more often than expected. Large packs take prey more closely in proportion to their abundance. Both large and small packs took zebras less often than expected, perhaps because they are dangerous. Overall patterns of prey selection are the product of three sequen-
P R E Y S E L E C T I O N ▪ 123
tial steps, with nonrandom patterns in rates of encounter with prey, in decisions to hunt when prey are encountered, and in the likelihood of killing if a hunt is initiated. A preference at one stage is usually reinforced by preferences at other stages (as expected), but preferences are offsetting in some cases. A quantitative model based on the assumption that wild dogs maximize their rate of food intake does a good job of predicting the actual diet. A model based on the assumption that wild dogs minimize the risk of energy shortfall does a poor job of predicting the actual diet. Neither model accurately predicts the probability of attacking prey types once they are encountered. Possible reasons for the poor performance of these models include the following: failure to incorporate the risk of injury as well as energetic costs; problems with measuring profitability for uncommon prey; and the effects of learned prey preferences.
6
Ungulate Herd Sizes and the Risk of Predation by Wild Dogs
Many studies have asked whether the risk of predation is affected by group size (Lima & Dill 1990; Krause & Godin 1995; Lima & Zollner 1996; Treves & Chapman 1996), and, in general, the risk of predation decreases as prey group size increases (Neill & Cullen 1974; Kenward 1978; Pulliam & Caraco 1984). Many observational and experimental studies of birds (Lima 1995a,b) and fish (Pitcher & Parrish 1993) have confirmed this pattern, but less is known about group size and predation risk for mammalian predators and prey (Lazarus 1995). There are limitations to the empirical data on group size and predation risk, even in such well-studied taxa as songbirds, because most studies of predation risk have good data on the prey, but little information about predators. Conceptually, an individual’s risk of predation is determined by the effect of herd size on four probabilities, which are: 1. Encounter rates—the probability of being encountered by a predator. 2. Attack preferences—the probability that the predator will hunt, upon encountering a group of prey. 3. Hunting success—the probability that the predator will make a kill, upon hunting. 4. Dilution of risk—the probability that a given individual will be the victim, upon a kill being made. The first two probabilities are conceptually straightforward, though the great majority of data sets combine them into a single measure (Inman & Krebs 1987; Uetz & Hieber 1994). In this chapter, we use the term probability of attack to mean the probability of being attacked once detected, and do not lump this measure with the probability of being detected. The third probability incorporates many details of the behavioral interaction between predator and prey, including collective vigilance, defense, evasion, and confusion (Milinski 1977; Lima 1995a,b). The fourth probability accounts for the fact that, as group size increases, there is a decline in the odds that any specific individual will be the victim if a kill is made (Hamilton 1971; Turner & Pitcher 1986). Many studies have tested whether group size affects one or more of these components of predation risk, but few studies have considered all four components simultaneously. In particular, surprisingly few studies have tested whether predators encounter prey groups of different sizes in proportion to their abundance.
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 125
Scheel (1993) found that Serengeti lions encountered some species of prey more often than expected based on their abundance (also see Chapter 5). A similar effect could operate within a single prey species, so that large or small herds are encountered by predators more often than expected by chance. Limited data suggest that predators are more likely to detect large groups of prey. Other studies have shown that the per-capita risk of predation (measured on a per-encounter basis) decreases as group size increases, but very few studies have examined how risk per encounter trades off with the likelihood of an encounter occurring (Calvert et al. 1979; Foster & Treherne 1981; Cresswell 1994; Uetz & Heber 1994). In general, it has been assumed that antipredator benefits, as measured on a per-encounter basis, are not overturned by an increased likelihood of encounter or attack (Krause & Godin 1995). Attack rates are determined jointly by encounter rates and the probability that a predator will choose to hunt upon encountering prey. Prey group size can affect the decision to hunt (Fitzgibbon 1990; Uetz & Hieber 1994), but as Krause & Godin (1995) note, “very few studies have investigated the choice of predators for prey groups of different sizes” (p. 465). For cheetahs and lions hunting ungulates, attack probabilities decrease as herd size increases (van Orsdal 1984; Fitzgibbon & Lazarus 1995), but cichlids hunting guppies in tank experiments preferred to attack larger schools (Krause & Godin 1995). Several field studies have shown that large colonies of birds and mammals receive more attacks (Hoogland 1995; Brown & Brown 1996), but it is not clear whether this is due to increased conspicuousness (and thus higher encounter rates) or an increased probability of attack once detected. Hunting success measures the probability that an attack will end in a kill. Many studies of predation risk and group size effects do not examine hunting success, because they record some measure of predator detection (e.g., flight distance, sometimes in response to an artificial predator), or measure the rate of attack by a predator that is not allowed to kill the prey (Lima 1995a,b; Krause & Godin 1995). Direct data on hunting success are recorded in two broad types of field studies—those focusing on the predator, and those focusing on the prey. Few field workers who focus observations on prey see enough kills to test whether group size affects the likelihood that a hunt will succeed. Studies focusing on a predator are more likely to produce direct data on hunting success (Schaller 1972; Mills 1990; Caro 1995), and several have collected enough data to test for a relationship between prey group size and the success of hunts. In general, these studies suggest that hunting success decreases as herd size increases (Kenward 1978; Fitzgibbon 1990; Mills 1990), though there are exceptions (Mills 1985). Based on broad patterns for the four probabilities that determine predation risk, individuals in large groups are generally less vulnerable to predation. Dilution of risk is always important, relative to other effects. Mechanisms such as collective detection or confusion of attackers are more variable, but
126 ▪ C H A P T E R 6
often favor large prey groups. It is noteworthy that essentially all prior data on these issues come from studies in which the predator relies on stealth or surprise. Combinations that have been studied include fish preying on fish (Pitcher & Parrish 1993), birds preying on birds (Kenward 1978; Lima 1995a,b; Brown & Brown 1996) and mammals preyed on by a stalking mammal (Fitzgibbon 1990) or by a bird (Hoogland 1995). To our knowledge, group size and the risk of predation have not been studied for the prey of a cursorial predator, where hunting success does not depend on stealth. Because wild dogs hunt at high rates, in daylight, concentrating most of their attention on a few prey species, they provide a good opportunity to examine the effect of herd size on predation risk. In this chapter, we measure the per-capita risk of predation as a function of herd size for wild dogs’ two most common prey, wildebeest and impala (Chapters 4 and 5). For each prey species, we produce a single measure of predation risk that incorporates the ways in which herd size affects encounter rates, the likelihood of attack, the success of attacks, and the dilution of individual risk. We described basic methods for collecting data on hunts in Chapters 2, 4, and 5. Briefly, we followed wild dogs to obtain rates of encounter with prey, hunts, and kills. In some encounters between dogs and prey, we could not count the prey herd, and, for most prey species, kills were too rare to test for effects of herd size on predation risk (see Figure 5.1). Consequently, out of 2015 prey encounters, the analyses in this chapter are restricted to 623 encounters with wildebeest and 280 encounters with impala, during which we counted the prey herd accurately. For wildebeest, the data were sufficient to simultaneously partition the analyses by herd size and by dog pack size. For impala, we did not have enough data to partition by dog pack size. To determine whether herd size affects encounter rates, we need a distribution of herd sizes encountered by dogs and a second, independent distribution of herd sizes for the prey population at large. The former distribution came from the 623 encounters with wildebeest and the 280 encounters with impala that we just described (also see Chapter 5). The latter distribution came from ground transects in which we recorded the size and composition of all wildebeest and impala herds that we encountered (N ⳱ 661). We stratified these counts by habitat type and season.
6.1 Probability of Being Encountered Wildebeest Wildebeest form small herds in Selous relative to other ecosystems, though temporary herds of more than 50 are common in the dry season, when ungulates congregate during daily movements to water. One occasionally sees herds of 500 or more, but, as Figure 6.1a shows, herds of 30 or less are the
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 127
Figure 6.1 (a) The frequency distribution of wildebeest across herd sizes. Open bars show the distribution of herd sizes for the population at large, and shaded bars show the distribution for herds encountered by wild dogs. (b) Log-odds tests to determine whether wild dogs encountered a herd-size class more or less often than expected. A log-odds ratio of 0 indicates random encounter. A positive log-odds ratio indicates that herds of that size class were encountered nonrandomly often. *indicates P ⬍ 0.05; **indicates P ⬍ 0.01.
128 ▪ C H A P T E R 6
Figure 6.2 (a) The frequency distribution of impala across herd sizes. Open bars show the distribution of herd sizes for the population at large, and shaded bars show the distribution for herds encountered by wild dogs. (b) Log-odds tests to determine whether wild dogs encountered a herd size class more or less often than expected. Details as in Figure 6.1.
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 129
rule, and herds of 10 or less are most common (singletons are included in this class). Small herds also predominate in the distribution of herd sizes encountered by wild dogs (Figure 6.1a: correlation between herds available and encountered: Kendall’s τ ⳱ 0.60, N ⳱ 6, P ⳱ 0.091), but wild dogs do not encounter herds of varying size in proportion to their abundance (χ2 ⳱ 714.5, df ⳱ 5, P ⬍ 0.001). Figure 6.1b shows the results of log-oddsratio tests in which we calculated log (proportion of herds encountered ⳰ proportion of herds in population) for each herd size class. Small herds were encountered less often than expected, herds of intermediate size were encountered at expected rates, and large herds were encountered more often than expected. The expected distribution of herd sizes is based on our eyesight. Because we drove these transects specifically to determine the distribution of herd sizes, we are fairly confident that small herds were no more likely to be missed than large herds. If this assumption is correct, then the expected distribution of herd sizes is unbiased. If the assumption was false, we might have overestimated the true frequency of large herds. Overestimating the frequency of large herds would lead to a false conclusion that large herds are encountered less often than expected—just the opposite of the pattern we found. The fact that wild dogs detected large herds more often than expected suggests that they use smell, hearing, or rules about habitat selection and movement to detect large herds. We have no data to isolate which of these mechanisms the dogs actually use to encounter large herds, but all of them seem plausible. In addition, small herds might detect approaching wild dogs before the dogs have detected them, and take action to avoid detection. From the behavior of wildebeest and impala in our observations, we do not think this is likely, but it remains a possibility. Considering the evolution of wildebeest herd size, nonrandom encounter rates impose a cost on wildebeest in large herds. Wildebeest that remain alone or in a herd of less than ten are less likely than wildebeest in large herds to encounter wild dogs. Other aspects of predation risk might offset this cost, but it is a cost of large herd size nonetheless. Impala Impala typically form smaller herds than wildebeest, and the majority of impala sightings are herds of 1 to 5 (Figure 6.2a). There is a positive relationship between the availability of herd size classes and the rate at which wild dogs detect them (Figure 6.2b: correlation by Kendall’s τ ⳱ 0.80, N ⳱ 5, P ⳱ 0.050), but herds are not encountered randomly with respect to size (χ2 ⳱ 13.82, df ⳱ 4, P ⬍ 0.008). As with wildebeest, log-odds ratios show that wild dogs encounter large herds (⬎ 50 impala) more often than expected, while they encounter small herds (6–10) less often than expected.
130 ▪ C H A P T E R 6
Encounters with very small (1–5) and intermediate-sized (11–50) herds occur at expected rates. Wildebeest and impala differ greatly in their foraging ecology. Wildebeest are grazers, while impala combine browsing and grazing. Because foraging ecology affects an ungulate’s distribution, one might expect differences between the two species in patterns of encounter with predators. In our data, however, an increased likelihood of encountering wild dogs is a cost of living in large herds for both wildebeest and impala.
6.2 The Probability of Being Hunted upon Encounter Wildebeest Logistic regression shows that larger wildebeest herds are more likely to be attacked upon detection (regression fit by maximum likelihood: β ⳱ 0.35, ⳮ2*log-likelihood ⳱ 820.0, χ2 ⳱ 5.71, df ⳱ 1, P ⳱ 0.017, N ⳱ 623). Looking at more detailed patterns, wildebeest herds of intermediate size are most likely to be attacked (Figure 6.3: second order polynomial regression of hunts/encounter on herd size, F2,22 ⳱ 12.75, P ⳱ 0.0002, R2 ⳱ 0.54). The preference for attacking herds of intermediate size is seen for all wild dog pack sizes, but it is stronger for large packs than for small ones (Figure 6.3: Compare the shape of a cross-section at low pack size to a cross-section at high pack size). Impala For impala, probabilities of attack varied with herd size in a pattern similar to that of wildebeest (Figure 6.4), but the relationship was not significant (logistic regression β ⳱ 0.011, ⳮ2*log-likelihood ⳱ 379.0, χ2 ⳱ 2.26, df ⳱ 1, P ⳱ 0.13, N ⳱ 280). The probability of being hunted is lowest for herds of 1–5 impala, where only 1 of 3 herds is attacked upon encounter. The odds of attack increase as herd size grows to 11–20 (for which 1 of 2 herds is attacked upon encounter), then change little with further increases in herd size.
6.3 Hunting Success Wildebeest Once a wildebeest herd is detected and a hunt is initiated, there is a striking increase in the probability that a kill will be made as herd size increases (Figure 6.5). Using logistic regression fit by maximum likelihood, this increase is significant (β ⳱ 0.08, ⳮ2*log-likelihood ⳱ 300.9, χ2 ⳱ 4.27,
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 131
Figure 6.3 The probability of attack for wildebeest herds of varying size, upon encounter with wild dog packs of varying size.
df ⳱ 1, P ⳱ 0.039, N ⳱ 235). A polynomial regression of herd size on hunting success does not fit significantly better than a simple linear increase, in part because the shape of the relationship depends on the size of the wild dog pack that is hunting (Figure 6.5). Hunting success increases in a concave curve for small packs and in a convex curve for large packs.
Impala As with wildebeest, hunts of impala herds are more likely to end in a kill as herd size increases (logistic regression: β ⳱ 0.025, ⳮ2*log-likelihood ⳱ 148.2, χ2 ⳱ 5.49, df ⳱ 1, P ⳱ 0.019, N ⳱ 118). A polynomial regression does not fit significantly better than a straight line (Figure 6.6). These are notable results. Once a hunt begins, a kill is significantly more likely in large herds of wildebeest or impala than in a small herds. By itself, this suggests that many widely discussed antipredator benefits of grouping
132 ▪ C H A P T E R 6
Figure 6.4 The probability of attack for impala herds of varying size, upon encounter with wild dogs. Bars show 95% CI for each point.
(early detection of predators, collective defense, and the confusion effect) do not operate effectively in interactions between wild dogs and their primary prey. If early detection, collective defense, or the confusion effect operated effectively, then hunting success would be lower for large herds, rather than higher. Second, most empirical studies have found that large groups are less vulnerable when attacked. As we mentioned earlier, prior studies have worked with predators that hunt by stealth or surprise; we are not aware of other studies with cursorial predators. Our results suggest that the antipredator effects of grouping run in opposite directions for stealthy and cursorial predators. If so, then prey species that face predators of both types will be subject to opposing selection that yields a distribution of herd sizes that trades off the risk of predation by coursers and stalkers. An interesting implication is that the herd size distribution is not likely to minimize the risk of predation by any single predator. Antipredator effects are variable, depending on the predator involved, a point that should probably be more prominent in models of grouping.
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 133
Figure 6.5 The probability of a hunt ending in a kill, for wildebeest herds of varying size, if attacked by wild dog packs of varying size.
6.4 Kills per Encounter, Dilution of Risk, and Combined Measures of Vulnerability Wildebeest In this section we start moving to a more complete measure of risk by combining two measures: the probability of a kill occurring once a hunt begins, and the probability that a hunt will occur once a herd is encountered (formally, Pr兵killⱍencounter其 ⳱ Pr兵huntⱍencounter其 * Pr兵killⱍhunt其). Once a herd is encountered, the probability that at least one wildebeest will be killed increases as herd size increases (Figure 6.7; logistic regression: β ⳱ 0.27, ⳮ2*log-likelihood ⳱ 300.9, χ2 ⳱ 4.26, df ⳱ 1, P ⳱ 0.039, N ⳱ 623). This analysis incorporates any mechanisms that deter an attack or alter its outcome. In relating kills per encounter to herd size, a second order polynomial regression (F2,22 ⳱ 4.21, P ⳱ 0.028, R2 ⳱ 0.28) fits significantly
134 ▪ C H A P T E R 6
Figure 6.6 The probability of a hunt ending in a kill, for impala herds of varying size. Bars show 95% CI for each point.
better than a linear regression, because the probability of a kill occurring initially increases, then levels out or declines slightly above a herd size of roughly 40 wildebeest. The herd size at which risk stops increasing depends on the size of the hunting pack (Figure 6.7). Earlier, we noted that the effect of herd size on predation risk is probably dependent on the species of hunter. This result extends that point by showing that herd size effects can also depend on the size of the hunting group. A herd size that yields low risk of predation by a small pack of wild dogs may yield a high risk of predation by a large pack. Once wild dogs have been encountered, the risk that at least one wildebeest will die increases as herd size increases. Nonetheless, the probability of dying decreases for each herd member through the dilution of risk (Figure 6.8). Risk dilution is strongest in small herds (logically, it must be), and dilution yields little reduction in individual risk above a herd size of 20 wildebeest (Figure 6.8). Consequently, a reciprocal regression of diluted risk on herd size fits better than a linear regression (reciprocal regression, β ⳱ 0.022 Ⳳ 0.005, t23 ⳱ 4.75, P ⳱ 0.00009, R2 ⳱ 0.5, vs. R2 ⳱ 0.14 for linear regression). From the last paragraph, one might conclude that wildebeest in larger
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 135
Figure 6.7 The probability of an encounter ending in a kill, for encounters between wildebeest and wild dogs. This measure combines the relationships shown in Figures 6.3 and 6.5.
herds run a smaller risk of predation, but this does not allow for the fact that larger herds are encountered by wild dogs more often than expected by chance (Figure 6.1). To account for this, we measured diluted, encountercorrected risk as a function of herd size (Figure 6.9). This measure, which we call overall risk, takes into account all the processes by which herd size might affect an individual’s vulnerability to predation: detection, the decision to hunt or not, the success of hunts, and dilution of risk. Overall risk increases as herd size increases (linear regression: t22⳱2.99, P ⳱ 0.007, R2 ⳱ 0.31). Wildebeest in large herds are less likely to die if encountered by wild dogs, but they are more likely to be encountered, and the encounter effect is strong enough to negate the dilution of individual risk in large herds. The relationship between overall risk and herd size is affected by the size of the attacking pack (Figure 6.9). For large dog packs, overall risk for wildebeest increases almost monotonically as herd size increases, though
136 ▪ C H A P T E R 6
Figure 6.8 The diluted risk of death for individual wildebeest in herds of different size, based on the probability of a kill occurring when a wildebeest herd is encountered. To show the surface clearly, the axis labeled wildebeest herd size runs in the opposite direction, when compared to Figures 6.3, 6.5, 6.7, and 6.9.
risk begins to asymptote at a herd size of 40–60. For small packs, overall risk initially drops as herd size increases, then begins to rise again around a herd size of 40 wildebeest. From the perspective of the wildebeest, we can ask if herd-size preferences are related to the risk of predation. The distribution of individual wildebeest among herd sizes (Figure 6.10, not the same as the frequency distribution of herd sizes, Figure 6.11) shows that most wildebeest are in herds of intermediate size (21 to 30), with fewer individuals in small or large herds. The distribution has the shape expected if selection to minimize the risk of predation by small wild dog packs is important, but it does not relate to the risk of predation by larger packs. However, wildebeest face predation by spotted hyenas, lions, and leopards, which are all far more common than wild dogs (Stander 1991; Creel & Creel 1996), and the selection pressure applied by each carnivore should be proportional to its abundance. More
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 137
Figure 6.9 The diluted risk of death for individual wildebeest in herds of different size, corrected for nonrandom rates of encounter between wild dogs and herds of different size (see Figures 6.1 and 6.8).
exactly, the strength of selection would be proportional to the number of individuals killed by each species, so leopards probably have little effect on wildebeest (Bothma & LeRiche 1989; Bailey 1993; Stander et al. 1997), but the effects of lions and hyenas should be dominant. In general, comparing the distribution of individuals among herds to the risk of predation is not a strong test, unless data on the risk of predation is available for all the important predators. Our data show that wildebeest in large herds are at high risk of falling prey to wild dogs, which are coursers. Most data for stalking predators suggest that prey in large herds are at low risk (Fitzgibbon & Caro 1992; Scheel & Packer 1995). Together, these patterns suggest that intermediate herd sizes may be a good solution for trading off the risk of predation by stalkers and coursers. In Selous, spotted hyenas (also coursers) are the most common large carnivore (⬃ 0.3 adults/km2; Creel & Creel 1996), followed by lions (0.1 adults/km2; Creel & Creel 1996), and wild dogs are the least common
138 ▪ C H A P T E R 6
Figure 6.10 The distribution of individual wildebeest among herd-size categories.
Figure 6.11 The frequency distribution of wildebeest herd sizes, relative to the likelihood of wild dogs making a kill.
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 139
(⬃0.04 adults/km2). Leopards are also common but very rarely take wildebeest. Cheetahs are not present. Taking the higher food intake of lions into account (Kruuk 1972; Scheel 1993; Scheel & Packer 1995), the rate of predation on wildebeest by stalkers and coursers is probably similar, and the preference of wildebeest for herds of intermediate size would be a good strategy to minimize the overall probability of being killed. This suggestion is very speculative at this point, but it is an interesting hypothesis for future research. To pursue this line of inquiry, it would be interesting to see data on herd-size effects for the prey of other coursers. It also would be interesting to see an integrated analysis of herd size in relation to risk of predation by a full set of carnivores. Shifting from the perspective of the wildebeest to the perspective of the wild dogs, we can relate the herd-size distribution to the likelihood of making a kill. Although wild dogs probably do not exert strong selection on wildebeest herd size, wildebeest are the single most important prey species for wild dogs in Selous (Chapter 5). For each herd-size category in Figure 6.11, we calculated the “likelihood of killing,” by multiplying the probability of a kill per encounter (Figure 6.7) by the ratio of encounters to herds available (Figure 6.1). We then divided each category’s likelihood by the maximum likelihood to produce a relative measure. This procedure gives a measure of how likely wild dogs are to kill a wildebeest in a herd of given size, relative to the availability of herds of that size, and shows that wild dogs are disproportionately likely to make kills in large herds (Figure 6.11: correlation between the likelihood of killing and herd size, Kendall’s τ ⳱ ⳮ1, Z ⳱ 2.82, P ⳱ 0.004). Unfortunately for the wild dogs, the most vulnerable herds are also the least common, and 70% of wildebeest herds are in the smallest and least-killed herd-size class (Figure 6.11: correlation between the likelihood of killing and proportion of herds, Kendall’s τ ⳱ –1, Z ⳱ 2.82, P ⳱ 0.004). At the population level, the distribution of wildebeest herd sizes reduces the vulnerability of the wildebeest population to predation, compared to an even or normal distribution of herd sizes. Predation by wild dogs would be a stronger limiting factor for the wildebeest population if the distribution of herd sizes were less highly skewed in favor of small herds. Impala When impala encounter wild dogs, the probability of a kill occurring increases as herd size increases (Figure 6.12: logistic regression fit by ML: β ⳱ 0.02, ⳮ2*log-likelihood ⳱ 229.35, χ2 ⳱ 7.37, df ⳱ 1, P ⳱ 0.007, N ⳱ 280). However, the dilution of risk is strong enough that an individual impala’s chance of dying decreases monotonically as herd size increases (Figure 6.13: correlation between diluted risk per encounter and herd size: Kendall’s τ ⳱ –1, z ⳱ 2.45, P ⳱ 0.14). Both of these results are similar to those for wildebeest—once an encounter occurs, large impala herds are good
140 ▪ C H A P T E R 6
Figure 6.12 The probability of an encounter ending in a kill, for encounters between impala and wild dogs. This measure combines the relationships shown in Figures 6.4 and 6.6. Bars show 95% CI for each point.
Figure 6.13 The diluted risk of death for individual impala in herds of different size, based on the probability of a kill occurring when an impala herd is encountered.
for both the predator and the prey, because the probability of a kill occurring is maximal, yet the risk for individual herd members is minimal. However, recall that rates of encounter with impala herds were not random with respect to herd size. We incorporated nonrandom encounters to give a measure of overall risk (diluted, encounter-corrected risk), as de-
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 141
Figure 6.14 The diluted risk of death for individual impala in herds of different size, corrected for nonrandom rates of encounter between wild dogs and herds of different size (see Figures 6.2 and 6.12). Bars show 95% CI for each point.
scribed above for wildebeest. For impala, overall risk drops steeply as herd size increases from small to intermediate sizes (Figure 6.14) Once herd size increases beyond 20 impala, overall risk remains almost constant. The nonparametric correlation between overall risk and herd size is not significant at P ⳱ 0.05 (Kendall’s τ ⳱ ⳮ0.6, z ⳱ 1.47, P ⳱ 0.14), but this does not allow for nonlinearity (Figure 6.14), and a reciprocal regression suggests that increasing herd size yields a decline in overall risk (β ⳱ 0.08 Ⳳ 0.007 SE, t3 ⳱ 12.51, P ⳱ 0.001). We can ask whether impala tend to live in herd sizes that minimize the risk of predation by wild dogs. As with wildebeest, this analysis carries the caveat that selection on herd size will respond to the risk of predation by an entire suite of predators, not just wild dogs, and that factors unrelated to predation will also affect the optimal herd size. However, impala are less important prey for lions and spotted hyenas (Creel & Creel 1996), so predation by wild dogs constitutes a greater portion of the total selection on herd size. If impala distribute themselves among herds to minimize their individual risk of predation, we would expect a negative relationship between the proportion of impala found in herds of a given size and the overall risk for individuals in that herd size (Figure 6.14). To test this, we examined three monotonically declining functions by regression, and found a significant relationship for two. The regression coefficient was most strongly significant for y ⳱ 1/x2 (B ⳱ 000033 Ⳳ .000008 SE, t4 ⳱ 4.249756, P ⳱ .013, R2 ⳱ 0.21), though the variance explained was greatest for y ⳱ a Ⳮ 1/x2 (R2 ⳱ 0.46). To summarize, impala select herd sizes that are not random with
142 ▪ C H A P T E R 6
Figure 6.15 The frequency distribution of impala herd sizes, relative to the likelihood of wild dogs making a kill.
respect to the risk of predation by wild dogs. Wild dogs probably impose an appreciable selection pressure on impala herd sizes, but we cannot conclude that the risk of predation by wild dogs causes the pattern. A similar pattern would be expected if impala in large herds had a low risk of predation by stalkers (as in Thomson’s gazelle; Fitzgibbon 1988). For wildebeest, individuals in small herds faced a lower overall risk of predation, when the probabilities of encounter, hunting, killing, and risk dilution were combined (Figure 6.9). For wild dogs hunting wildebeest, the likelihood of making a kill was highest for large herds (Figure 6.11). Together, these two results produce a conclusion that is typical for predator-prey relationships—what is good for the predator (large herds) is bad for the prey. For impala, the pattern is more interesting. Large herds are better than small ones from the perspective of the impala (Figure 6.14) and from the perspective of the wild dogs, because the probability that wild dogs will detect a herd and make a kill is an increasing function of herd size (Figure 6.15: Kendall’s τ ⳱ 1, z ⳱ 2.45, P ⳱ 0.014). These two results produce a thought-provoking scenario. The relationship between overall risk and herd size will favor impala that live in larger herds, yet an increase in mean herd size will also favor the wild dogs, increasing the total impact of wild dog predation on the impala population. This pattern will only arise for predatory interactions in which active antipredator mechanisms like early detection of predators, collective defense, or the confusion effect (which would reduce the collective risk of the herd) are ineffective. If collective risk rises but individual risk drops as herd size in-
H E R D S I Z E S A N D T H E R I S K O F P R E D AT I O N ▪ 143
Figure 6.16 Large herds favor both the predator and the prey when corporate risk rises, but individual risk drops with increasing herd size.
creases, then large herds are better from the perspectives of both the prey and the predator (Figure 6.16). Why does corporate risk increase in larger impala and wildebeest herds? First, large herds are more likely to hold a vulnerable individual, even if herd size does not affect any mechanism of interaction between the predator and prey. Earlier detection by many eyes is one of the major antipredator benefits of grouping in many species (Lima 1995a,b; Lima & Zollner 1996), but probably does not play a major role in interactions between coursing predators and their prey. Coursers initiate a hunt by testing for vulnerable individuals, often by inducing the prey to flee. Because coursers do not depend on a stealthy approach followed by a rush, early detection of their approach probably has limited benefits. Collective defense is another potential benefit of herding, but our data suggest that it is not effective for interactions between wild dogs and their major prey. For impala, this is not surprising, because impala avoid predation almost exclusively by running, and we never observed active defense against wild dogs. Wildebeest actively defend themselves by forming “pinwheels” of adults that surround calves and face outward (Creel & Creel 1995b). Even a single wildebeest female will aggressively defend her calf if it is grabbed, by returning to its side and trying to butt or trample the attackers. This strategy may be effective against solitary predators such as cheetahs (Caro 1994), but it rarely dissuades predation by a wild dog pack. In fact, most cases in which Selous wild dogs killed adult wildebeest came
144 ▪ C H A P T E R 6
in attacks on a female defending a calf. While the data in this chapter do not demonstrate that collective defense by wildebeest against wild dogs is ineffective, they suggest that collective defense by large herds is no more effective than collective defense by small herds. Several studies suggest that larger prey groups may confuse an attacking predator when they scatter at the beginning of an attack, and this confusion may reduce the likelihood of a kill (Neill & Cullen 1974; Kenward 1978). We believe that confusion can act against the prey as well. Individuals in large herds often impede one another at the start of their flight, particularly in thick habitats, where the vegetation constrains escape options. Fast flight through woodland requires many decisions about how to avoid obstacles. When many individuals are making these decisions simultaneously, coordination becomes difficult. Confusion is more apparent for impala than for wildebeest, because impala bolt in many directions, while wildebeest are likely to flee as a herd, running parallel. Impala fawns sometimes became sufficiently confused that they were captured after coming to a halt even though all of the wild dogs were behind them. It is interesting that confusion appears to favor the prey in fish and birds (Neill & Cullen 1974; Kenward 1978), but may favor the predator among mammals. Predatory birds and fish pursue prey in three dimensions, while mammals operate in two dimensions. All else equal, the loss of one dimension would increase the odds of prey impeding one another, as often happens when wild dogs hunt in woodlands. These arguments relate to the evolution of cooperative hunting by wild dogs (Chapter 5). Most prior theoretical and empirical work suggests that increases in prey group size will decrease the odds of an attack occurring or succeeding (Fitzgibbon & Caro 1993). Most of the empirical work that has demonstrated antipredator benefits to grouping has involved solitary predators that hunt by stealth. Wild dogs differ in two ways, both probably important. As we’ve discussed, they do not rely much on stealth, which weakens the benefit to be had by collective vigilance and early detection of an attack. In addition, the benefit of collective defense is weakened because wild dogs almost invariably hunt in groups, and this often allows them to overwhelm active defense in circumstances that might dissuade a solitary predator. A mother might defend her calf against one predator, but is unlikely to defend it against two (Kruuk 1972). Extending this logic, the benefits of collective defense are likely to weaken as the size of the attacking group increases. Because we did not collect detailed data on the behavior of prey during interactions with wild dogs, further research on the behavioral mechanisms involved in detection, defense, and confusion could be quite productive. It seems likely that cursorial predators will yield interesting insights on these issues.
7
Demography—Survival and Reproduction
Population dynamics are determined by rates of birth, death, immigration, and emigration. Consequently, demography is the foundation for answering many questions in conservation and behavioral ecology. In this chapter, we present basic demographic data, discuss the evolution of wild dog life histories, and consider some implications for conservation. We use these data to examine reproductive cooperation and conflict in Chapter 9, and to model extinction risk in Chapter 13.
7.1 Survival Rates Raw Data and Methods of Calculation We determined age- and sex-specific annual rates of survival using 1,068 annual records from 365 dogs, following a three-step process. First, we determined rates of apparent survival by considering all known individuals that disappeared to be dead (Table 7.1). Second, we adjusted our estimates of apparent survival to allow for undetected emigration, by assuming that the number of undetected emigrants of a given age, sex, and rank (alpha/subordinate) was equal to the number of unknown immigrants of that age, sex, and rank (Table 7.1). Third, we made an adjustment based on the disappearance of socially dominant dogs, because no alphas were ever detected to have dispersed, except those displaced by same-sex immigrants and those involved in pack splits. Consequently, we considered any dominant dog that disappeared to be dead, except those involved in pack split or displacements. We calculated survival rates using data only from dogs of known age, and compared these results to calculations that included dogs whose age was estimated from toothwear, pelage, and overall appearance. The two sets of estimates agreed closely (correlations between age-specific survival rates using the two methods: r ⳱ 0.96 for females, r ⳱ 0.94 for males). This agreement is not surprising, given that the survivorship curves are quite linear for adults. We based all of the analyses in this chapter on the larger data set. Our survivorship curves are based on longitudinal records of dogs born in different years, so we have combined static and cohort life tables. On average, each individual contributed 2.9 annual data points to the analysis.
146 ▪ C H A P T E R 7 Table 7.1 Raw data for age, rank, and sex-specific annual survival of Selous wild dogs
Females Known Fate
Females Unknown Immigrants
Males Known Fate
Males Unknown immigrants
Age
␣
Sub
␣
Sub
␣
Sub
␣
Sub
0 1 2 3 4 5 6 7 8 9 10 11
0 0 2 7 11 10 7 6 2 2 2 0
97 70 58 37 23 17 6 3 0 0 0 0
0 0 0 2 1 0 0 0 0 0 0 0
0 3 1 2 1 1 0 0 0 0 0 0
0 0 2 9 9 10 7 4 2 0 0 0
101 67 63 43 35 14 12 11 9 6 3 0
0 0 0 1 1 1 0 0 0 0 0 0
0 3 1 2 1 0 0 0 0 0 0 0
Factors Affecting Annual Survival Basic patterns of annual survival are shown in Table 7.2. and Figure 7.1. We used Akaike’s Information Criterion (with sample-size correction, AICc) to determine which logistic regression models best explained patterns of survival (Akaike 1983; Burnham & Anderson 1998). We fit logistic regressions by maximum likelihood, with survival as a dichotomous dependent variable, and age, sex, rank, and year as independent variables. We compared the set of regression models shown in Tables 7.3 (all ages classes) and 7.4 (adults only), following Burnham and Anderson’s guidelines for the use of AICc values to assess support from the data for each competing model (1998, pp. 127–129). This approach quantifies the tradeoff between obtaining a good fit to the data while avoiding overparameterization. The data sets for all age classes and for adults gave similar results: Age and year strongly affected survival; sex and rank had weaker effects, but were included in the best model. Averaged across both sexes and all ages, annual survival was 0.75. For males, annual survival averaged 0.76, compared to 0.74 for females. Table 7.2 summarizes how survival rates varied with age for both sexes, with confidence limits. Males and females had the same general pattern of survival across age classes (Figures 7.2 and 7.3). Both sexes had relatively poor odds of survival their first year, survived well as yearlings, then faced a decline in annual survival as they aged. Though broadly similar, the details of this pattern differed between the sexes in two ways. First, male pups had lower
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 147 Table 7.2 Annual survivorship by age class for females (F) and males (M). Rates are corrected for disappearance (see text). Resighting probability is fixed at one for those still in study population
Number of Individuals
Annual Survival
Lower 95% Confidence Limit
Upper 95% Confidence Limit
Age Class
F
M
F
M
F
M
F
M
Pups Yearlings
97 73
101 67
0.75 0.84
0.66 1
0.66 0.74
0.56 0.94
0.83 0.92
0.75 1
145 41 15
169 45 35
0.77 0.54 0.40
0.75 0.76 0.57
0.68 0.40 0.19
0.68 0.62 0.39
0.84 0.67 0.67
0.82 0.86 0.73
Adults 2–4 yrs 5–6 yrs 7–11 yrs
survival rates than females (0.66 vs. 0.75), while male yearlings had higher survival rates than females (1.00 vs. 0.84). Consequently, males experienced a much greater increase in survival between the pup and yearling age classes. Second, adult females faced a steeper decline in survival than adult males did, as they aged. By the oldest age class (7–10 years), survival was
Figure 7.1 Survivorship (lx) curves for females and males. To correct for undetected emigration, disappearance was equated with death, then unknown immigrants were used to estimate numbers of disappeared animals that were alive but not counted. N0 ⳱ 101 males and 97 females.
148 ▪ C H A P T E R 7 Table 7.3 Comparison of logistic regression models of annual survival, using Akaike’s Information Criterion, with a sample size correction. Data from all age classes
Model
ⳮ2ln(L)
噛 Params
N
AICc
Conclusion
Year**, Age*, Sex, Rank Year**, Age*, Rank Year**, Age*, Sex Year** Year, Sex, Rank Age, Sex, Rank Age Intercept only (Null) Sex Rank
769.9421 772.0188 773.8038 779.8073 777.9578 792.8929 797.2654 803.8385 803.3105 803.4976
5 4 4 2 4 4 2 1 2 2
834 834 834 834 834 834 834 834 834 834
780.0146 780.0671 781.8521 783.8217 786.0061 800.9412 801.2798 805.8433 807.3249 807.512
best model no evidence to exclude no evidence to exclude weak evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude
For models that cannot be strongly excluded and outperform the null, the results of t-tests for each independent variable are given by **: P ⬍ 0.01; and *: P ⬍ 0.05.
1.7-fold higher for males than for females (0.67 vs. 0.4, N ⳱ 35 and 14, respectively). Unlike males, the proportion of females that reproduce increases steadily with age, so the steep decline in annual survival for adult females is associated with high reproductive effort (see “Reproduction” and Figure 7.9). Survival varied substantially from year to year (Figure 7.4a), from 0.74 to
Table 7.4 Comparison of logistic regression models of annual survival, using Akaike’s Information Criterion, with a sample size correction. Data from adults only
Model
ⳮ2ln(L)
噛 Params
N
AICc
Conclusion
Year**, Age*, Sex, Rank Year**, Age*, Sex Year** Year**, Age*, Rank Year**, Sex, Rank Age, Sex, Rank Intercept only (Null) Age* Sex Rank
418.924 421.6554 426.4417 422.761 423.1209 432.2674 438.8841 437.0381 437.121 438.1377
5 4 4 2 4 4 2 1 2 2
443 443 443 443 443 443 443 443 443 443
429.0613 429.7467 430.469 430.8523 431.2122 440.3587 440.8932 441.0654 441.1483 442.165
best model no evidence to exclude no evidence to exclude no evidence to exclude weak evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude strong evidence to exclude
For models that cannot be strongly excluded and outperform the null, the results of t-tests for each independent variable are given by **: P ⬍ 0.01; and *: P ⬍ 0.05.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 149
Figure 7.2 Age-specific annual survival rates (sx) for females, with shortest unbiased 95% confidence limits (dashed lines).
Figure 7.3 Age-specific annual survival rates (sx) for males, with shortest unbiased 95% confidence limits (dashed lines).
150 ▪ C H A P T E R 7
Figure 7.4 Comparison of annual survival across years, excluding the first and last years of study. (a) All age classes combined. (b) Adults only. Dotted lines show shortest unbiased 95% confidence limits.
0.96. Restricting the data to adults gave a similar pattern across years, with changes of greater magnitude, but the best years for adult survival were also the best years for other age classes (Figure 7.4b: Kendall’s τ ⳱ 0.8, P ⳱ 0.05). There is a tendency for survival to be density dependent, by several measures. The correlation between adult survival and adult density is ⳮ0.66 (P ⳱ 0.22). The correlation between adult survival and total density is also ⳮ0.66 (P ⳱ 0.22). The correlation between survival of all age classes and adult density is ⳮ0.73 (P ⳱ 0.16), and the correlation between survival of all age classes and total density is ⳮ0.54 (P ⳱ 0.34). A comparison of survival rates across populations (see below) also shows a negative
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 151 Table 7.5 Chi-squared tests for interaction between sex and rank in their effects on survival of adults Category
N
Sex Rank Rank by Sex Male Female
551 529 521 282 239
2 by maximum likelihood 2 2 2 2 2
⳱ ⳱ ⳱ ⳱ ⳱
2.41, 1.61, 4.57, 0.34, 2.31,
P P P P P
⳱ ⳱ ⳱ ⳱ ⳱
0.12 0.45 0.10 0.84 ␣ and subordinates equal 0.13 ␣ tend to survive better
correlation between density and adult survival rates (Spearman’s r ⳱ ⳮ0.54, P ⳱ 0.27). Although none of these five correlations is significant with 95% confidence, they are all strongly negative, and it appears that some of the variation in survival among years and populations is due to density. Density-dependent survival may be related to patterns of dispersal. Dispersers face a substantial risk of death, relative to nondispersers, and wild dogs typically emigrate from large packs, which are more common in years of high density (see Chapter 8). We treated social rank as a dichotomous variable (alpha/subordinate) for analysis of survival rates. AICc values supported the inclusion of rank, which was included in the two models with strongest support from the data (Table 7.3). When pups and yearlings were excluded from the analysis, rank was still included in the best model of adult survival, but generally received less support (Table 7.4: The second and third best-supported models do not include rank). These results make it a bit difficult to interpret the relationship between rank and survival. Social rank had a relatively weak effect on the survival of adults, mainly due to an interaction between the effects of rank and sex. Restricting the analysis to adults, the survival rates of alpha males and subordinates were almost identical, but alpha females survived slightly better than subordinate females (Tables 7.1 and 7.5). Rank had a stronger effect on survival when all ages were considered (Table 7.3), because pups and yearlings were never dominant, and survival was strongly age-dependent. Apparently, much of the effect of rank on survival is due to collinearity between rank and age, but rank still enters the best regression model when age is included as a regressor, due to the interaction between rank and sex (Table 7.5). Comparison of Survival Rates across Populations We summarized data on annual survival for five wild dog populations in Table 7.6. The entries in Table 7.6 are ordered so that large, stable, dense
0.36 (154) 0.71 (56) 0.43 (100) 0.45 (84) 0.31 (16) — 35
(198)A (140) (450) (314) (86) (50)
0.70 0.92 0.71 0.76 0.66 0.52 38
Botswana 1 1989–1994
Selous 1991–1996 0.60 (75)E 0.75 (79) — — — — 15–17
Kruger 2 1974–1978 0.30 (240) 0.64 (65) 0.69 (129) — — — 17
Kruger 3 1990–1993 0.24 (103) — 0.80 (175) — — — 15
Serengeti 4,5 1967–1977 0.40 (16)B — 0.73C,D — — — 7
Serengeti 6,7 1985–1991 0.73 (5)B — 0.55–0.64C,D — — — 36F
Masai Mara8 1985–1990
A
: Large, high density, stable populations that have persisted are to the left. Small, low density, unstable populations that ultimately disappeared are to the right. Table entries are survival estimate as a proportion, followed by sample size in parentheses. B Sample size is number of litters, rather than number of individuals. C Original data combine adults and yearlings into a single class. D Sample size not given in original. E Original data restricted to nine litters counted within 4 months of birth. F This estimate comes from a small area defined by the ranges of few packs, so it is high for reasons of methodology. Sources: 1: McNutt (1995); 2: Reich (1981); 3: van Heerden et al. (1995); 4: Malcolm (1979); 5: Frame et al. (1979); 6: Fanshawe (1987); 7: Burrows et al. (1994); 8: Fuller et al. (1992a).
†
Pup Yearling Adult 2–4 years 5–6 years 7–11 years Density (Ad/1,000 km2)
Age Class
Population and Time Span†
Table 7.6 A comparison of age-specific annual survival and population density in five wild dog populations
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 153
populations are to the left, and small, unstable, low-density populations are to the right. Three patterns emerge by comparing populations: 1. The survivorship curve for Selous lies substantially above the survivorship curves of other wild dog populations. Clearly, good survival is a driver of high wild dog density in Selous. To quantify this difference in a simple way, we calculated the percentage of newborns that reach adulthood (2 years) in each population. In Selous, 64% of newborns reached breeding age, compared to a range of 19% to 49% in other populations (Botswana: 26%; Kruger: 19% to 45%; Serengeti: 19% to 29%; Masai Mara: 40% to 47%). 2. The survival of pups varies more than the survival of yearlings and adults. Estimates of annual survival for pups vary among ecosystems by a factor of 2.9, while yearling survival varies 1.4-fold and adult survival varies 1.9-fold. The apparently low variation in yearling survival is at least partly due to a lack of published estimates of yearling survival (some studies pooled yearlings with adults). Pup survival also varies widely within ecosystems. For example, annual pup survival in Serengeti was only 15% (N ⳱ 73 pups) between 1975 and 1977 (Malcolm & Marten 1982), but was 40% (N ⳱ 16 litters) between 1985 and 1991 (Burrows et al. 1994). 3. Adult survival relates negatively to population performance, while pup survival relates positively. The correlation between adult survival and population density is ⳮ0.54 (Spearman’s r, P ⳱ 0.27) suggesting that a high rate of adult survival does into necessarily translate into favorable population dynamics. In contrast, the correlation between pup survival and population density was 0.57 (Spearman’s r, P ⳱ 0.18). Pup survival averaged 47% in Selous, Botswana, and Kruger, which are large, stable populations with relatively high density, and 38% in Serengeti and the Masai Mara, which are small, low-density, unstable populations. Populations in decline have poor pup survival (with the exception of Masai Mara), and stable populations generally have good pup survival (with the exception of Botswana). Frame et al. (1979) noted that no pups were raised in Serengeti between 1974 and 1976, in a period with 15 pack breeding cycles. In this case, poor survival of pups clearly drove a population crash, because the reproduction was normal (Malcolm & Marten 1982) and adult survival was 80% (Table 7.6; Frame et al. 1979), the highest adult survival rate on record (Table 7.6). Only pup survival was poor during the population’s decline. Collectively, these points suggest that juvenile survival is an important determinant of wild dog population dynamics and conservation. Juvenile survival is a strong determinant of population dynamics in other well-studied taxa, for example passerine birds (Legendre & Clobert 1995). In contrast, simulations using VORTEX suggested that juvenile survival had a substantially smaller impact than adult survival on wild dog population dynamics (Ginsberg & Woodroffe 1997). These contradictory conclusions warrant further investigation.
154 ▪ C H A P T E R 7
Sensitivity of Population Growth to Demographic Parameters: Demographic Perturbation Analyses with Matrix Models We measured the sensitivity of population growth to variation in age-specific vital rates, using demographic perturbation analysis with matrix models of population dynamics (Caswell 2001). We built a Leslie matrix from the agespecific survival and reproductive rates in Table 7.7, and used the program Unified Life Models (ULM, Legendre & Clobert 1995) to project population dynamics and measure the population’s growth rate, λ. One population projection assumed that population growth was exponential, as in traditional life-table analyses. Because there is some evidence that survival is density dependent (see above), we ran a second projection, in which survival was linearly density dependent, with proportionately equal reductions in survival for all age classes. Still using ULM, we measured the sensitivity and elasticity of λ to changes in each age-specific demographic parameter (Caswell et al. 1997; Horvitz et al. 1997). We measured sensitivities and elasticities after the population had been projected through a sufficient number of generations to reach its stable age distribution. For applications of this approach to conservation-oriented questions with other species, see Menges (1990), Doak et al. (1994), and Schemske et al. (1994). Horvitz et al. (1997) note that the question “Which of the vital rates is most important to population growth?” can have different answers, depending on the measure of importance that one picks. Here, we are interested in identifying the vital rates that, when changed, have the greatest impact on population growth. Sensitivity and elasticity are appropriate measures of importance for this question. The sensitivity of λ gives the effect of a small additive change in one of the vital rates on population growth. The elasticity of λ gives the effect of a small proportional change in one of the vital rates on population growth (Caswell 1997). We report both sensitivities and elasticities (Tables 7.8 and 7.9), but we are most concerned with elasticities. The sensitivity and elasticity of λ to changes in age-specific survival and reproductive rates is shown in Table 7.8 for exponential population growth, and in Table 7.9 for density-dependent population growth. Exponential and density dependent models yielded very similar conclusions. Under either model of population growth, the largest change in λ comes with a change in the survival of pups (Figure 7.6). Elasticities are also high for changes in the survival of yearlings and young adults. Changes in the survival of older adults and changes in reproductive rates yield small elasticities (Figure 7.6). In a demographic perturbation analysis like this one, it is important to consider how much variability each demographic variable shows in the real world (Brault & Caswell 1993; Wisdom & Mills 1997). If λ is highly sensitive to a changes in a vital rate that is not free to vary (except on evolutionary time scales), the result is of little importance to conservation. For wild
97 73 61 48 36 28 13 9 2 2 2 0
0 1 2 3 4 5 6 7 8 9 10 11
0 0 0.03 0.19 0.33 0.36 0.54 0.67 1 1 1
Proportion of Females Dominant 0.75 0.84 0.79 0.75 0.78 0.46 0.69 0.22 1 1 0
Annual Survival (sx) 1 0.75 0.63 0.49 0.37 0.29 0.13 0.09 0.02 0.02 0.02 0
Survival from Birth (lx) 0 0 9 20 32 47 78 29 10 8 0
Pups Produced 0 0 0.07 0.21 0.44 0.84 3.00 1.61 2.50 2.00 0
Fecundity (mx)
With exponential growth: r ⳱ 0.044 Generation time ⳱ 5.42 years. If survival is linearly density dependent: r ⳱ 0.000 in ⬍ 2 generations Generation time ⳱ 5.35 years. A Calculation of Vx as left eigenvector of Leslie matrix assumes that population has reached stable age distribution.
Females
Age
Table 7.7 Life table for female wild dogs in Selous, combining cohort and static methods
1 1.41 1.69 2.07 2.41 2.38 1.57 1.43 0.86 0 0
Reproductive Value (Vx)A 3.82 3.75 3.30 2.92 2.56 2.00 2.15 1.67 3.00 2.00 1.00
Life Expectancy (Ex)
156 ▪ C H A P T E R 7 Table 7.8 Sensitivity (a) and elasticity (b) of population growth rate to changes in age-specific survival and reproduction, assuming population growth is exponential, with demography shown in Table 7.7A,B,C a. Sensitivity 0.186 0.134 0.26 0 0.222
b. Elasticity 0 0.008 0.186 0 0.178
0.107 0 0 0.217
0.08 0 0 0 0.187
0.058 0 0 0 0 0.133
0.043 0 0 0 0 0 0.066
0.022 0 0 0 0 0 0 0.03
0.011 0 0 0 0 0 0 0 0.009
0.005 0 0 0 0 0 0 0 0 0
0.002 0 0 0 0 0 0 0 0 0 0
0.001 0 0 0 0 0 0 0 0 0 0
0.017 0 0 0.162
0.026 0 0 0 0.136
0.036 0 0 0 0 0.099
0.066 0 0 0 0 0 0.033
0.018 0 0 0 0 0 0 0.015
0.011 0 0 0 0 0 0 0 0.004
0.003 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0
Tables are organized as Leslie matrices, so entries in the top row are sensitivity/elasticity of to changes in agespecific reproduction, ascending from age class 1 year in the first column, to age class 11 years in the last column. B Tables are organized as Leslie matrices, so entries in the subdiagonal are sensitivity/elasticity of to changes in age-specific survival, from age class 1 year in the first column, to age class 11 years in the last column. C Demographic calculations assume a birth-pulse population with a postbreeding census, both true for wild dogs in Selous. A
dogs, λ is quite sensitive to changes in the reproductive rate of pups, yearlings, and two-year-olds (Tables 7.8 and 7.9). However, the reproductive rate of pups and yearlings is fixed at 0. Even two-year-olds rarely breed (Table 7.7), due to reproductive suppression by older, socially dominant packmates (see below). In contrast, empirical data from several populations show that pup survival is highly variable (Table 7.6). Because pup survival is variable and has a strong impact on λ, it is the most important demographic variable for the conservation of wild dog populations.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 157 Table 7.9 Sensitivity (a) and elasticity (b) of population growth rate, , to changes in age-specific survival and reproduction, assuming population growth shows linear density dependence, with demography shown in Table 7.7A,B,C,D a. Sensitivity 0.187 0.133 0.263 0 0.225
b. Elasticity 0 0.008 0.187 0 0.179
0.106 0 0 0.219
0.078 0 0 0 0.188
0.056 0 0 0 0 0.134
0.042 0 0 0 0 0 0.065
0.021 0 0 0 0 0 0 0.03
0.011 0 0 0 0 0 0 0 0
0.004 0 0 0 0 0 0 0 0 0
0.002 0 0 0 0 0 0 0 0 0 0
0.001 0 0 0 0 0 0 0 0 0 0
0.017 0 0 0.162
0.026 0 0 0 0.135
0.037 0 0 0 0 0.099
0.066 0 0 0 0 0 0.033
0.018 0 0 0 0 0 0 0.015
0.011 0 0 0 0 0 0 0 0.004
0.004 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0
Tables are organized as Leslie matrices, so entries in the top row are sensitivity/elasticity of to changes in agespecific reproduction, ascending from age class 1 year in the first column, to age class 11 years in the last column. B Tables are organized as Leslie matrices, so entries in the subdiagonal are sensitivity/elasticity of to changes in age-specific survival, from age class 1 year in the first column, to age class 11 years in the last column. C Demographic calculations are for a birth-pulse population with a postbreeding census, both true for Selous wild dogs. D For this simulation, survival was linearly density dependent, with an equal impact of population density on all age classes, and reproduction was not density dependent (see text). Sensitivity and elasticity are reported for a population that has stabilized at an arbitrary carrying capacity, K. The numerical value of K does not affect these results. A
Survival and Sex Ratios Survival to independence was slightly higher for males than for females [l2 ⳱ 0.67 vs. 0.63], perhaps implying greater parental investment in the less dispersive sex. Males are significantly less likely to disperse than females (see Chapter 8). Malcolm (1979) suggested that wild dogs invest more
158 ▪ C H A P T E R 7
Figure 7.5 Sex ratios across age classes. Dotted lines show shortest unbiased 95% confidence limits.
heavily in male pups than in females because males are more likely to repay parental investment by serving as helpers once they are independent. Malcolm was applying this argument to his observation that the sex ratio of pups in Serengeti is male-biased. In Selous, the sex ratio of pups at first count does not differ from 1:1, so the stage at which investment is biased differs, but the selection pressure is the same (Hrdy 1987). Of course, any benefit due to “repayment” of parental investment would have to exceed the number of gene-copies passed on by offspring that opted to reproduce rather than helping. A more clear-cut effect of survival on sex ratios can be seen in Figure 7.5. Adult males survive better than adult females (Tables 7.2 and 7.4), and this has a cumulative effect on the sex ratio of progressively older age classes. Across all ages from two years on up, the sex ratio was 0.55M:0.45F. Males outnumbered females at most ages (Figure 7.5), and this difference was statistically significant for dogs that were seven years old or older (Z-test comparing sample and point, P ⬍ 0.05). A very similar pattern of increasing male bias among older age classes has been observed in Kruger (Reich 1981) and Serengeti National Parks between 1967 and 1978 (Frame et al., 1979; Malcolm 1979). The Serengeti population was small in this period (26 dogs in 1978), and its male-biased sex ratio (64% male) substantially increased the risk of extinction by reducing the number of breeding females (see “Demographic Effective Population Size” below). The sex ratio of wild dogs in Serengeti provides a good example of demographic stochasticity compounding the problems of a declining population.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 159
7.2 Reproduction In most wild dog packs, a single litter is produced each year by the dominant female. Mean litter size was 7.9 Ⳳ 0.5 for 36 litters in Selous. Including two cases in which an intact pack went through a breeding cycle without producing a litter, the mean number of pups born per pack year was 7.5 Ⳳ 0.6 (N ⳱ 38). Gestation lasts about 70 days, and the female whelps in an underground den. The den is usually an unoccupied warthog hole, which the female expands by digging with her forepaws—wild dogs are strong diggers, but rarely dig except at dens. Most dens are in sandy soil, and many dens consist of a complex of several holes, sometimes leading to a single chamber, sometimes leading to separate chambers. Only one den is normally occupied at any time, but it is common for the pups to be moved locally (within a few tens of meters). Dens are occasionally moved longer distances (maximum ⳱ 2.0 kilometers straight-line distance). More than one adult may carry pups in a den move, but the mother usually does most of the ferrying. The reasons for den moves are not obvious. Feces, bones, and scraps of regurgitated meat accumulate over time, and the smell is eventually palpable even to a human nose, so it is possible that dens are moved to avoid attracting other carnivores (see Chapter 11). Congruent with this explanation, dens were usually sited in dense thickets, where the likelihood of detection by other carnivores was low (see Chapter 3). Dens might be moved in response to the death of a pup, because we sometimes counted fewer pups after a move than just before, but it is also possible that pups were lost during the move. (As an interesting aside, wild dog mothers seemed well aware of whether all the pups had been moved or not. Their behavior was highly directed from the beginning of a move until abruptly stopping at the new den after the last carry.) Finally, dens might be moved in response to a buildup of parasites, but we found no evidence of this when we inspected freshly abandoned dens. Reproduction by wild dogs is highly seasonal in the Selous, as it is in other populations (Malcolm 1979; Reich 1981). Of 36 litters, 32 were whelped in June, July, or August (Figure 7.7) All litters but one were born between June and October. The timing of reproduction is closely related to rainfall (see Figure 2.3), so that pups appear at the driest part of the year. The peak of the birth season is in July (the driest month), and rainfall remains low throughout the three-month denning period. The denning period is the only time when wild dogs return to the same location each day—at other times, it is extremely rare for a pack to sleep in the same place for two consecutive days. Why is denning tied to the dry season? Reich (1981) observed the same pattern in Kruger National Park, and concluded that pups were born in the dry season so that dens could be located near permanent water, which at-
160 ▪ C H A P T E R 7
Figure 7.6 Elasticity of population growth () to changes in age-specific survival and reproduction, if survival is density dependent in a linear fashion. See Tables 7.8 and 7.9.
Figure 7.7 The frequency distribution of whelping dates.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 161
tracts high densities of ungulates. If wild dogs can den near a predictable food supply, this might reduce the cost of losing their mobility while denning, at least in terms of maintaining high rates of encounter with prey. Our data support Reich’s explanation. Hunting success did not decrease detectably during the denning period, and daily travel distances did not increase (see Chapter 5). In the wet season, ungulates were widely dispersed and relatively difficult to locate. Water was readily available throughout the area in thousands of water holes and korongos (small seasonal rivers). Long grasses in floodplains that were not edible in the dry season were palatable in the wet season, especially in areas that had recently burned. Because the distributions of food and water were less constraining for ungulates in the wet season, large herds were uncommon, and the locations of herds was unpredictable. In the dry season, ungulates congregated in large herds near permanent water, in predictable locations. On the Serengeti plains, wild dogs breed in the rainy season (February to May; Malcolm 1979), but this can also be explained by the distribution of ungulates. Unlike the Selous and Kruger, Serengeti ungulates follow a largescale annual migration. The density of ungulates in the short grass area where wild dogs were studied goes up by a factor of thousands in the rainy season, when the wildebeest, zebra, and Thomson’s gazelles arrive. From a physiological perspective, this flexibility in the timing of reproduction is interesting, because it shows that reproductive activity and quiescence are tied directly to the food supply, rather than environmental variables such as rainfall or day length. Factors Affecting Reproduction age A female wild dog’s age has a strong effect on her fecundity (Figure 7.8). Only the socially dominant female in each pack is assured of breeding, and few females attain dominance while they are young. The probability of being socially dominant increases steadily as a female ages, until age eight. All females that survived to eight years or older were socially dominant (Figure 7.9; Table 7.7). Like females, males were steadily more likely to be dominant as they aged, but only until the age of five years. After this, males were increasingly likely to lose their rank as they aged. It is not clear why males have a social and reproductive “prime” in middle age, while females continue to be dominant and highly fecund once they are old. Apparently, the sexes differ in the properties that a dog must have to stay at the top of the hierarchy, but we do not fully understand the behavioral mechanisms underlying this difference. As in most mammals, male wild dogs fight more often than females. Most reversals of dominance happened during mating periods, when alpha males are involved in more than twice as many fights as alpha females (see Figure 9.4). Mechanistically, the fact that
162 ▪ C H A P T E R 7
Figure 7.8 Fecundity as a function of mothers’ age.
Figure 7.9 The proportion of males and females that were socially dominant as a function of age. Circles ⳱ female, squares ⳱ male.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 163
Figure 7.10 Pups raised to one year as a function of mothers’ age.
males fight more than females would predict that alpha males would be more likely to lose their rank, simply by being tested more often. From an evolutionary perspective, this simply shifts the question back down a level—why do males fight more often? This may be due to phylogenetic inertia. Perhaps male wild dogs fight more often than females simply because their androgen levels are higher, traits they share with the great majority of vertebrates. We will return to this difference in our discussion of sex-biased dispersal in Chapter 8. Age-related reproductive suppression is the primary explanation for the relationship between a female’s fecundity and her age, but there was also a relationship between reproductive success and age, considering only those females that did breed. Litter size tended to increase with age, but the relationship was not significant (Table 7.7). The relationship was stronger when we regressed pups raised to independence on their mother’s age (Figure 7.10: second order polynomial regression through the origin, F2,29 ⳱ 8.42, P ⬍ 0.001). Middle-aged breeding females (three to six years old) raised more pups than breeders that were younger or older. pack size As pack size increased, so did the number of offspring born and raised. Nonbreeding adults of both sexes made contributions to the reproductive success of breeders in several ways. First, dogs in large packs obtain more meat with less effort than dogs in small packs (Figure 7.11a: Creel & Creel 1995b; Creel 1997). Second, when pups were young, nonbreeders fed them
164 ▪ C H A P T E R 7
Figure 7.11 Some of the ways that nonbreeders help to increase reproductive success. (a) Hunting is more successful and less costly in large groups. (b) A subordinate male (Deets) regurgitates meat to begging pups after returning to the den from a hunt. Genetic data excluded this male from paternity of these pups. (c) A subordinate male (Chip) guards the entrance to a den, into which the pups have retreated. (d) Adults watch for hyenas and lions while pups eat (note that only one vulture has located the carcass).
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 165
regurgitated meat and guarded them at the den (Figure 7.11b, c: Malcolm & Marten 1982). When pups were older and could travel with the pack, nonbreeders of both sexes allowed the pups to eat first, guarding the carcass while the pups ate (Figure 7.11d: Malcolm 1979). Through these mechanisms, packs with 10 adults or more produced litters
166 ▪ C H A P T E R 7
Figure 7.12 Comparison of pups produced in large and small packs.
Figure 7.13 Comparison of pups raised to one year in large and small packs.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 167
twice as large as litters in packs with 9 adults or fewer (Figure 7.12: 13.2 vs. 6.1 pups born, t36 ⳱ 2.77, P ⳱ 0.009). Moreover, packs with 10 adults or more raised three times more yearlings than packs with 9 adults or fewer (Figure 7.13: 3.4 vs. 10.4 pups raised to one year, t29 ⳱ 3.56, P ⳱ 0.001). To estimate the contribution of one adult helper, we regressed annual change in the number of pups raised on annual change in pack size (Creel & Waser 1994). This method addresses the concern that a pack might be large because it raises many young, rather than causation running the opposite direction. By regressing changes in reproductive success on changes in group size, one tests whether a reduction in pack size leads to a reduction in reproductive success (time ensures the directionality), regardless of initial pack size. By this method, each nonbreeding helper contributed 0.57 Ⳳ 0.25 pups raised to one year (t11 ⳱ 2.31, P ⳱ 0.04). Another line of reasoning suggests that the correlation between pack size and reproductive success is causal. No pack smaller than five succeeded in raising pups in Selous, and no pair without helpers has been recorded to raise pups in any wild dog population (to our knowledge). Wild dogs are truly obligate cooperative breeders. From an evolutionary perspective, it is logical that reproductive effort would increase to make use of helpers, if helpers are reliably present as part of the demographic environment (Creel & Creel 1991). If reproductive effort does increase over evolutionary time, this increase would reinforce selection on reproductive suppression of subordinates, because subordinates are more likely to accept suppression when the costs of a breeding attempt are high. Thus, obligate cooperative breeding may arise by parallel increases in reproductive effort and the degree of reproductive suppression. Wild dogs give birth to enormous litters (as many as 21; M. Mills, personal communication), and their reproductive effort is the highest on record for social carnivores, when measured as mass of the litter divided by female body mass raised to the 0.75 power (to adjust for allometry; Creel & Creel 1991). prey availability We gathered data on rates of encounter between wild dogs and their prey for six packs over 11 pack-years (see Chapter 5). Both the number of pups born and the number of yearlings raised were positively correlated with the number of prey herds encountered per kilometer traveled (pups born: r ⳱ 0.83, t10 ⳱ 4.67, P ⬍ 0.001; yearlings raised: r ⳱ 0.79, t10 ⳱ 2.85, P ⬍ 0.01). Both of these relationships were driven mainly by wildebeest, which are the dogs’ most important prey in Selous (Chapters 4 and 5), and the correlations remain significant when reproductive success is regressed on encounters with wildebeest (pups born: r ⳱ 0.89, t10 ⳱ 650, P ⬍ 0.001; yearlings raised: r ⳱ 0.87, t10 ⳱ 3.94, P ⬍ 0.01). In contrast, neither measure of reproductive success was correlated significantly with the rate of encounter with impala herds. Adult group size did not correlate significantly
0.633097 0.295855 0.217503 0.028011 0.006627
0.62778 0.603383 0.621106
R2 38 38 38 38 38 38 38 38 38
N 5 4 5 1 6 2 2 2 2
Parameters 266.5396 266.7016 267.8903 269.8434 270.2808 301.4574 309.4758 325.9568 327.6107
AICc Best model No evidence to exclude No evidence to exclude Weak evidence to exclude Weak evidence to exclude Strong evidence to exclude Strong evidence to exclude Strong evidence to exclude Strong evidence to exclude
Conclusion
For models that cannot be strongly excluded and outperform the null, the results of t-tests for each independent variable are given by **: P ⬍ 0.01; and *: P ⬍ 0.05
Year**, Females**, Females *, Males* Year**, Females**, Males* Year**, Females**, Females2*, Males2* Intercept Only (Null) Year, Females, Females2, Males, Age Females Year Males Age
2
Variables in Model
Table 7.10 Comparison of linear regression models of pups produced, using Akaike’s Information Criterion, with a sample size correction
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 169
with rates of encounter with all prey, wildebeest, or impala, suggesting that pack size and prey encounters have independent effects on reproductive success. best regression model for reproduction using AICc We used Akaike’s Information Criterion (with sample-size correction, AICc) to determine which multiple regression models best explain variation in annual reproductive success (Akaike 1983; Burnham & Anderson 1998). We fit linear regressions by maximum likelihood, with reproductive success as the dependent variable, and the following set of independent variables: year, number of adult males, number of adult females, and the alpha female’s age. (We did not include prey availability because data were not available for the majority of pack-years.) We examined two measures of reproductive success as independent variables: pups produced and yearlings raised. Following Burnham and Anderson’s guidelines on the use of AICc values to assess support from the data for each competing model (1998, pp. 127–129), we compared the set of regression models shown in Tables 7.10 (pups born) and 7.11 (yearlings raised). Whether measured as pups born or yearlings raised, the data support the hypothesis that reproductive success increases as the number of adult males or adult females increases. The data also show that the reproductive benefits due to nonbreeding helpers become smaller as pack size increases, because the regression coefficients for females2 and males2 are negative. Finally, there is significant variation among years in reproduction, populationwide. Unlike survival (see above), year-to-year variation in reproduction within the Selous population shows no detectable relationship with population density. A comparison across populations (Table 7.12) also yields no indication that litter size relates to population density (see next section). Age is a strong determinant of annual reproductive success because it is highly collinear with social dominance. Age is also correlated with the number of yearlings raised by breeding females, but does not enter the strongest multiple regression models. Comparisons with Reproduction in Other Populations Table 7.12 summarizes data on reproduction and population density from several studies of wild dogs. The mean litter size for 148 litters in six populations was 8.5 pups. Litter size varied among populations by a factor of 1.7, but this variation does not show a significant relationship with population density (Kendall’s τ ⳱ ⳮ0.35, P ⳱ 0.38). If anything, there is a weak tendency for litters to be smaller in dense populations, which suggests that litter size is not a critical variable for population dynamics and conservation. This conclusion is supported by perturbation analyses using age-structured models of population growth. As discussed above (Tables 7.7 and 7.8), pop-
0.378574 0.288909 0.012347 0.001941
0.714181 0.730347
R2 31 31 31 31 31 31 31
N 4 5 1 2 2 2 2
Parameters 176.3871 179.6389 204.3607 211.430 219.7867 240.1555 240.8054
AICc
Best model Weak evidence to exclude Strong evidence to exclude Strong evidence to exclude Strong evidence to exclude Strong evidence to exclude Strong evidence to exclude
Conclusion
For models that cannot be strongly excluded and outperform the null, the results of t-tests for each independent variable are given by **: P ⬍ 0.01; and *: P ⬍ 0.05.
Year**, Females**, Males ** Year**, Females**, Males**, Males2* Intercept Only (Null) Females Year Males Age
2
Variables in Model
Table 7.11 Comparison of linear regression models of yearlings raised, using Akaike’s Information Criterion, with a sample size correction
7.9 (36) 38
7.7 (21) 35C
Botswana1 1990–1991
Kruger 4 1990–1993 11.9 (9) 17
Kruger 2,3 1974–1978 9.1 (10)B 15–17D 7.2 (39) 15
Hwange 5 1968–1985 9.9 (11) 15
Serengeti 6,7 1967–1977 10.4 (16) 6.7
Serengeti 8,9 1985–1991 8.8 (6) 36e
Masai Mara10 1985–1990
A
For populations studied in more than one period (not continuously), data are shown separately. Table entries give mean litter size at first count with number of litters in parentheses. B Mean for 10 litters counted at 4 months of age or younger, to retain compatibility with other studies. C Density calculated from reported mean of 90 adults on 2,600 km2 area, 1989–1994. D Density estimate comes from 1964. E This estimate comes from a small area defined by the ranges of few packs, so it is high for reasons of methodology. Sources: 1: J. Bulger & J. McNutt in Fuller et al. (1992); 2: Reich (1981); 3: Pienaar 1969; 4: Maddock & Mills (1994); 5: Childes (1988); 6: Malcolm (1979); 7: Frame et al. (1979); 8: Fanshawe (1987), 9: Burrows et al. (1994); 10: Fuller et al. (1992).
†
Litter Size Density (Ad/1000 km2)
Selous 1991–1996
Population and Time Span†
Table 7.12 A comparison of litter sizes and population density in six wild dog populationsA
172 ▪ C H A P T E R 7
ulation growth is more sensitive to changes in survival rates than to changes in reproductive rates, given the demography of wild dogs in Selous (Table 7.6). Population growth would respond strongly (surprise, surprise!) if pups or yearlings were able to breed (Tables 7.7 and 7.8: entries in columns 1 and 2 of row 1), but the sensitivity of λ to changes in the reproductive rate of sexually mature adults is low. In Selous, larger packs produced more pups and raised more yearlings. Nonbreeding adult males and females both made significant contributions to reproductive success. When considering the number of pups produced, the effectiveness of nonbreeding helpers decreased as pack size increased, for both male and female helpers (Table 7.10). When considering the number of yearlings raised, the effectiveness of male helpers decreased as pack size increased, but the effectiveness of female helpers remained linear (Table 7.11). Two other studies have examined the relationship between reproductive success and pack size. In Serengeti, Malcolm & Marten (1982) found a “positive but nonsignificant relationship” between the number of yearlings raised and the number of adult nonbreeders in a pack, using ordinary least squares regression (r ⳱ 0.45, P ⳱ 0.07, N ⳱ 17). Considering the P-value, these data give some support to the hypothesis that reproductive success depends on pack size. Malcolm’s (1979) data are unusual because the Serengeti population was declining rapidly during his study, and 65% of the litters in his data set failed completely. In Kruger, the number of pups surviving at 10 months was not correlated with adult pack size or with the number of adult males, but was positively correlated with the number of adult females (r2 ⳱ 0.85, P ⬍ 0.05, N ⳱ 9; Maddock & Mills 1994). Maddock & Mills did not discuss nonlinear relationships between reproduction and group size, but a stronger linear effect of females on reproductive success is consistent with results from Selous, where the effectiveness of male helpers was asymptotic. Combining data from Selous, Kruger, and Serengeti yields a data set with 57 pack-years (Figure 7.14), and the combined data reveal a strong relationship between yearlings raised and pack size. The linear regression of yearlings raised on pack size is highly significant (b ⳱ 0.36 Ⳳ 0.08, t55 ⳱ 4.27, r2 ⳱ 0.25, P ⬍ 0.0001), but inspection of Figure 7.14 suggests that the effectiveness of nonbreeding helpers declines for very large packs, and the quadratic term of a second order polynomial regression is highly significant, with a negative coefficient (linear b ⳱ 1.12 Ⳳ 0.27, t54 ⳱ 4.22, P ⬍ 0.0001, quadratic b ⳱ ⳮ0.03 Ⳳ 0.01, t54 ⳱ 3.01, P ⬍ 0.004; total r2 ⳱ 0.36). Collectively, there is strong support for the hypothesis that reproductive success increases as pack size increases (up to 16 adults, which would be a large pack in any population). We discuss some implications of this relationship for the evolution of reproductive cooperation and conflict in Chapter 10.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 173
Figure 7.14 Reproductive success as a function of pack size, combining data across populations.
7.3 Density Dependence Our data suggest that survival is density dependent, but reproduction is not. In reading this section, keep in mind that the evidence suggesting that adult survival is density dependent is rather weak. However, variation among populations and year-to-year variation within Selous both suggest that adult survival may be density dependent. We consider it worthwhile to consider why survival might be density dependent in wild dogs, but this section is speculative. Why should survival be density dependent, while reproduction is not? A common argument from life-history theory suggests that an individual in a high-density population, faced with declining resources, should forgo reproduction if the effort involved in reproducing would threaten its own survival. The data for wild dogs contradict this argument, which is based on the principle of allocation, and assumes that density-dependent changes in vital rates are mediated by competition for food. We suspect that this is not the case for African wild dogs. Although the wild dog population in Northern Selous was at high density relative to other populations of wild dogs, they were at low density relative to populations of sympatric large carnivores (Creel & Creel 1995b). Spotted hyenas, which largely depend on the same resources, can attain densities more than 100 times higher than those of sympatric wild dogs (Chapter 11). This strongly suggests that wild dogs are not limited by
174 ▪ C H A P T E R 7
Figure 7.15 Pack size as a function of population density.
the density of their prey. One might salvage the idea that wild dogs were prey limited if dogs were less efficient hunters than hyenas, but this is not the case: Wild dogs hunt at least as successfully as sympatric carnivores, usually more successfully (Chapters 4–6). In support of the hypothesis that wild dogs are not generally prey limited, there is no detectable relationship between the densities of wild dogs and their prey (Creel & Creel 1998), but there are significant relationships between the densities of wild dogs and dominant competitors (Creel et al. in press; Chapter 11). If the population density of wild dogs is typically low enough that a change in their density is unlikely to affect intraspecific competition for prey, this raises a question: Why is survival density dependent at all? One possibility is that changes in population density affect competition for breeding opportunities, rather than competition for food. If density increases through an increase in the size of packs, then more individuals will be in competition for each breeding opportunity. Models of reproductive cooperation and conflict (Vehrencamp 1983; Reeve et al. 1998) suggest that this would increase the frequency of dispersal, which carries a high risk of mortality (Chapter 8), and might increase the frequency of fighting for reproductive opportunities by nonbreeders. In contrast, if density increases through an increase in the number of packs, then competition for breeding opportunities could remain constant or even decrease as density increases. We explore these ideas further in Chapters 8 and 10. Here, our point is that high density could increase mortality by increasing the frequency of dispersal or by increasing the rate and intensity of fighting for reproductive opportunities.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 175
Figure 7.15 shows that pack size is not related to population density. Because larger packs produce and raise more offspring (Figures 7.12–7.14), one might expect that reproductive rates would increase as density increased, through an increase in pack size. The observation that mean pack size holds quite constant as density changes does not support this expectation, and is consistent with our finding that reproduction is not density dependent.
7.4 Genetic Effective Population Size Real populations rarely match the assumptions of population genetic theory. The concept of effective population size (Ne) was developed to describe how a population can lose genetic variability (through genetic drift) more quickly than would be predicted by population size (N) alone (Wright 1931, 1938, 1943). Ne is defined as the size of an “ideal” population that would lose genetic variation at the same rate as a given real population (Wright 1969; Lande & Barrowclough 1987). In this definition, an ideal population meets five assumptions: 1. 2. 3. 4. they 5.
Population size is constant through time. The number of breeding males and females are equal. Individuals’ reproductive success is Poisson distributed. Individuals mate at random throughout the entire population (that is, are not isolated by distance). Generations do not overlap.
Because few real populations meet these assumptions, Ne is usually less than the censused population size (Nunney & Elam 1994). Quantitative methods exist to measure the reduction in Ne when each assumption is not met (Wright 1943, 1969; Crow & Kimura 1970; Lande & Barrowclough 1987). A review of effective population size in social carnivores showed that the largest reductions in Ne come from violations of conditions 2 and 3 above (Creel 1998). Skewed reproductive success, together with the biased sex ratio among breeders that arises when males and females show different degrees of reproductive suppression, resulted in Ne/N ratios of 0.2 to 0.5. Violations of conditions 1 and 4 resulted in smaller reductions, with Ne/N ratios of 0.7 to 1.0. What effect does the social structure of wild dogs have on effective population size? We evaluated the effects of skewed sex ratios and skewed reproductive success. Although the sex ratio of adults was 55M:45F, this alone would have little impact on Ne, if all of these adults were reproductive. Ne/N is 0.99 in this case (Ne ⳱ 4NmNf/(Nm Ⳮ Nf). However, the majority of these adults do not breed, and this has a profound effect on Ne. Because we do not have direct data on lifetime reproductive success, we must use an indirect method to measure the impact of reproductive suppres-
176 ▪ C H A P T E R 7
sion on Ne. Only 25.7% of adult males and 25.3% of adult females attain dominance at some point in their lives. If we assumed that no subordinates bred, and that all breeders had equal reproductive success, this would give Ne/N ⳱ 0.25. However, some subordinates do breed. In 30 pack-years of observation, there were 8 litters by subordinates and 31 litters by alphas, or 1.27 breeding females per alpha female. Multiplying 25.3% by 1.27 suggests that 32% of adult females breed at some point in their lives. Genetic data (Girman et al. 1997) showed that 3 of 29 pups in Kruger were not fathered by the alpha male. In Selous, genetic data showed that at least 2 of 9 pups were not fathered by the alpha male (see Chapter 10). Combining the data, if 5 of 38 pups had subordinate fathers, then there are 0.15 subordinate fathers per alpha male, or 1.15 fathers per litter, on average. Multiplying 25.7% by 1.15 suggests that 30% of adult males breed at some point in their lives. If 30% of adult males and 32% of adult females breed during their lives, we get an Ne/N ratio of 0.31. This measure does not incorporate the reduction in Ne due to variation in lifetime reproductive success among breeders, which is poorly known but probably substantial. With an Ne/N ratio of 0.31, we can estimate the reserve size needed to protect a population with Ne ⳱ 500. Franklin (1980) suggested that an Ne of 500 will maintain a population’s current level of heritable genetic variation, though Lande (1995) suggests that a better target might be Ne ⳱ 5000. (If Lande’s number is correct, then it is difficult to understand how any large carnivore maintains genetic variation, because their population sizes are almost always much smaller than 5000.) For wild dogs, the average density of five populations (Selous, Botswana, Kruger, Hwange, Serengeti: Table 7.12) was one adult dog per 48 km2. At this density, and with Ne/N ⳱ 0.31, a protected area of 77,420 km2 would be needed to maintain a population with Ne ⳱ 500. All of the protected areas in Africa except the Selous ecosystem are substantially smaller than this (see next section for more discussion of reserve sizes).
7.5 Demographic Effective Population Size Just as population structure can affect the loss of genetic variation, it can also affect a population’s dynamics. Caughley (1994) introduced equations for demographic effective population size (Ned), which parallel methods for measuring genetic effective population size. Caughley’s equations quantify how deviations from an even sex ratio and stable age distribution will affect a population’s growth rate (λ), assuming that growth is exponential. Because populations will often go extinct due to demographic problems before the loss of genetic variability can become a problem (Lande 1988), problems that affect λ will generally be a more pressing concern for conservation than problems that purely affect the loss of genetic variability.
D E M O G R A P H Y — S U R V I VA L A N D R E P R O D U C T I O N ▪ 177
In response to this problem, Caughley (1994) defined the demographic effective population size, Ned, as “the size of an ‘ideal’ population with an even sex-ratio and a stable age-distribution that has the same net change in numbers over a year as the population of interest” (p. 223). Caughley gives the effect of a skewed sex ratio on Ned as: N(Pf ∗ s ∗ m Ⳮ s Ⳮ 1)
Ned ⳱
1 2
(Eq. 7.1)
∗ (s ∗ m Ⳮ s ⳮ1)
where N⳱ Pf ⳱ s⳱ m⳱
population size population sex ratio (proportion female) annual survival, weighted average across ages annual fecundity, weighted average across years (births/female/ year)
Equation 7.1 weights the productivity of the population by the observed sex ratio (proportion female) in the numerator, and by an ideal sex ratio of 0.5 in the denominator. If the proportion female is less than 0.5, then Ned will be less than N. Applying this equation to Selous wild dogs (with an adult sex ratio of 55M:45F) gives an Ned/N ratio of 0.94. Because the population is male-biased, its effective size, from a demographic perspective, is 94% of censused population size. Caughley (1994) gives the effect of a deviation from the stable age distribution as: N N⳱
[
ω
兺 fx(Sxmx Ⳮ Sx ⳮ 1) x⳱0
]
ω
兺 lxe
ⳮrx
x⳱0
(Eq. 7.2)
(sxmx Ⳮ sx ⳮ 1)
where N⳱ x⳱ sx ⳱ mx ⳱ lx ⳱ fx ⳱ r⳱
population size age in years annual survival at age x annual fecundity at age x survivorship from birth to age x actual age distribution, scaled so that f0 ⳱ 1 intrinsic rate of increase
Applying this equation to data from Selous wild dogs gives an Ned/N ratio of 0.84. Equation 7.2 weights the productivity of the population by the observed age distribution (fx) in the numerator, and by the stable age distribu-
178 ▪ C H A P T E R 7
tion (lxeⳮrx) in the denominator. If the population has an excess of individuals in highly fecund age classes (relative to the expected stable age distribution), then Ned will be greater than N. For Selous wild dogs, there is an excess of individuals in young age classes that reproduce little, so Ned is only 84% of N. Combining these two effects, the total Ned/N ratio is 0.79. The average density for five populations (Selous, Botswana, Kruger, Hwange, Serengeti: Table 7.12) was one adult dog per 48 km2. Together, these data can be used to determine what area would be needed to harbor a wild dog population with a demographically effective size of 500 adults. That area is 30,380 km2, larger than all of the protected areas in Africa that hold wild dogs, other than Selous. Excepting Selous, the largest protected areas that now hold wild dogs at reasonable densities are the Ruaha/Rungwa/Kisigo complex in southern Tanzania (24,560 km2), Kruger National Park (22,000 km2), and Hwange National Park (15,219 km2). Among these areas, the largest known population is in Kruger (Maddock & Mills 1994), with 357 dogs (including pups), which gives a demographically effective size of 282 dogs. With an adult:pup ratio of about 1:1 (Maddock & Mills 1994), the demographically effective size of the population is about 141 adults. See Chapter 13 for an analysis of extinction risk for small wild dog populations. For comparison, the Selous Game Reserve is 43,600 km2, and the Selous ecosystem covers 78,650 km2. By the method of calculation used above, the reserve would have Ned ⳱ 718 adults, and the ecosystem would have Ned ⳱ 1,294 adults, respectively. Using a collection of sightings and photographs, we estimated that the Selous Game Reserve actually holds 880 adults, which gives Ned ⳱ 695 adults. Overall, our analyses of effective population size offer mixed news for wild dog conservation. From a genetic perspective, no populations are large enough to be confident that heritable genetic variability will be maintained in the long run. However, several populations are large enough that low levels of migration between populations would maintain genetic variability, and wild dogs are very capable dispersers (Chapter 8). Direct genetic data suggest that variability has been maintained, despite low densities in the past (Girman & Wayne 1997). From a demographic perspective, the Selous population is very well buffered from stochastic events, but most other populations are small enough that demographic and environmental stochasticity remains a concern. The population in Kruger is a good example. At 350–450 dogs, this is one of the stronger populations on the continent, but its demographically effective size is only about 140 adults. Plausible combinations of environmental and demographic bad luck could quickly reduce this number to a dangerously low level. In Chapter 13, we use an individual-based simulation model to quantify the effects of ecological and demographic variables on extinction risk.
8
Dispersal
Interpack dispersal has been relatively well-studied for carnivores in general, and for African wild dogs in particular. Among carnivores, males are generally the more dispersive sex, following the general rule for mammals (Waser & Jones 1983; Waser 1996). In a recent review of dispersal in carnivores, Waser (1996) found that males were significantly more likely to disperse in five of six species (dwarf mongooses, European badgers, lions, spotted hyenas, and coatis), while males and females dispersed at equal rates in one species (wolves). Because male-biased dispersal is such a robust phenomenon, there has been considerable interest in a study by Frame & Frame (1976), who found that females were the more dispersive sex in Serengeti wild dogs. As McNutt (1996) noted, this unusual reversal of the normal sex bias in mammalian dispersal has had an impact on subsequent discussions of canid social organization. Subsequent to the Frames’s study in Serengeti, other wild dog studies have cast doubt on generality of female-biased dispersal in wild dogs. In Kruger National Park, dispersal was not sex-biased, as three female groups and four male groups emigrated in 16 pack-years (the number of dispersers was not reported; Girman et al. 1997). Fuller et al. (1992b) report that dispersal was not biased by sex in the Masai Mara (again, the number of dispersers was not reported). The most complete analysis of dispersal by wild dogs is from northern Botswana. McNutt (1996) observed dispersal by 57 known-age individuals in 23 groups, dispersing from 8 packs. The proportions of males and females that emigrated did not differ, because essentially all subordinates of both sexes eventually left their pack of birth, although males dispersed in larger groups, moved farther, and waited longer before emigrating (McNutt 1996). In general, these studies suggest that the dispersal in wild dogs is often not sex-biased. The two major hypotheses to explain sex-biased dispersal are based on competition for mates and inbreeding avoidance ( Johnson & Gaines 1990; Moore 1993; Waser 1996). These hypotheses are not mutually exclusive. In its simplest form, the inbreeding avoidance hypothesis is very simple indeed—if one sex always disperses prior to breeding and the other sex never disperses, then close inbreeding will usually be avoided. In reality, the situation is much more complex, because dispersers will often be related to those they join after immigration (Creel & Waser 1994; Keane et al. 1997). For example, suppose male X leaves territory 1 and settles in a nearby territory 2, then produces a daughter who settles in territory 2. Meanwhile, back in territory 1, X’s parents produce a second son, Y, who disperses to territory 2 upon X’s death, and mates with X’s daughter. This is a mating between
180 ▪ C H A P T E R 8
closely related individuals (r ⳱ 0.25), even though they were born in different groups, are unfamiliar, and one of them has dispersed. This poses an interesting question for empirical studies of dispersal in social species: How common are events such as this one? In some carnivores, it is clear that dispersal reduces the frequency of close inbreeding to very low levels (Ralls et al. 1988), as in lions (Pusey & Packer 1987). However, increasing evidence demonstrates that short-distance dispersal, group dispersal, and nonrandomness in dispersal destinations can produce a significant degree of relatedness between dispersers and those they join (dwarf mongooses: Creel & Waser 1994; Keane et al. 1997; wild dogs: Girman et al. 1997). Genetic data and behavioral observations show that wild dogs generally avoid mating with individuals that are familiar and related by r ⱖ 0.25 (Reich 1978; Frame et al. 1979; Girman et al. 1997), but we know little about the rate of inbreeding between more distant relatives, or unfamiliar close relatives. If weak inbreeding is common for wild dogs, we do not know what effect, if any, it has on fitness. The mate-competition hypothesis is also quite simple in its basic form. Where one sex competes more intensely for mates, that sex should be less likely to disperse. Why? Because breeding opportunities are more limited, so dispersal is less likely to provide a breeding opportunity. In essence, this is very similar to the “ecological constraints model” of delayed dispersal among cooperative breeders (Emlen 1982). Mate competition is an important factor in the dispersal patterns of carnivores with reproductive suppression of subordinates, because subordinates generally disperse from large groups to small ones, where competition for mating opportunities is less intense (dwarf mongooses: Rood 1987; Creel & Waser 1994; wolves: Ballard et al. 1987; Fritts & Mech 1981; discussion by Waser 1996). By contrast, dispersal is apparently unrelated to mate competition in social carnivores that lack reproductive suppression of subordinates (lions: Pusey & Packer 1987; coatis: data from Gompper in Waser 1996). The adult sex ratio is a dominant factor affecting the level of mate competition for each sex in a population. Given this, observations from wild dogs are broadly consistent with the mate-competition hypothesis, because females were more dispersive in the highly male-biased Serengeti population (Frame & Frame 1976), but dispersal was not sex-biased in Kruger, where the sex ratio was unbiased (Maddock & Mills 1994; Girman et al. 1997). In Botswana, dispersal was not sex-biased, although the adult sex ratio was strongly male-biased (McNutt 1996). However, McNutt argued that competition for food, rather than mates, may have put equal pressure on males and females to disperse. While this explanation was not tested, it is compatible with our observation that adult survival decreases with increasing pack size in Selous (Chapter 7). Although sex bias in dispersal tendency has varied among wild dog populations, other patterns are constant (Frame & Frame 1976; Fuller et al. 1992; Girman et al. 1997; McNutt 1996). In all populations, wild dogs are most
D I S P E R S A L ▪ 181
likely to disperse at one to two years of age, though some natal dispersal occurs among older dogs, and some dogs undertake secondary emigrations. Both sexes are likely to disperse in groups, though singletons occasionally disperse. No dispersing units of mixed sex have been reported prior to this study, but we observed four cases of mixed-sex dispersal in Selous. These events are essentially the fission of one breeding pack into two. In this chapter, we describe patterns of dispersal in Selous, estimate the mortality risk of dispersal, present data on patterns of relatedness within and among packs (using microsatellites), and evaluate the roles of reproductive competition and inbreeding avoidance in driving dispersal.
8.1 Defining Dispersal in Social Carnivores We found it impossible to give a short, simple definition of dispersal in wild dogs, so we will give the (long, complex) set of rules by which we scored dispersal events. Although the definition of dispersal seems clear in principle, it can be difficult to apply to social species (see Waser 1996 for a discussion of operational difficulties). For example, suppose that all the females of a pack depart from their home range, leaving the males behind in a unisex group. These females form a new pack by joining other males, and occupy a range that overlaps their old range by 25%. After several months, a previously unknown group of females joins the original males to restore a breeding pack in a range that overlaps the males’ old range by 35%. By most definitions, both sets of females have dispersed, but what about the males? The distinction between social and locational dispersal is useful here, but it doesn’t fully resolve the ambiguities (Isbell et al. 1990; Waser 1996). The female groups have dispersed both socially (by leaving the group in which they lived) and locationally (by leaving the territory on which they lived). The males could be classified as nondispersers, because they were left behind as a subset of their original group. They have not joined new individuals (social nondispersal), nor did they ever clearly leave their home range (locational nondispersal). In this view, the males are the patrilineal core of a pack that has retained its identity. But once the new packs are established, the end result for the males that were left behind is very similar to the result for the females that left them, both socially and locationally. The only substantial difference is a transient one: Who wandered around in the search for new mates? Two important properties of the population are not affected by this transient difference: (1) the genetic consequences for the mating system, and (2) spatial patterns of gene flow within the population. Two other properties probably are affected. First, the intensity of mate competition may be greater for the sex that moves, because they are more likely to encounter packs from which they might displace the residents of their sex. Such pack takeovers are common in some social carnivores; for instance, dwarf mongooses (Rood 1987) and wild dogs (Frame et al. 1979; see below). Second,
182 ▪ C H A P T E R 8
the risk of death may be greater for the sex that moves into unfamiliar territory to search for mates. Because of the complexity of these issues, we are not certain that any simple definition of dispersal would work well for all social carnivores. Frame et al. (1979) and McNutt (1996) have grappled with operational definitions of dispersal for wild dogs in Serengeti and northern Botswana. As McNutt notes, most of the apparent difference in dispersal patterns between these populations (female-biased in Serengeti, unbiased in Botswana) is explained by a difference in the distance moved by males. Frame et al. (1979) considered many of the males in Serengeti (17 males, 5 packs) to be nondispersers after they were joined by unrelated females on a home range that heavily overlapped the males’ old range. In Botswana, more than 90% of subordinates of both sexes dispersed, but McNutt (1996) notes that the main difference from Serengeti is in the degree of overlap between old and new ranges. Lower overlap between old and new ranges led McNutt to classify more individuals as dispersers in Botswana, but patterns of relatedness within packs were probably similar in the two populations, despite dispersal rates that appear very different at first glance. McNutt’s (1996) operational definition of dispersal was this: “Successful dispersal by an individual or group is defined as an individual or single-sex group which established a new pack by pairing and breeding with an opposite sex individual or group in a post-dispersal home range that differs from their natal home range” (p. 1069). This is a good definition, but some problems remain. First, dispersal by mixed-sex groups (which we observed in Selous) is not considered a possibility under this definition. Second, this definition only considers successful dispersal, so an individual that died after emigrating but without immigrating would not be considered a disperser. Along the same lines, an individual that dispersed to join an existing pack (which we observed in Selous) would not be considered a disperser because it did not establish a new pack. Third, this definition leaves vague the criteria by which the postdispersal home range is considered “different” from the predispersal home range. Home ranges sometimes change substantially from year to year, and the home ranges of neighboring packs can overlap very substantially (Chapter 3). Consequently, the spatial relationships between packs can change substantially from one year to the next, even when there are no changes in pack membership. Pre- and postdispersal ranges also overlap in many cases, and this definition does not specify how much the ranges must have changed to be considered different. To deal with these complexities, we scored dispersal in Selous using the following set of rules: 1. Any dog that permanently left an intact pack was considered an emigrant. 2. Any unknown dog that appeared on the study site in a single-sex group
D I S P E R S A L ▪ 183
was considered an immigrant. Generally, we did not consider unknown mixed-sex groups to be immigrants, because in most cases pups were present, suggesting that these groups were established packs that we simply had not located previously. There was one exception. We located an unknown male-female pair with no pups, moving widely through the ranges of several well-established packs, in the final year of the study, in areas used by at least one known pack (Chapter 3), and classified them as immigrants. 3. Wild dog packs are highly cohesive, in comparison to many other social carnivores. It is not unusual to see a lion or a wolf away from its group mates, but a wild dog is rarely away from its pack for more than a few minutes, except when a pack kills several prey at once. A wild dog that sleeps away from its pack for just one night will generally remain apart from the pack for many days, usually permanently. Some individuals detach from their pack and, failing to immigrate elsewhere, eventually return home. We scored five such cases as unsuccessful dispersal events. One event involved four dogs that remained out of their pack for 155 days. The other four events involved single dogs that floated for 85, 20, 14, and 5 days. 4. When a breeding pack split into two groups that each held members of both sexes, we considered at least one of the new packs to have dispersed. In principle, both of the new packs could have been scored as having dispersed, if both packs occupied new ranges (more on this in point 6). In practice, we observed two of these pack splits. In both splits, one group remained on the old range and was scored as nondispersing. 5. When all of the dogs of one sex left a pack, we scored the dogs that remained behind as nondispersers in all but two cases, described here. In the first case, eight males remained as a single-sex group for 131 days. When these males joined new females, the home range of the new pack changed the local configuration of home ranges dramatically. In the second case, four females remained as a single-sex group for 314 days. The configuration of home ranges also changed, though less dramatically, when the four females formed a new pack. 6. Rules 4 and 5 relate to locational dispersal. In deciding whether an event involved locational dispersal, we compared the pre- and postdispersal configuration of home ranges for the putative dispersers and neighboring groups. If the configuration changed so that pairs of overlapping ranges were altered, we scored the event as a case of dispersal. That is, movements that created new pairs of overlapping ranges were always considered dispersal events. We emphasize that this criterion involves the spatial relationships of groups to one another before and after dispersal, rather than the distance traveled by a dispersing individual or group (which we also report). A rule based on distance alone would be too simple, due to changes in home-range configurations from year to year. If an existing pair of ranges shifted spatially, the individuals involved have physically moved, but have not dispersed in the normal sense of the term.
184 ▪ C H A P T E R 8
8.2 Number and Size of Dispersing Groups By the above criteria, we observed 58 dispersing groups—more than one per pack per year. Dispersing groups held an average of 5.1 Ⳳ 0.22 individuals per group. Dispersing units ranged in size from 1 to 12 dogs.
8.3 Rates of Dispersal Effects of Sex On an annual basis, females were significantly more likely than males to change pack membership (Figure 8.1). Among females, we observed 71 cases of dispersal (in which a female immigrated, emigrated, or both) in 214 individual-years of observation. For males, we observed 56 cases of dispersal in 258 individual-years of observation. The sample of potential dispersers for these calculations (214 females, 258 males) included all the adults and yearlings in packs that were monitored for a full year. These data give annual probabilities of dispersal of 0.21 for males (95% CI ⳱ 0.175– 0.265) and 0.33 for females (95% CI ⳱ 0.275–0.405), which differ significantly (comparison of Poisson rates, Z ⳱ 2.49, P ⳱ 0.013). On average, females were 11⁄2 times more likely than males to move between packs in a given year. Effects of Age Dogs of both sexes were most likely to disperse in their first year of sexual maturity (as two-year-olds), though a substantial number of yearlings also dispersed (Figure 8.2). As noted above, dispersal was more common among females, and Figure 8.2 shows that females dispersed more frequently than males at all ages but two. Males showed a very pronounced increase in the likelihood of dispersal as two-year-olds, compared to females, but the probability of dispersal declined rapidly for males beyond the age of two years. Overall, sex did not affect the mean age of dispersal, which was 2.55 Ⳳ 0.15 years for males and 2.72 Ⳳ 0.15 years for females. For the sexes combined, the mean age of dispersal was 2.64 Ⳳ 0.11.
8.4 Size of Dispersing Groups Twenty dispersing units held only males, averaging 3.95 Ⳳ 0.57 males per group (range 1 to 8). Thirty-four dispersing units held only females, averaging 2.35 Ⳳ 0.27 females per group (range 1 to 7). Four dispersing units were of mixed sex, averaging 7.5 Ⳳ 2.06 dogs per group. These means are
D I S P E R S A L ▪ 185
Figure 8.1 Annual rates of dispersal for females and males, with 95% confidence limits.
Figure 8.2 Frequency distribution for the age of dispersal in males and females. Natal and secondary dispersal events are included in these distributions.
useful for comparison to other species, but they do not directly reveal the group sizes in which most individuals disperse. For example, suppose we observed two dispersing groups of 10, and two groups of 2. The average group size is 6, even though 20 of 24 individuals (83%) dispersed in groups of 10. To consider the selection pressures affecting dispersal, the number of
186 ▪ C H A P T E R 8
Figure 8.3 Frequency distribution for the size of dispersing groups of males and females.
individuals that dispersed in each group size is a better measure than the mean. By this measure (Figure 8.3), males dispersed in groups that averaged 5.3 dogs (95% CI ⳱ 4.8–5.8). Females dispersed in significantly smaller groups, averaging 3.4 dogs (95% CI ⳱ 3.1–3.9).
8.5 Linear Dispersal Distance We recorded dispersal distances for 77 individuals, measured as the linear distance between the geometric centers of their predispersal and postdispersal home ranges. The mean dispersal distance was 29.6 kilometers, dropping to 27.3 kilometers if we include six values of zero for dogs that returned to their pack of origin (Table 8.1). The maximum dispersal distance was 76.5 kilometers, by a group of four females. It is likely that some individuals moved farther but we did not relocate them. We routinely checked for signals from radiocollared animals that had disappeared when we tracked from the air, but we did not fly often and some dispersers were never relocated, suggesting that they had moved well off the study site (or the transmitter had failed, but no transmitters failed among nondispersers). The minimum dispersal distance was zero kilometers, for 6 females (not all in one group) that returned to the group from which they had emigrated after failing to immigrate elsewhere. Excluding zeros, minimum dispersal distance was 5 kilome-
D I S P E R S A L ▪ 187 Table 8.1 Dispersal distances for males and females, measured as the linear distance from the geometric center of the predispersal and postdispersal home ranges N
Mean Ⳳ SE
Minimum
Maximum
Males Including returnees Excluding returnees
39 39
26.7 Ⳳ 1.8 26.7 Ⳳ 1.8
5.0 5.0
47.2 47.2
Females Including returnees Excluding returnees
38 32
27.8 Ⳳ 3.9 33.1 Ⳳ 4.0
0 9.4
76.5 76.5
Sexes Combined Including returnees Excluding returnees
77 71
27.3 Ⳳ 2.1 29.6 Ⳳ 2.1
0 5
76.5 76.5
ters for a single male. There was no detectable difference in the distances dispersed by males and females (Figure 8.4 and Table 8.1).
8.6 The Duration and Circumstances of Floating For 49 individuals whose dates of emigration and immigration were known to within a few days, we recorded the duration of floating between packs, which we defined as time spent alone or in a single-sex group, usually traveling widely. The composition of floating groups sometimes changed as floaters joined one another or split up. For individuals in floating groups of shifting composition, joining another single-sex group of the same sex did not count as an immigration. We made the decision to adhere to a definition of floating as “time spent outside of a potential breeding pack.” The mean duration of floating was 145 Ⳳ 18 days (Figure 8.5). Female dispersers floated more than twice as long as male dispersers (females: 187.8 Ⳳ 26.2 days; males: 82 Ⳳ 11.2 days, t48 ⳱ 3.71, P ⳱ 0.007: see “Mortality Risk of Dispersal,” below), and old dispersers floated for longer periods than young ones (Figure 8.6; b ⳱ 47.0 Ⳳ 12.9, t47 ⳱ 3.63, P ⳱ 0.007). It is unlikely that this effect of age is due to sex differences, because males and females dispersed at similar ages (Figure 8.2). A possible explanation is that older dispersers are less likely to immigrate into subordinate positions, which do not allow them to breed immediately. If so, older dispersers would have to search more widely for a suitable opportunity (see “Escape from Reproductive Suppression,” below). On three occasions, dogs transferred directly between packs in a single day. In one case, three females (all two-year-olds) transferred to a pack that was recently formed by littermates. The other two cases occurred simul-
188 ▪ C H A P T E R 8
Figure 8.4 Frequency distribution of linear distances moved by dispersing males and females.
Figure 8.5 The duration of floating between packs, for males and females.
D I S P E R S A L ▪ 189
Figure 8.6 The duration of floating between packs, as a function of age.
taneously, when two packs fought seriously in the overlap zone of their ranges. Pack 1 held a single female, and Pack 2 held three females (the alpha and two of her two-year-old daughters). The alpha female of Pack 2 was mortally wounded in the fight, but it took her several days to die. During this time, the alpha female from Pack 1 steadily integrated herself into Pack 2. The dying alpha female of Pack 2 constantly attempted to attack the immigrant, but was too badly wounded to harm her. The males of the pack interacted normally with both females. Meanwhile, the two daughters from Pack 2 established themselves in Pack 1 on the day after the fight. The longest float was 485 days, by a single female, Blondie. Blondie’s history illustrates the loose social ties of some floaters, both to their original pack and to other floaters. Blondie emigrated by herself as a three-year-old, from a large pack that included a cohort of 11 yearling females. She was often harassed by coalitions of yearling packmates prior to her departure, and while this might not be a forced emigration, it appeared that Blondie was becoming subordinate to the entire yearling cohort, which is unusual (females rarely lose rank to younger packmates; Chapter 7). Over the next year and a half, Blondie associated with at least four other females that had emigrated from her natal pack at different times, all of them using an area that overlapped their former home range. The composition of this group changed frequently. Subsets of the floating group interacted amicably with their old pack on two occasions, sharing a carcass on one occasion. Several of these floaters eventually re-immigrated into their natal pack, but Blondie
190 ▪ C H A P T E R 8
ultimately formed a new pack with three other females from her natal pack, one of whom had floated for 354 days, and two for 173 days. Unlike Blondie’s group, some floating groups are very cohesive, as illustrated by a group of five two-year-old males. These littermates emigrated together, and 23 days later emigrated en masse into a pack 18.2 kilometers away, which was composed of an adult pair and two large female pups. Interestingly, the immigrants did not kill or displace the resident alpha male, though he was immediately subordinate to all of the immigrants. It is very unusual for immigrants to join a pack with unfamiliar adult residents of the same sex, indeed we are not aware of any other cases on record (Frame et al. 1979; McNutt 1996). A plausible interpretation for this behavior is that small packs are undesirable with respect to foraging success and reproductive success (Chapters 5 and 7). Perhaps immigrants are more tolerant of same-sex residents—provided they are socially submissive—in situations where pack size is near the lower threshold for successful reproduction. After one month, these males emigrated again. After floating for 92 days, they immigrated into another pack, 41.2 kilometers from their natal home range. This pack held two adult males and three adult females. On the same day that the five floating males immigrated, a solitary adult female also joined the group. We do not know whether this female was associated with the males prior to that day, or was attracted by the loud hoo calls exchanged by the immigrants and residents as they joined each other. This event was extraordinary, because none of the resident adult males or females were displaced by the immigrants, so a single, one-day event merged two lineages of unrelated males and two lineages of unrelated females.
8.7 Comparison with Dispersal in Other Wild Dog Populations As discussed at the beginning of this chapter, female-biased dispersal was originally reported for wild dogs in Serengeti (Frame et al. 1979), but subsequent studies in northern Botswana (McNutt 1996), Masai Mara (Fuller et al. 1992b), and Kruger (Girman et al. 1997) found no detectable sex bias in dispersal rates. In Selous, the annual probability of dispersal was significantly higher for females than for males, muddying the waters a bit. Of these five study populations, female-biased dispersal has been found in Serengeti and Selous, the two populations that are the most dissimilar in ecology, demography, and population density. Given this, the cause of sex-biased dispersal is difficult to identify. In Serengeti and northern Botswana, males dispersed farther than females (Frame et al. 1979; McNutt 1996), but the dispersal distances of males and females in Selous were similar. In Botswana, males dispersed in significantly larger groups than females, and the same was true for wild dogs in Selous. Overall, many aspects of dispersal vary among wild dog populations.
D I S P E R S A L ▪ 191
These differences are likely to be tied to differences in population density, demography, and social structure, all of which mediate the fitness effects of dispersal. For instance, adult sex ratios differ among populations, and this should affect the intensity of competition for mating opportunities for males and females. Differences among populations in mean pack size could also affect the intensity of mate competition. Differences in adult survival rates could affect the likelihood of close inbreeding, by altering the probability that parents are still alive when offspring reach breeding age. A quantitative analysis of these factors is not possible with published data, but would be interesting.
8.8 Mortality Risk of Dispersal Dispersal is typically dangerous for species that move long distances and float for long periods. Waser (1996) summarized evidence that dispersers have higher mortality rates than nondispersers in mongooses, wolves, badgers, and lions. Carnivores are well-armed, and some of the mortality risk of dispersal is due to escalated fights between would-be immigrants and residents of the groups that they try to join (Fritts & Mech 1981; Packer & Pusey 1982; Rood 1987; Creel et al. 1993). The risk of death might also increase for dispersers because they are on unfamiliar territory. Finally, if foraging success depends on group size, as in wild dogs, then small groups of dispersers must work harder to obtain (and defend) sufficient food. Smaller groups might also be more vulnerable to predation. The mortality risk associated with dispersal is difficult to measure, particularly in large carnivores, which are elusive and mobile (Waser et al. 1994). Wild dogs are known to move more than 250 km (Fuller et al. 1992b) so it is always possible that an individual that has disappeared is alive somewhere off the study site. In most studies (including ours) only one or two dogs were radiocollared in each pack, so most emigrants cannot be systematically relocated by aerial tracking. The most common method of dealing with undetected emigration is to assume that the study area is neither a source nor a sink for dispersal. If this assumption is true, then the number of previously unknown immigrants into the study population will on average be equal to the number of undetected emigrants leaving the study population. With this assumption, one can estimate the survival rate of dispersers as the ratio of immigrants to emigrants, or (噛 of individuals that emigrated and immigrated within the population Ⳮ 噛 of previously unknown immigrants) ⳰ (噛 individuals known to emigrate Ⳮ 噛 of previously unknown immigrants). Unfortunately, it is almost impossible to test the source-sink assumption upon which this calculation rests. Waser et al. (1994) discuss methods to produce confidence limits for the mortality rates of dispersers in the face of uncertainty about the fates of animals that disappear.
192 ▪ C H A P T E R 8
Because the density of wild dogs on our study area was higher than on the surrounding areas, it would probably be false to assume that our population was not a source of emigrants. Consequently, we adopted a different method to estimate the mortality risk of dispersal, by comparing the sizes of emigrating and immigrating groups. During our study, we recorded the immigration of 33 groups or individuals and the emigration of 32 groups or individuals. Because the number of emigrating and immigrating groups is almost identical, we are confident that we did not miss many cases in which a floating group was present on the site, but went undetected, with all members dying before they joined another pack. Obviously, undetected deaths of entire floating groups would bias our estimate of dispersal risk if they were common. However, if such cases were common, then we should have recorded more cases of emigration than immigration. This logic is reinforced by the high density of wild dogs on our main study site. An area of high density is unlikely to be a sink for immigration. Consequently, it is unlikely that undetected deaths of entire floating groups were masked by a high rate of immigration onto the site. If the undetected failure of complete dispersing groups is uncommon, then the ratio of immigrating group size to emigrating group size (including singletons in both cases) is a good measure of the probability of surviving dispersal. In Selous, dispersing groups were smaller when they immigrated than when they emigrated. A one-way ANOVA revealed significant variation in the sizes of groups of emigrants, immigrants, previously unknown transients, and groups that transferred directly between packs without floating (F4,182 ⳱ 5.95, P ⬍ 0.001). A post-hoc comparison of immigrant and emigrant groups (Tukey’s honest LSD, P ⬍ 0.001) confirmed that group size decreased significantly between emigration and immigration (Figure 8.7). By this method, mortality is 20% for dispersing males (shortest unbiased 95% CI ⳱ 0.12–0.28), and 29% for dispersing females (shortest unbiased 95% CI ⳱ 0.19–0.37). The risk of dying during dispersal is substantial (with upper confidence limits of about 1 in 3 dispersers dying), and tends to be higher in females, though not significantly so. Recall that females floated between packs for substantially longer periods than males (0.52 and 0.23 years, respectively), which might underlie the trend for mortality to be higher for female dispersers. After adjusting for differences in the duration of floating, the annual risk of death was significantly higher for male dispersers (0.88: 95% CI ⳱ 0.80–0.94) than for females (0.56: 95% CI ⳱ 0.46–0.66). For comparison, the annual risk of death for nondispersers between the ages of two and six (an interval that includes the majority of dispersers) was 0.13 for males and 0.21 for females. Thus, dispersal increased annualized mortality rates by a factor of 6.8 in males, and by a factor of 2.7 in females. These estimates of risk reveal three properties of dispersal in wild dogs:
D I S P E R S A L ▪ 193
Figure 8.7 The sizes of emigrating and immigrating groups of males and females, which reflects the risk of death during dispersal (see text).
1. Dispersal is dangerous. For both sexes, the probability of dying was far higher for a dog involved in dispersal than for one living in a stable pack. 2. The absolute risk of dying during dispersal was 1.5 times greater for females than for males. This sex difference is explained by more prolonged exposure to the risks of dispersal, because females floated for longer periods. Because risk is related to the duration of floating, this suggests that part of the cost of dispersal arises from day-to-day life in a small group (rather than the risks of fighting during immigration). Dispersing groups are much smaller than most packs (compare Table 8.2 with data on pack sizes in Chapter 5). Small groups hunt less effectively (Chapter 5) and have difficulty defending carcasses (Fanshawe & Fitzgibbon 1993). They also might be less effective at defending themselves against attacks by lions, but we have no data to test this idea. 3. Expressed per unit time, dispersal was 1.6 times riskier for males than for females. Some of the risks of dispersal do not depend on the duration of floating—for example, the costs of fighting with residents during an immigration attempt (Frame & Frame 1979). Our data suggest that these time-independent risks of dispersal are more serious for males than for females.
194 ▪ C H A P T E R 8 Table 8.2 Risk of death during dispersal. See text for methods
Sex
Emigrant Group Size
Immigrant Group Size
Survival
Mortality
Float time (yrs)
Annualized mortality
Males Females
5.63 3.87
4.51 2.76
0.80 0.71
0.20 0.29
0.23 0.52
0.88 0.56
8.9 Dispersal and Escape from Reproductive Suppression Because social subordinates rarely breed, most adult wild dogs are reproductively suppressed at any given moment (Chapters 7, 9, 10). For species that live in groups with a single breeding pair, the most direct test of the hypothesis that mate competition drives dispersal is to ask whether dispersers improve their odds of becoming breeders, relative to nondispersers. For a subordinate female, the annual probability that the alpha female will die is 0.21 (Table 7.1). On average, there were 3.24 subordinate females per pack, so the annual probability of becoming dominant by succession was 0.065 ( ⳱ 0.21 ⳰ 3.24). For males, the same calculation yields an annual probability of succession of 0.058. Unlike females, males sometimes fought and displaced an older male to become alpha. The annual probability of obtaining dominance by displacement, rather than succession, was 0.04. Altogether, the annual probability of attaining a breeding position by a mechanism other than dispersal was 0.065 for females, and 0.098 for males (Figure 8.8). Among dispersers, 11 of 78 immigrating males (0.124) became dominant in the pack they joined or formed within the year of immigration. Because females dispersed in smaller groups, a higher proportion, 0.178 (16 of 74), of immigrating females became dominant within the year of immigration. Females that disperse improve their annual odds of becoming dominant by a factor of 2.7 (⳱ 0.178 ⳰ 0.065), relative to nondispersers, so escape from reproductive suppression is a strong selection pressure favoring dispersal (Figure 8.8). For males, selection to escape from reproductive suppression is a much weaker force (Figure 8.8), increasing the odds of becoming dominant only by a factor of 1.3 (⳱ 0.124 ⳰ 0.098). Of course, these benefits must be weighed against the substantial cost of dispersal (see “Mortality Risk of Dispersal,” above), but the fact that successful dispersers are more likely to breed indicates that escape from reproductive suppression is a selection pressure favoring dispersal. As a mechanism to escape from reproductive suppression, dispersal is twice as effective in females as in males. This difference depends very strongly on the fact that alpha males can be supplanted by subordinate pack-
D I S P E R S A L ▪ 195
Figure 8.8 The annual probability of attaining dominance via dispersal and nondispersal, for males and females.
mates, but alpha females were never supplanted by packmates. As we discussed in Chapter 7, we do not understand why the sexes differ in this aspect of social dominance. Somehow, the properties required to maintain dominance differ in males and females (see Figure 7.9). Although we do not understand this difference between the sexes, it has a clear impact on the utility of dispersal as a means of escaping reproductive suppression.
8.10 Dispersal and Escape from Inbreeding Demographic Data For 80 emigrants, we knew whether the dominant individual of the opposite sex had or had not changed since the birth of the emigrant. Although the alpha male and female are not the parents of every pup, genetic data confirm that the alphas produce most of the offspring in wild dogs (Girman et al. 1997) and other cooperatively breeding carnivores (Keane et al. 1994). We do not have genetic data for most dispersers, so we simply consider the alphas to be putative parents for this analysis. For females, only 1 of 44 dogs left a pack in which the alpha male was still her putative father (Figure 8.9). For males, 17 of 35 dogs left a pack in which the alpha female was still his putative mother (Figure 8.9). All 17 of the males that dispersed from packs in which their mother was still the breeder did so as yearlings or two-year-olds. Altogether, 61% of one- and two-year-old emigrant males dispersed from a pack in which their mother was the breeder (95% CI ⳱ 42–77%). If the emigration of one- and twoyear-olds was not affected by the survival of their mothers, the random expectation (based on survival rates) is that 72% would emigrate from a pack in which their mother was still the breeder (95% CI ⳱ 65%–78%). Overlap
196 ▪ C H A P T E R 8
Figure 8.9 The number of cases in which emigrants avoided potential parent-offspring matings.
between the observed and expected confidence intervals suggests that males’ dispersal decisions are not actively conditioned upon the survival of their mother. Nonetheless, males are likely to emigrate at age two (Figure 8.2), and they consequently emigrate away from potential parent-offspring matings at significantly higher rates than females. Forty-nine percent of dispersing males avoided a potential parent-offspring mating, compared to only 2% of dispersing females (χ 2 ⳱ 23.8, P ⬍ 0.001). The low proportion of females that face father-daughter matings is a consequence of higher turnover for breeding males than for breeding females (Chapter 7). Genetic Data We can also use genetic data to relate dispersal to inbreeding avoidance. While our genetic data are limited, this approach has the distinct advantage of considering relatedness as a continuous variable, rather than looking only at putative parent-offspring matings. We have genotypes for three microsatellite loci from 64 individuals in 10 packs (labwork was done by Robert Wayne, Derek Girman, and Carles Vila). These loci had 9, 8, and 6 alleles, with allele frequencies for the most common allele of 0.33, 0.25, and 0.62, respectively. We used Queller & Goodnight’s (1989) method of calculating relatedness with jackknifed standard errors, using the program Relatedness 4.2. See Queller & Goodnight (1989) for a description of the interpretation of these r values, including negative values, or, for the purposes of this analysis, simply note that it is differences in r values with which we are concerned, not absolute values. Of these 64 dogs, 22 dispersed during the study. To test for the effect of dispersal on patterns of relatedness within packs, we used two approaches. First, we calculated mean relatedness for all dyads within a pack, dyads of
D I S P E R S A L ▪ 197
the same sex, and dyads of the opposite sex. We calculated relatedness within these classes first with dispersers in their pack of origin (Figure 8.10a), and then with dispersers in their pack of destination (Figure 8.10b). Not surprisingly, mean relatedness within packs decreased when dispersers were scored in the pack of destination, from r ⳱ 0.11 Ⳳ 0.033 to r ⳱ 0.05 Ⳳ 0.051. This overall decline was due to a decrease in relatedness between individuals of opposite sex, from r ⳱ 0.12 Ⳳ 0.072 to r ⳱ ⳮ0.07 Ⳳ 0.065. For dyads of the same sex, relatedness dispersal did not decrease relatedness at all (predispersal r ⳱ 0.10 Ⳳ 0.025; postdispersal r ⳱ 0.12 Ⳳ 0.104). This set of changes is consistent with behavioral observations of dispersers (Frame et al. 1979; McNutt 1996; Girman et al. 1997). Usually, groups of closely related individuals of a single sex (usually littermates or litters from two consecutive years) join an unrelated or distantly related lineage of the opposite sex. The immigrants may either join a group of the opposite sex to form a new pack, or may join an existing pack and displace the lineage of their own sex, but the genetic result is the same. In either case, relatedness to the opposite sex declines (decreasing the degree of inbreeding), but relatedness to individuals of the same sex remains high (facilitating kin selection; see Chapter 10). To examine the effect of dispersal on inbreeding more directly, we calculated relatedness of alphas of each sex to packmates of the same sex and of the opposite sex. Figure 8.11 shows the degree of relatedness for these dyads before dispersal (part a) and after dispersal (part b). For alpha males and for alpha females, dispersal considerably decreased the degree of relatedness to opposite-sexed packmates. For alpha females, relatedness to potential mates declined by a factor of 6.25 (from r ⳱ ⳮ0.04 Ⳳ 0.062 to r ⳱ ⳮ0.25 Ⳳ 0.118; t ⳱ 2.33, df ⳱ 9, P ⬍ 0.02). For alpha males, relatedness to potential mates declined by a larger amount, from r ⳱ 0.06 Ⳳ 0.014 to r ⳱ ⳮ0.24 Ⳳ 0.12 (t ⳱ 2.31, df ⳱ 22, P ⬍ 0.02). For males, genetic data confirmed the conclusion from demographic data that dispersal was related to inbreeding avoidance. For females, demographic data did not suggest that dispersal was related to avoiding parent-offspring matings, but genetic data did show that dispersal was effective in avoidance of more distant inbreeding. Dispersal does not decrease the degree of relatedness of alphas to individuals of the same sex (Figure 8.11). For alpha males, dispersal has very little effect on relatedness to packmate males. For alpha females, dispersal actually increases relatedness to packmate females, because groups of closely related females generally immigrate together. Patterns of relatedness within sexes are of interest from the perspective of kin selection and the evolution of reproductive suppression (see Chapter 10), but for this discussion, they strengthen the conclusion that dispersal is related to inbreeding avoidance. That relatedness to potential mates declines, while relatedness to dogs of the same sex holds level (males) or increases (females), suggests inbreeding avoidance.
198 ▪ C H A P T E R 8
Figure 8.10 The effect of dispersal on patterns of relatedness between individuals living in the same pack, using Queller & Goodnight’s (1989) estimator of relatedness for microsatellite data. Mean Ⳳ SE from jackknife procedure in program Relatedness 4.2.
D I S P E R S A L ▪ 199
Figure 8.11 The effect of dispersal on patterns of relatedness between dominant individuals (breeders) and packmates of the same sex and of opposite sex (potential mates), using Queller & Goodnight’s (1989) estimator of relatedness for microsatellite data. Mean Ⳳ from jackknife procedure in Relatedness 4.2. Dyad codes: aFF ⳱ alpha female/packmate females, aFM ⳱ alpha female/packmate males, aMM ⳱ alpha male/ packmate males, aMF ⳱ alpha male/packmate females.
200 ▪ C H A P T E R 8
8.11 Integrating Forces that Drive Dispersal Our data show that both mate competition and inbreeding avoidance play a role in dispersal, but the balance of forces driving dispersal differs for males and females. For both sexes, dispersal carries a substantial risk of death. The risk per unit time is significantly greater for males than for females, which perhaps explains why males disperse less often than females and float for shorter periods. Offsetting the cost of dispersal, females gain a substantial increase in the probability of becoming dominant (and therefore breeding). The annual reproductive success of breeders is high (Chapter 7), so escape from reproductive suppression provides a substantial fitness benefit for dispersing females. Dispersal also allows escape from reproductive suppression for males, but the benefit is smaller than for females (Figure 8.8). Based on demographic data, the effect of dispersal on inbreeding avoidance also shows a sex bias. Only 2% of females avoid potential parentoffspring matings by dispersal, compared with 49% of males. If incest between parent and offspring carries a cost for wild dogs, this cost is more often avoided by dispersing males than by dispersing females. Based on genetic data, which also accounts for mating between more distant relatives, dispersal patterns reduce the degree of inbreeding significantly for both males and females. There is substantial evidence to show that closely inbred offspring often have low viability or fertility in many species (Pusey & Wolf 1996). However, there are also examples that reveal no evidence of inbreeding depression or avoidance, despite substantial sample sizes (e.g., dwarf mongooses: Keane et al. 1997), and the ubiquity and impact of inbreeding depression in the wild is subject to debate (Pusey & Wolf 1996). For wild dogs, our analysis is limited by a lack of data on the costs of inbreeding depression, if there are any. Despite this limitation, our data suggest that dispersal in wild dogs is driven by a combination of inbreeding avoidance and competition for mating opportunities. Indirectly, this suggests that inbreeding probably does carry a fitness cost for wild dogs. Because most wild dogs are reproductively suppressed, competition for mating opportunities is intense, relative to most other species (Chapter 9). With the mating system of wild dogs, most adults are not able to breed in a given year, and, consequently, one might expect that any mating opportunity, even one that involves inbreeding, would be taken. That wild dogs avoid inbreeding despite intense competition for mating opportunities is indirect evidence that inbreeding would carry a fitness cost.
9
Reproductive Suppression, Social Stress, and the Behavioral and Endocrine Correlates of Rank
Animal groups are often stratified by a dominance hierarchy (Figure 9.1). Because dominant individuals are more likely to win contests for food or mates, they should generally have greater reproductive success than subordinates (Frank 1986; Le Boeuf & Reiter 1988; Frank et al. 1995; Pusey et al. 1997), though dominance may also carry costs (Packer et al. 1995; Wasser 1995; Creel et al. 1996). Social rank can have a dramatic effect on reproductive success in cooperatively breeding species, where dominant individuals sometimes monopolize reproduction completely. Among carnivores, dominance is generally determined by age and body size (Creel et al. 1992), and in some species by the rank of relatives (spotted hyenas: Holekamp & Smale 1993). Studies in captivity show that differences in dominance are often mirrored by differences in circulating levels of sex steroids and glucocorticoids (Abbott 1993; Blanchard et al. 1995), but relatively little is known about the endocrine correlates of dominance in the wild (Sapolsky 1985; Wingfield et al. 1991; Mays et al. 1991). Dominance may directly affect sex-steroid levels, though relationships between dominance and steroid hormones are notoriously variable (Zielinski & Vandenbergh 1993; Berkovitch & Clark 1995). Less directly, subordination can increase glucocorticoid stress hormone levels (Christian & Davis 1964; Bronson & Eleftheriou 1964; Blanchard et al. 1995), which can inhibit the secretion of sex steroids and impair reproductive behavior (Welsh & Johnson 1981; Wingfield & Silverin 1986). It is widely assumed that glucocorticoid stress hormones mediate reproductive suppression in cooperatively breeding societies (“social stress” or “psychological castration”: Brown 1978; Wingfield et al. 1991), but few studies have tested this hypothesis with data from the wild (Mays et al. 1991; Schoech et al. 1991; Wingfield et al. 1991). Most studies of social stress have been conducted in captivity (Berkovitch & Clarke 1995), with frequent and intense aggression that subordinates cannot avoid. For example, dominance and subordination in rodents is commonly assessed by pairing individuals that are unfamiliar with one another and observing the fight that
202 ▪ C H A P T E R 9
Figure 9.1 Agonistic and aggressive behavior in wild dogs. (a) Overt signals: The alpha male stares directly with ears forward, erects and flares tail, while subordinates redirect gaze, hold ears and tails low, cringe, and fall to ground. (b) A subtle signal: Ear position indicates low-intensity submission while being approached by a higher-ranking dog. (c) Ambiguity: Agonistic behavior sometimes grades into play, even among adults. During play bouts, rank relationships are often difficult to assess.
R E P R O D U C T I V E S U P P R E S S I O N ▪ 203
ensues (Brain 1983; Zielinski & Vandenbergh 1993). Such contests may be a good model of short-term events in the wild (e.g., immigration events: Alberts et al. 1992; Creel et al. 1993) but they do not mimic agonistic interactions in a stable group with an established hierarchy (Creel et al. 1996). Allout fighting is rare in most social groups. Thus the behavioral and endocrine consequences of social subordination in the wild may not conform to patterns reported from captivity. More is known about the behavioral and endocrine correlates of dominance in males than in females (Sapolsky 1985; Zielinski & Vandenbergh 1993; Arnold & Dittami 1997; but see Harcourt 1987). This imbalance is unfortunate, because the effects of subordination are likely to differ between males and females (Kleiman 1980; Creel et al. 1992, 1996). The sexes differ in basic reproductive function (cyclical ovulation vs. continuous production of sperm), in the energetic costs of reproduction (costly gestation and lactation in females only), and in life-history profiles (e.g., likelihood of dispersing, which allows for escape from reproductive suppression). Through evolutionary time, any of these differences could alter the ways in which the neuroendocrine system responds to social status. In this chapter we describe the behavioral and endocrine correlates of subordination in both sexes, and ask how sex differences in mechanisms of reproductive suppression relate to social organization. We address six questions: (1) What is the effect of social subordination on mating rates? (2) To what degree is reproductive suppression of subordinates due to aggression from dominants? (3) Is reproductive suppression of subordinates strictly a behavioral process, or is it physiologically mediated by depressed sex-steroid levels? (4) Is suppression of subordinates mediated by stress? (5) How do mechanisms of suppression differ between males and females? (6) How do behavioral and endocrine patterns relate to wild dogs’ social organization? Most wild dog packs produce a single litter in the season of prey abundance (Malcolm 1979; Reich 1981), with only the dominant female assured of becoming pregnant (Figure 9.2). Alpha females produced 81% of 85 litters in Kruger National Park (Reich 1981; M.G.L. Mills, personal communication) and 75% of 57 litters in Serengeti National Park (Malcolm & Marten 1982; Burrows 1995). If a subordinate female gives birth, her pups are usually born several days after those of the dominant. Subordinates’ pups are sometimes killed by the dominant female, but are often creched with the dominant female’s pups and raised (see Chapter 10; van Lawick 1973; Frame et al. 1979; Fuller et al. 1992a). The estrogen and progesterone profiles of two captive subordinate females suggested that they were capable of ovulation, but they did not ovulate during the annual mating period and did not become pregnant (van Heerden & Kuhn 1985). For males, the degree to which subordinates are excluded from reproduction is not well known. Anecdotal data suggest that dominant males defend access to estrous females
204 ▪ C H A P T E R 9
Figure 9.2 An alpha female near the end of her 70-day gestation. Wild dogs produce very large litters, which are extremely heavy relative to the female’s body mass.
and mate more often than subordinates (Goodall & van Lawick 1971; Malcolm 1979; Reich 1981). A genetic study in Kruger found that the alpha male fathered most pups, but at least 2 of 10 litters had multiple paternity (Girman et al. 1997). Limited genetic data from Selous also detected multiple paternity. Multiple paternity is common in other social carnivores, creating the possibility of hidden reproduction by subordinate males (dwarf mongooses: Keane et al. 1994; lions: Packer et al. 1991; Ethiopian wolves: Gotelli et al. 1994). Mating spans three to seven days, but estrous behavior builds gradually over several weeks prior to mating. A strengthening of the bond between the alpha male and female is the first overt behavioral sign that a mating period is approaching. The alpha pair are the only opposite-sexed adults that normally rest together (slightly apart from the others; for example, under a tree adjacent to the rest of the pack) and they interact often, relative to other dyads of opposite sex. The association between alphas becomes more overt as the mating season begins. The female’s vulva becomes swollen during estrus, sometimes accompanied by a slight bloody discharge. Mating attempts by the male are generally not tolerated by the female for several days. She rejects mounts by moving away, lying down, or snapping at the male. The female gradually stops these behaviors, and eventually stands firmly with her tail averted to one side as the male mounts. Many mounts do not lead to a full copulation. Full copulations include a copulatory lock in at
R E P R O D U C T I V E S U P P R E S S I O N ▪ 205
least some cases, in which the bulb of the penis swells so that the mating pair cannot disengage for a period of up to a minute. Either dog may lick its genitals after a full copulation. Gestation lasts 70 days (Cade 1967; Dekker 1968; Reich 1981) and the interbirth interval is one year, perhaps slightly longer in some populations (Frame 1985). Litters average 8–11 pups but can be as large as 16 (Fuller et al. 1992a). Wild dog litters are heavy relative to the female’s body mass, and wild dog pups grow quickly, so that alpha females are energetically in need of help during gestation and lactation (Creel & Creel 1991). All adult pack members help to raise pups by feeding them, guarding them, and retrieving them when lost on hunts (Chapter 10; Malcolm & Marten 1982). Pack members also regurgitate food to the alpha female (Malcolm & Marten 1982), who produces a unique (and ear-splitting) begging call during gestation and lactation.
9.1 Are Dominants More Aggressive? During nonmating periods, the rate at which female wild dogs initiated aggression was not related to social dominance (Figure 9.3a: all pairwise comparisons ns). Surprisingly, rates of aggression tended to drop during mating periods (Figure 9.3a: t ⳱ 1.92, P ⳱ 0.058, df ⳱ 60), but did not drop equally for females of all ranks. Aggression dropped 36% for alpha females, 37% for beta females, and 65% for those of lower rank. As a consequence, dominant females were significantly more aggressive than subordinates during mating periods (alpha vs. all others: t ⳱ 1.73, one-tailed P ⳱ 0.046, df ⳱ 38). Rates of aggression were higher and more variable in males than in females. As with females, aggression and rank were not significantly associated during nonmating periods (alphas ⳱ 1.78 Ⳳ 0.88 acts/h; betas ⳱ 2.53 Ⳳ 1.46 acts/h; lower ranks ⳱ 0.98 Ⳳ 0.26 acts/h), and males did not become more aggressive during mating periods (Figure 9.3b: t ⳱ 0.75, P ⳱ 0.45. df ⳱ 119). Aggression declined very little for alpha males during mating periods (4% decrease), but decreased substantially for betas (40%) and lower ranks (21%). During mating periods, alpha and beta males were more aggressive than males of lower rank (alphas: t ⳱ 2.99, Bonferroniadjusted one-tailed P ⳱ 0.004, df ⳱ 68; betas: t ⳱ 1.81, Bonferroni-adjusted one-tailed P ⳱ 0.074, df ⳱ 64), but rates of aggression in alpha and beta males were similar (Figure 9.3b). During mating periods, low-ranking dogs of both sexes avoided high-ranking dogs in situations likely to provoke aggression. For males, a lower rate of aggression was accompanied by a striking increase in the severity of aggression. In a typical mating period, several males suffered bite wounds on the face and neck (Figure 9.4). In nonmating
206 ▪ C H A P T E R 9
Figure 9.3 Rates of aggression plotted against dominance rank during nonmating and mating periods for (a) females and (b) males. Note that ordinate scale differs in (a) and (b). Open bars ⳱ alpha; shaded bars ⳱ beta; black bars ⳱ dogs ranked third or below. Error bars show one standard error.
R E P R O D U C T I V E S U P P R E S S I O N ▪ 207
Figure 9.4 Wounds on the head and shoulders, typical for males during a mating period, when mildly escalated fights are common.
contexts, escalated fights were rare (e.g., we never observed injuries in contests over food). Of 10 fights in which a packmate deposed the alpha male, 6 (60%) occurred during mating periods, although mating takes up only 1–5% of the year. Three additional cases weren’t observed, but occurred during an interval that included a conceptive mating period. The single clear exception involved an alpha male who lost his rank after being badly injured by a wire snare. For females, lower rates of aggression during mating periods were not associated with an increase in the severity of fights. Females rarely fought over access to males during mating periods, and the alpha female was largely indifferent to interactions between the alpha male and other females. No alpha female was deposed in a fight related to mating.
9.2 Do Dominants Mate More Often or More Effectively? Mating Rate When a pack enters a mating period, alphas of both sexes mate at higher rates than subordinates (Figure 9.5), but subordinates (particularly betas) do mate occasionally. Dominance had a stronger affect on mating rates in females than in males. Alpha females mated at 10 times the rate of betas (0.75 Ⳳ 0.17 mounts/h vs. 0.07 Ⳳ 0.05 mounts/h: t ⳱ 3.74, Bonferroniadjusted P ⬍ 0.002, df ⳱ 16), and ⬎30 times the rate of females ranked third
208 ▪ C H A P T E R 9
Figure 9.5 Mating rates plotted against dominance rank for (a) females and (b) males. Open bars ⳱ alpha; shaded bars ⳱ beta; black bars ⳱ dogs ranked third or below. Error bars show one standard error.
or lower (0.002 Ⳳ 0.002 mounts/h: t ⳱ 6.56, Bonferroni-adjusted P ⬍ 0.002, df ⳱ 28). Alpha males mated 5 times more often than betas (0.54 Ⳳ 0.04 mounts/h vs. 0.11 Ⳳ 0.08 mounts/h: t ⳱ 2.77 Bonferroni-adjusted P ⳱ 0.04, df ⳱ 16) and 9 times more often than males ranked third or lower (0.06 Ⳳ 0.02 mounts/h: t ⳱ 2.78, Bonferroni-adjusted P ⳱ 0.006, df ⳱ 58). The distribution of matings has important implications for patterns of relatedness and population genetic structure. If rank had no effect, most matings would be by subordinates because they outnumber dominants (Figure 9.6). For females, the great majority of matings were actually made by alphas (Figure 9.6a: 2 ⳱ 398, P ⬍ 0.001). For males, the monopolization of mating was less complete (Figure 9.6b). Alpha males mated more often than expected by chance, beta males mated as often as expected, and lower-ranking males mated less often than expected ( 2 ⳱ 33.2, P ⬍ 0.001). Nonetheless, the majority of mounts were made by males other than the alpha (Figure 9.6b). This raises the question: “How effective are matings by males of different rank?” Mating Duration Wild dogs go through a copulatory lock in which the bulb of the penis enlarges and prevents the pair from disengaging (Goodall & Van Lawick 1971; Reich 1981), so ejaculatory matings are likely to last longer than nonejaculatory mounts. For females, dominance and the duration of matings
R E P R O D U C T I V E S U P P R E S S I O N ▪ 209
Figure 9.6 Comparison across ranks of observed numbers of mounts (stippled bars) and the number expected if dominance rank did not affect mating rates (open bars): (a) females, (b) males. Alpha females monopolized mounts more effectively than alpha males.
210 ▪ C H A P T E R 9
were not significantly related (alphas ⳱ 11.7 Ⳳ 2.9 seconds, lower ranks ⳱ 5.1 Ⳳ 0.7 seconds; t ⳱ 1.13, P ⳱ 0.23, df ⳱ 37). For males, mounts by betas (33.1 Ⳳ 13.6 s) were significantly longer than those of lower-ranking males (7.7 Ⳳ 1.3 s; t ⳱ 2.06, P ⳱ 0.04, df ⳱ 42), but were also significantly longer than mounts by alpha males (5.1 Ⳳ 0.8 s; t ⳱ 2.91, P ⳱ 0.01, df ⳱ 16). The duration of beta males’ mounts was also more variable (ⱖ 10-fold difference) than mounts by other males, because a subset of beta males’ mounts were unusually long. Rank of Mate Two males with equal mating rates and durations would still accrue different reproductive success if one had better access to dominant females, who are more likely to become pregnant and raise young. The ranks of mating pairs were positively correlated (t ⳱ 3.18, P ⳱ 0.002, df ⳱ 77), though the variance explained was small (r 2 ⳱ 0.05). The variance explained is probably low because only one female is likely to be at the peak of estrus on a given day. Subordinates that enter estrus do so slightly after the dominant, so malemale competition for estrous females is not greatly affected by the female’s rank.
9.3 Do Hormonal Differences Accompany Behavioral Differences? Males Dominant male wild dogs mate and fight more than subordinates. Because testosterone facilitates mating and aggression (Brain 1983; Wingfield & Moore 1987; Baum 1992; Monaghan & Glickman 1992), these behavioral differences might reflect variation in testosterone levels. In Selous, males’ testosterone levels did not change significantly between nonmating and mating periods (Figure 9.7: using an unpaired test, partial F1,108 ⳱ 3.08, P ⳱ 0.08, controlling rank and pack identity). Testosterone levels also did not decrease during denning (partial F1,122 ⳱ 0.46, P ⳱ 0.51, controlling rank and pack identity) as would be expected if paternal care were associated with low androgen levels (Wingfield et al. 1990). Dominance and testosterone were not significantly associated during nonmating periods (Figure 9.7: alpha males ⳱ 33.3 Ⳳ 6.9 ng T/mg feces; betas ⳱ 57.5 Ⳳ 16.4 ng T/mg feces; lower ranks ⳱ 30.9 Ⳳ 6.6 ng T mg feces). At times of mating, testosterone levels were unchanged in alpha males (Ⳮ4%: Figure 9.7) but decreased in beta males (ⳮ21%) and those of lower rank (ⳮ33%). Consequently, testosterone levels at mating were higher in alpha males (34.3 Ⳳ 8.0 ng T/mg feces) and beta males (45.3 Ⳳ 8.4 ng
R E P R O D U C T I V E S U P P R E S S I O N ▪ 211
Figure 9.7 Fecal testosterone concentration (ng T/mg dry feces) as a function of male dominance during (a) nonmating and (b) mating periods. Open bars ⳱ alpha; shaded bars ⳱ beta; black bars ⳱ dogs ranked third or below. Error bars show one standard error.
T/mg feces) than in males of lower ranks (20.8 Ⳳ 2.8 ng T/mg feces; partial F vs. alphas ⳱ 4.56, P ⳱ 0.048, df ⳱ 17; partial F vs. betas ⳱ 7.13, P ⳱ 0.01, df ⳱ 19). Testosterone levels in alphas and betas did not differ significantly during mating periods, but betas tended to have higher levels, as in nonmating periods (Figure 9.7). Overall, the relationship between dominance and testosterone parallels relationships between dominance and behavior. Low testosterone levels during mating periods are associated with low rates of aggression and mating in males ranked third or below, but alpha and beta males do not differ significantly in this regard. For wild dogs in Kruger National Park, rank and testosterone were associated after controlling for age (partial F ⳱ 2.78, P ⳱ 0.047, df ⳱ 1,15, onetailed test). Limited samples precluded testing mating and nonmating periods separately. It is not clear why age is an important variable in Kruger but not in Selous, but this may be related to differences in the two populations’ age structure. Males older than 4 are rare in Kruger (M.G.L. Mills, personal communication) but not in Selous (Chapter 7). Females Estrogen levels (Figure 9.7) were low during nonmating periods (0.070 Ⳳ 0.034 mg E/mg feces), increased at estrus (0.196 Ⳳ 0.056 mg E/mg feces) then declined to low levels during gestation (0.044 Ⳳ 0.148 mg
212 ▪ C H A P T E R 9
Figure 9.8 Concentrations of estrogens (open circles, with units of g per g dry feces) and progestins (filled circles, with units of g per mg dry feces) in female wild dog feces during nonmating (“baseline”), estrus, gestation, and lactation. Error bars show Ⳳ one standard error.
E/mg feces) and lactation (0.040 Ⳳ 0.128 mg E/mg feces). Mating occurs during and just after the estrogen peak, which causes the vulva to swell, sometimes with a trickle of bloody discharge. Progestin levels (Figure 9.8) were low during nonmating periods (12.5 Ⳳ 4.9 mg P/mg feces) and estrus (11.4 Ⳳ 7.8 mg P/mg feces), increased greatly during gestation (127.1 Ⳳ 20.7 mg P/mg feces) then returned to low levels during lactation (13.5 Ⳳ 18.0 mg P/mg feces). Estrogen levels were highly variable during gestation and lactation (Figure 9.8), so we restricted our analyses of rank effects to samples from baseline (nonmating, nonpregnant, nonlactating females) and estrous periods, testing whether dominance was associated with endocrine processes involved in becoming pregnant. However, by three criteria we observed no cases in which a subordinate female became pregnant then aborted in the second or third trimester. First, wild dogs produce exceptionally heavy litters (Creel & Creel 1991) and the belly of a pregnant female is visibly distended from the 20th day of the 70-day gestation. Second, wild dogs’ nipples swell during gestation. Third, the vulva becomes flaccid. First-trimester abortions would go undetected by these criteria, but other failures of pregnancy would be detectable (Figure 9.2). Estrogens are the primary endocrine stimulus of sexual behavior in females. During nonmating periods, the estrogen levels of subordinates were higher than those of alpha females (Figure 9.9a: partial F ⳱ 6.91, P ⳱ 0.012, df ⳱ 1,47). During estrus this relationship was reversed; alpha
R E P R O D U C T I V E S U P P R E S S I O N ▪ 213
Figure 9.9 (a) Fecal estrogen levels (g per mg dry feces) for dominance classes of female wild dogs during nonmating (“baseline”) and estrus periods. Open bars ⳱ alpha; black bars ⳱ subordinates. Error bars show one standard error. (b) Fecal estrogento-progestin ratios of female wild dogs during nonmating (“baseline”) and estrus periods, as a function of dominance. Open bars ⳱ alpha; black bars ⳱ subordinates.
214 ▪ C H A P T E R 9
females had significantly higher estrogen levels than subordinates (Figure 9.9a: partial F ⳱ 7.62, P ⳱ 0.014, df ⳱ 1,16). Progestins are also needed for the full expression of mating behavior (Turner & Bagnara 1976) and the ratio of estrogen to progestin alters hypothalamic-pituitary-gonadal feedback that regulates sex-steroid secretion and ovulation (Turner & Bagnara 1976). When not mating, alpha females and subordinates had similar progestin levels (t ⳱ 0.52, P ⳱ 0.60, df ⳱ 52), but at estrus, progestin levels were higher in alphas (t ⳱ 2.18, P ⳱ 0.04, df ⳱ 19). Treating rank as a continuum, estrogen-to-progestin ratios were significantly lower in high-ranking females during nonmating periods (Figure 9.9b: partial F ⳱ 4.36, P ⳱ 0.04, r 2 ⳱ 0.18, df ⳱ 1,47), but did not differ at estrus (Figure 9.9b: partial F ⳱ 0.69, ns).
9.4 Nonbreeder Lactation One subordinate female apparently lactated and nursed an older relative’s pups without a preceding pregnancy. This female was nulliparous and four years old at the time. She was subordinate to a female estimated to be six years old. Together with two other females, they dispersed and formed a new pack with three males from another group. The alpha female became pregnant three months after the pack formed. The subordinate that lactated did not appear pregnant on any of the five occasions that we observed her in the 55 days prior to the onset of her lactation. Pregnancy is normally very easy to observe in wild dogs (Figure 9.2), so we consider it unlikely that she was pregnant, but we have no endocrine data to confirm this. The subordinate female began lactating within 12 days of the litter’s birth. By the 40th day of lactation, the subordinate female’s udder was large and heavy, and she did the majority of the suckling (Figure 9.10). She was with the nine pups more often than their mother, and was more tolerant of suckling attempts than the mother. Her udder was visibly turgid prior to suckling, and became smaller and slack after suckling, just as in a lactating dominant female. Although non-nutritive suckling is common in some species, we never observed it in wild dogs, nor have other authors, so we believe that this was a nonbreeder lactation, as occurs after pseudopregnancy in other carnivores, including dwarf mongooses, meerkats, domestic dogs, and domestic cats (Smith & McDonald 1974; Verhage et al. 1976; Creel et al. 1991; S. Doolan, personal communication). In dwarf mongooses, nonbreeder lactation is fairly common among young nulliparous females, for whom it provides a substantial indirect fitness benefit by improving the survival of closely related offspring (Creel et al. 1991). If nonbreeder lactation is possible in wild dogs, one wonders why it is not seen in other social canids, because it would probably provide an indirect fitness benefit for them, too. Mechanistically, pseudopregnancy is the trigger for lactation in dwarf mongooses, and data from
R E P R O D U C T I V E S U P P R E S S I O N ▪ 215
Figure 9.10 Female wild dogs develop a pronounced udder during lactation. One female lactated without any overt sign of pregnancy, which is normally easy to detect.
captivity show that wild dogs undergo pseudopregnancy if they ovulate but do not conceive (Monfort et al. 1997). Alternative mechanisms exist, because lactation can be induced in nulliparous primates by suckling itself (humans: Cohen, 1971; spider monkeys: Estrada & Patterson 1979; rhesus monkeys: Holman & Goy 1980; ring-tailed lemurs: Pereira & Izard 1989). A few other species have been noted to lactate without pregnancy, including Indian elephants (Rapaport & Haight 1987), and male Dayak fruit bats (Francis et al. 1994). Although most reported cases have occurred in captivity, it is possible that cases in the wild go unreported, due to a lack of confidence about pregnancy diagnosis. Given the taxonomic breadth of species for which nonbreeder lactation has been reported, it may not be as rare as it currently seems.
9.5 Does Social Stress Mediate Reproductive Suppression of Subordinates? Social subordination is a well-known trigger of the glucocorticoid stress response (Christian & Davis 1964; Bronson & Eleftheriou 1964; Sapolsky 1985; Blanchard et al. 1995), and chronic glucocorticoid elevation can inhibit reproduction (Welsh & Johnson 1981; Wingfield 1988; Sapolsky 1992).
216 ▪ C H A P T E R 9
Figure 9.11 Fecal basal corticosterone levels (ng per g dry feces) as a function of rank in female and male wild dogs. Open bars ⳱ alpha; black bars ⳱ subordinates.
For Selous wild dogs of both sexes, basal corticosterone levels were higher in dominants than in subordinates (Figure 9.11: partial F ⳱ 6.76, P ⬍ 0.01, df ⳱ 1,207; controlling for sex and reproductive state), so subordinates’ reproduction is clearly not suppressed by chronic glucocorticoid elevation. To the contrary, dominants reproduce well despite their chronically elevated glucocorticoid levels (Creel et al. 1996).
9.6 How Effective Is Reproductive Suppression of Subordinates? In Selous, alpha females produced 22 litters in 27 pack-years (11 packs), giving an annual probability of breeding of 81.5% Ⳳ 7.0%. The five cases in which the alpha did not breed were newly formed packs that waited until the next rainy season to breed, rather than mating out of season. In stable packs, the alpha female invariably gave birth once a year. Subordinates’ annual probability of breeding was far lower (6.4% Ⳳ 2.4%; z ⳱ 7.36, P ⬍ 0.001). Only seven litters were born to subordinate females in 110 individual-years (in 5 of 27 pack-years). Although wild dogs are capable of breeding at one year of age, it is likely that many one-year-olds are not ready to reproduce (Cade 1967; Dekker 1968). Excluding one-year-olds, seven subordinates gave birth in 67 individual-years (10.4% Ⳳ 3.7%), still much lower than alpha females (z ⳱ 5.41, P ⬍ 0.001). Subordinate females in Kruger became pregnant at rates similar to (but slightly higher than) subordi-
R E P R O D U C T I V E S U P P R E S S I O N ▪ 217
nates in Selous (8.5% Ⳳ 2.1%; 17.4% Ⳳ 4.3% excluding one-year-olds, n ⳱ 44 pack-years).
9.7 Similarities and Differences between the Sexes in the Correlates of Rank The correlates of rank were broadly similar for male and female wild dogs. For dogs of both sexes, rates of aggression were not related to rank outside of mating periods, but dominant animals were more aggressive during mating periods. Dominant dogs’ greater aggression was paralleled by higher mating rates and better access to mates of high rank. For both sexes, behavioral differences were underpinned by differences in sex-steroid levels. In males, alphas and betas had higher testosterone levels than lower-ranking males, associated with higher rates of aggression and mating. Low mating rates in subordinate females were associated with high baseline estrogens and a high baseline estrogen-to-progestin ratio, which may prevent follicular development and ovulation by altering feedback to the hypothalamus and anterior pituitary (similar to the action of estrogen-based birth control pills in humans). Like other canids (Concannon et al. 1975), wild dogs become pseudopregnant if they ovulate but do not become pregnant (Monfort et al. 1997). Subordinate females in Selous did not normally show the elevated progestins of pseudopregnancy, suggesting that they did not ovulate. For females, endocrine and behavioral patterns yield highly skewed reproductive success. Alpha females produced most litters in Selous (76%) and in Kruger (81%), similar to the proportion reported for Serengeti (75%: Malcolm & Marten 1982). Genetic data will be needed to determine how many pups survive when a subordinate gives birth, but behavioral observations show that subordinates’ young are sometimes killed (Goodall & van Lawick 1971; Malcolm 1979; Reich 1981; Fuller et al. 1992a). Of seven litters by subordinate females in Selous, one litter was killed immediately, one litter was raised, and five litters were creched with the dominant’s litter so the fate of each mother’s pups was not directly observable. A comparison of the sizes of joint litters and single-female litters suggests that pups mothered by subordinates often survive (Chapter 10). Alpha males do not monopolize matings as effectively as alpha females (Figure 9.6), and the testosterone levels of alpha males are no higher than those of beta males (Figure 9.7). Matings by beta males are longer than those by alpha males. These patterns suggest that shared paternity is more common than shared maternity. In Kruger, this prediction was not supported by microsatellite analysis, which showed that 10% of 29 pups had subordinate fathers and 8% of 51 pups had subordinate mothers (Girman et al. 1997). Genetic data for Selous are discussed in the next chapter.
218 ▪ C H A P T E R 9
9.8 Interspecific Comparisons Relationships between dominance and sex steroids have been studied in several cooperatively breeding species, both in the wild (Reyer et al. 1986; Creel et al. 1991, 1992, 1993; Mays et al. 1991; Schoech et al. 1991; Schmidt et al. 1991; Vleck et al. 1991; Wingfield et al. 1991; Poini & Fletcher 1994) and in captivity (Abbott 1984, 1987; French et al. 1984, 1989; Faulkes et al. 1991). These studies suggest that sex steroids and social status are more closely associated in females than in males. Androgens are associated with rank in four of seven cooperative breeders studied in the wild. Four populations of cooperatively breeding birds showed an association between male rank and testosterone [white-browed sparrowweavers, Plocepasser mahali (Wingfield et al. 1991); Florida scrub jays, Aphelocoma coerulescens (Schoech et al. 1991); noisy bell-miners, Manorhina melanorhys (Poini & Fletcher 1994); and pied kingfishers Ceryle rudis (Reyer et al. 1986)]. No relationship between dominance and testosterone was found in two cooperatively breeding birds [Australian magpies, Gymnorhina tibicea (Schmidt et al. 1991), and Harris’s hawk, Parabuteo unicinctus (Mays et al. 1991)] and one mammal, the dwarf mongoose, Helogale parvula (Creel et al. 1993). For cooperatively breeding females, correlations between rank and reproductive physiology are the rule. During the period of reproduction, estrogen levels were higher in dominants than in subordinates for Florida scrub jays (Schoech et al. 1991), Harris’s hawks (Mays et al. 1991) and dwarf mongooses (Creel et al. 1992). As in wild dogs, associations between rank and estrogens in all of these species differed during breeding and nonbreeding periods. In white-browed sparrow-weavers, both dominants and subordinates had undetectable estradiol levels, but subordinates did not ovulate (Wingfield et al. 1991). In captive primates, female subordinates were hormonally suppressed in common marmosets (Callithrix jacchus: Abbott 1984) and cottontop tamarins (Sanguinus oedipus: French et al. 1984), but not in golden lion tamarins (Leontopithecus rosalia: French et al. 1989). Overall, interspecific comparisons suggest that males are hormonally suppressed less often than females. Weaker suppression among males is predicted by the hypothesis that subordinates resist suppression when the costs of reproduction are low (Creel & Creel 1991).
9.9 Dominance and Stress Circulating levels of sex steroids and glucocorticoids often show a negative correlation (Welsh & Johnson 1981; Moore et al. 1991), though not always (Wingfield & Silverin 1986; Hegner & Wingfield 1987). Based on this nega-
R E P R O D U C T I V E S U P P R E S S I O N ▪ 219 Table 9.1 Associations between social status and adrenal glucocorticoid levels in cooperatively breeding species. Subordinates do not have chronically elevated glucocorticoids in any study to date
Species
Conditions
African wild dog
Field
Dwarf mongoose
Field
Florida scrub-jay
Field
Alpine marmot
Field
Harris’s hawk
Field
White-browed sparrow weaver
Field
Common marmoset
Captive
Cotton-top tamarin
Captive
Association between Rank and Glucocorticoids F: Subordinates lower M: Subordinates lower F: Subordinates lower M: No association F: Subordinates lower M: No association F: No data M: Subordinates lower F: No association M: No association F: No association M: No association F: Subordinates lower M: No data F: Subordinates lower M: No data
Source 1, 2 1 3, 4 5 6 7 8, 9 10
Sources: 1, Creel et al. 1996; 2, Creel et al. 1997a; 3, Schoech et al. 1991; 4, Schoech et al. 1997; 5, Arnold & Dittami 1997; 6, Mays et al. 1991; 7, Wingfield et al. 1991; 8, Abbott et al. 1997; 9, Saltzman et al. 1994; 10, Zeigler et al. 1995.
tive relationship, it is often hypothesized that “low levels of reproductive hormones in subordinate individuals are . . . the result of stress-induced reproductive suppression” (Wingfield et al.; p. 44). This hypothesis is logical, and has support from studies of wild baboons (Sapolsky 1985; 1990; 1992), but no field studies of cooperative breeders support it, while four contradict it (Table 9.1). Basal glucocorticoid levels were not affected by dominance in Harris’ hawks (Mays et al. 1991), white-browed sparrow-weavers (Wingfield et al. 1991), male dwarf mongooses, and male Florida scrub jays. Subordinate females of the latter two species had significantly lower basal glucocorticoid levels than dominants (Schoech et al. 1991; Creel et al. 1996). In wild dogs of both sexes, subordinates had lower basal glucocorticoid levels than dominants. As in wild dogs, dominant male alpine marmots had higher androgen levels than subordinates, despite having higher corticosterone levels (Arnold & Dittami 1997). If chronic elevation of glucocorticoids is the proximate cause for reproductive suppression, one would expect a negative correlation between androgens and glucocorticoids. This correlation was positive for dominant males, negative for subordinate sons, and nonsignificant for subordinate non-kin (Arnold & Dittami 1997). The marmot data nicely
220 ▪ C H A P T E R 9
illustrate the more general conclusion that reproductive suppression in cooperative breeders is not often due to chronic elevation of glucocorticoids (Table 9.1). Rather than supporting the “chronic stress” hypothesis for reproductive suppression, studies of cooperative breeders raise a new issue: Dominant individuals maintain a functional hypothalmic-pituitary-gonadal axis despite chronically elevated glucocorticoids. If fighting itself is stressful, it is not surprising that glucocorticoids are elevated in dominant individuals, for species in which dominants are more aggressive than subordinates (Creel et al. 1996; Arnold & Dittami 1997), but it is surprising (from a mechanistic perspective) that chronic stress does not impair dominants’ reproduction (Moore et al. 1991). Even if losing is more stressful than winning, an individual that fights often and wins might have higher glucocorticoid levels than an individual that fights rarely and loses. It is critical to distinguish between aggressiveness and dominance when making predictions about the relationship between dominance and social stress. As we use the term, dominance is a characteristic of interactions between familiar individuals, which is manifested by the ability to win a contest without an escalated fight (Figure 9.1). When two unfamiliar individuals meet and simply fight until one wins, there is no need to bring the concept of social hierarchies into the picture: The fight occurs because there is no established relationship between the individuals. Based on current data, subordinates rarely suffer from chronic social stress in cooperatively breeding groups. This result differs strikingly from many studies of captive rats and mice, where subordinates have elevated basal glucocorticoid levels and enlarged adrenals (e.g., Bronson & Eleftheriou 1964; Blanchard et al. 1995). In some captive primates, subordinates have higher circulating glucocorticoid levels or larger adrenals than dominants, if the relationship is tested in a stable social group (e.g., squirrel monkeys: Manogue 1975). However, this relationship was not found in other studies, also conducted with stable social groups (e.g., squirrel monkeys: Steklis et al. 1985). Conflicting results have been reported within single species (see the squirrel monkey studies, cited above). In unstable primate groups, subordination is not associated with elevated glucocorticoid levels, and Sapolsky (1990; 1992) suggests that the psychological correlates of dominance differ in stable and unstable hierarchies. Hence, much of the variation among studies can be explained by variation in the stability of hierarchies. Sapolsky’s hypothesis clearly explains some of the variation among studies, but it does not fit most of the data from cooperatively breeding species. In the field studies described above, subordinates had glucocorticoid levels lower than or equal to those of dominants, without exception. This is striking because the field studies were generally conducted over several years, with several groups. Consequently, these studies probably incorporated short periods of instability within longer periods of stability. This is
R E P R O D U C T I V E S U P P R E S S I O N ▪ 221
true of the dwarf mongoose and wild dog studies, at least. Under natural conditions, Sapolsky’s hypothesis predicts either that subordinates will have higher glucocorticoid levels (if periods of stability predominate, which is likely for long periods), or that no relationship between rank and glucocorticoids will be found (if instability predominates, which is not likely for long periods). Four of six field studies of cooperative breeders contradict both predictions, and two match the latter prediction (Table 9.1). Studies of cooperatively breeding tamarins and marmosets in captivity also show that being subordinate is not chronically stressful, even in stable groups. French (1997) reviewed mechanisms of reproductive suppression in callitrichids and concluded that “in Callithrix and Sanguinus reproductive failure or delay in first ovulation is not produced by stress-induced pathology or by high levels of adrenal glucocorticoid” (p. 43). In female common marmosets, glucocorticoid levels are higher in dominants than in suppressed subordinates (Abbott et al. 1997; Saltzman et al. 1998). Overall, it now seems clear that chronic social stress is not a common cause of reproductive suppression in cooperatively breeding species (Creel et al. 1996). There may be phylogenetic differences in the effect of social instability on the correlation between dominance and glucocorticoid levels. In whitethroated sparrows, corticosteroid levels are high in free-living subordinates, but these differences disappear in stable captive hierarchies (Schwabl 1995). This differs from the pattern suggested for primates by Sapolsky (1992). More data are needed before phylogenetic patterns can be assessed properly, but phylogeny is a possible explanation for some of the variability in the studies described here. Captive studies often assess dominance by pairing unfamiliar individuals and observing the fight that ensues (see discussion in Zielinski & Vandenbergh 1993). The correlates of winning in this situation may bear little resemblance to the correlates of dominance in the wild (Creel et al. 1996). In a stable group of wild dogs, most disputes are settled by stylized displays or mild aggression, not by all-out fighting. “Paired contests” may be a good model for endocrine events that occur during immigration or the formation of new groups in the wild (Alberts et al. 1992; Creel et al. 1993). At these times, dominance relationships have not been established and fights may be severe. Even in more natural captive studies, subordinates have limited ability to avoid aggression, and cannot disperse. Given the patterns we have just reviewed, we suggest that differences among species and in the stability of hierarchies explain some of variation in studies of social stress, but “semifree-ranging” conditions will be necessary to reveal the endocrine patterns that operate in the wild for many species. Chronically elevated glucocorticoids cause a broad array of harmful effects, including immune suppression, inefficient energy metabolism, and neural death (Munck et al. 1984; Sapolsky 1992). Thus, in societies where reproductive suppression is normal, it is not surprising to find non–
222 ▪ C H A P T E R 9
glucocorticoid-mediated mechanisms. Costs not associated with reproduction would make glucocorticoid-mediated suppression evolutionarily invasible by mechanisms independent of glucocorticoids (perhaps mediated by the effects of opioids on gonadotropins: Abbott 1993).
9.10 Do the Correlates of Rank Relate to Dispersal and Social Organization? In evolutionary models of reproductive skew, shared parentage is used by dominants as an incentive to prevent subordinate helpers from dispersing (Chapter 10; Vehrencamp 1983; Creel & Waser 1991; Reeve & Ratnieks 1993). If females are more completely suppressed, theory suggests they should be more likely to disperse. In Serengeti, female wild dogs dispersed more often than males, an unusual pattern in mammals (Frame et al. 1979; Macdonald & Moehlman 1982; Chepko-Sade & Halpin 1987; Waser 1996). In Selous also, females dispersed more often than males (Chapter 8). Selous males were more “patient” with respect to dispersal, perhaps because dominant males are more likely than dominant females to share reproduction. In Kruger, genetic data suggested that the sexes were equally suppressed (Girman et al. 1997), and annual probabilities of dispersal were similar for females (19.2% Ⳳ 2.9%) and males (20.7% Ⳳ 2.8%: z ⳱ 0.21, P ⳱ 0.84). In summary, endocrine differences between dominants and subordinates suggest that natural selection has altered wild dogs’ physiology to reinforce behavioral mechanisms of reproductive suppression. Such behavioral and endocrine correlates of social status vary among species and populations, and these mechanistic differences can be related to basic features of social organization. For wild dogs, female-biased dispersal and a male-biased sex ratio may relate to differences between the sexes in the behavioral and endocrine correlates of rank.
10
Patterns of Relatedness and the Fitness Consequences of Dispersal, Philopatry, and Reproductive Suppression
The evolution of cooperative breeding has been an enduring subject in behavioral ecology, because it presents complicated and interesting quandaries about the evolution of social behavior. Most animals begin their breeding attempts at sexual maturity, usually just after dispersal from the site of birth, and continue until senescence or death. In cooperative breeders, many individuals forgo reproduction for the majority or entirety of their adult lives. For example, more than 65% of dwarf mongooses die without making a breeding attempt, despite relatively low juvenile mortality. This is an interesting situation. If natural selection favors traits that increase the number of descendants an individual leaves behind, what mechanisms create social systems in which the majority of adults do not reproduce? There are several possibilities. First, subordinates may delay reproduction simply to increase residual reproductive value (Lack 1954; Williams 1966), but this explanation is unlikely to hold for most cooperative breeders, because subordinates will normally breed immediately upon attaining dominance, regardless of their age. In wild dogs, as in many cooperative breeders (Hatchwell & Komdeur 2000), reproductive suppression seems to be the product of competition with same-sexed individuals for limited breeding opportunities, although inbreeding avoidance also affects decisions to disperse or remain on the natal territory, particularly in males (Chapter 7). The role of inbreeding avoidance in reproductive suppression is variable among cooperatively breeding carnivores, with important effects in some species (such as the meerkat: O’Riain et al. 2000), and no detectable effect in other species (such as the dwarf mongoose: Keane et al. 1997). When breeding opportunities are difficult to obtain, as in many cooperative breeders, then inbreeding depression must be substantial if inbreeding avoidance is to be favored by selection. Qualitatively, the sequence of events following the death of a breeder can reveal something about the relative strength of intrasexual competition and inbreed-
224 ▪ C H A P T E R 1 0
ing avoidance as evolutionary forces producing reproductive suppression. In dwarf mongooses the sequence of events that follows the death of a breeder is similar to that of wild dogs, and suggests that subordinates are suppressed primarily by the presence of a same-sexed dominant, rather than self-inhibited by relatedness to the opposite-sexed dominant. When an alpha dies, it is almost invariably the same-sexed beta that assumes the breeding role. The newly dominant individual begins breeding immediately, if possible, or in the next breeding season if it is too late to attempt a litter. The fact that the second individual in the hierarchy assumes the breeding position strongly suggests that intrasexual competition is the driving force behind failure to breed. If patterns of relatedness among potential mates were the driving force, then individuals lower in the hierarchy would often be predicted to inherit the breeding role. Also, when a breeder dies, it does not normally provoke changes in the dominance hierarchy of the opposite sex, which again suggests that competition is more important than inbreeding avoidance (mate choice) in controlling who breeds and who does not. In general, there is little evidence of active mate choice in cooperatively breeding carnivores. Patterns of survival and dispersal may make certain mating patterns uncommon (i.e., close inbreeding may be rare), even in the absence of active mate choice. In cases where competition underlies subordinates’ failure to breed, the distinction between externally imposed reproductive suppression and selfimposed reproductive inhibition is often a matter of semantics. Some authors (e.g., Schoech et al. 1996) propose that an endocrine mechanism of reproductive failure among subordinates implies externally imposed suppression, while the absence of endocrine mechanisms implies that the “absence of a stimulus” prevents subordinates from breeding, not the “presence of an inhibitor.” We disagree. For example, Schoech et al. (1996) argued that male Florida scrub jays were endocrinologically primed for breeding, but that a lack of behavioral interactions with sexually receptive females kept testosterone levels low. In their words, “Determining whether a seemingly low physiological measure (e.g., hormone level, gonad size) is a response to a negative stimulus rather than the baseline reading of a parameter that has yet to be stimulated to a higher level can be difficult” (p. 88). We agree with this mechanistic point, but not with the evolutionary conclusion they draw, that “breeding suppression or inhibition should not be invoked to explain delayed breeding in cooperatively breeding species” in this situation (p. 88). Clearly, one way that a dominant individual can influence a subordinate is by withholding access to critical resources (this is almost a definition of dominance), including access to sexually receptive members of the opposite sex. This creates an “absence of stimulus,” in the terminology of Schoech et al., but it is still a mechanism of reproductive suppression that arises because of the presence of the dominant individual. If a mechanism that causes reproductive failure would cease to function if the dominant were absent, it is
PAT T E R N S O F R E L AT E D N E S S ▪ 225
fundamentally a form of reproductive suppression driven by intrasexual competition. The proximate mechanisms underlying reproductive failure are interesting in their own right, and understanding proximate mechanisms can generally shed light on some evolutionary questions. However, mechanistic data do not clearly distinguish between suppression and self-inhibition in this case. For example, subordinate male dwarf mongooses have androgen profiles indistinguishable from those of dominant males, while subordinate females have significantly lower estrogen levels than dominant females, particularly during mating periods (Creel et al. 1992). Although the proximate mechanisms of reproductive failure differ substantially between the sexes (with important implications for the effectiveness of reproductive suppression), the selection pressures maintaining reproductive suppression are very similar (Creel & Waser 1991, 1994; Creel et al. 1992; Keane et al. 1997). Subordinates may tolerate reproductive suppression in order to remain on their natal territory, avoiding the costs of dispersal while waiting to “inherit” a breeding position by moving up the dominance hierarchy (Emlen 1982; Vehrencamp 1983). Dispersal can entail a substantial risk of death, particularly in species that saturate their habitat so that all suitable territories are occupied (Chapter 8; Waser et al. 1994; Waser 1996). Habitat saturation is common in cooperatively breeding vertebrates (Stacey & Koenig 1990; Hatchwell & Komdeur 2000), but apparently not common in eusocial insects (Brockman 1997). Even if dispersal itself is not costly, the likelihood of breeding independently may be low, and reproduction can be energetically costly, both of which can constrain dispersal in unsaturated habitats (Creel & Creel 1991; Brockman 1997). Finally, philopatric individuals may accrue indirect fitness benefits by helping to increase the number of offspring raised by relatives (Hamilton 1964), as has been shown in several species of birds (Emlen 1995), insects (Bourke & Franks 1995; Choe & Crespi 1997), and mammals (Creel & Waser 1994; Jennions & Macdonald 1994). In some species, cooperative breeding has become obligate, so that parents cannot produce and raise offspring without help from nonbreeders. Cooperative breeding is clearly obligate in the eusocial insects, where castes have evolved and many individuals are permanently sterile. Among vertebrates, there are only a few clear cases of obligately cooperative breeding, including stripe-backed wrens (Rabenold 1990), naked mole rats (Sherman et al. 1991), dwarf mongooses (Creel & Creel 1991), meerkats (CluttonBrock et al. 2001), and wild dogs. To our knowledge, no pair of wild dogs without helpers has been seen to raise pups to independence in the wild. Even if pairs did occasionally succeed, data make it clear that the probability of raising offspring to independence without help is close to zero in all wellstudied wild dog populations (Chapter 7; Maddock & Mills 1994; Malcolm 1979; Burrows 1995). There seem to be three reasons that unaided pairs and small packs have difficulty raising pups. First, foraging success improves as
226 ▪ C H A P T E R 1 0
pack size increases, at least initially (Chapter 4; Creel 2001). Second, because small packs require all of their members for hunting, they are more likely to leave their pups unattended during hunts, increasing their vulnerability to predation. Finally, wild dogs produce unusually large litters, and the ratio of litter mass to female mass is the highest recorded among carnivores (Creel & Creel 1991). The growth rate of the litter is also unusually fast, and the energetic burden of providing enough food to offset these costs is large (Gorman et al. 1998), and these costs can be shared among more individuals in large packs. Because wild dogs are obligately cooperative, some authors classify them as effectively eusocial (Sherman et al. 1995). This classification is debatable because the reproductive status of wild dogs can change during their adult lives, so that life-history decisions are not irreversible, as in eusocial insects (Crespi & Yanega 1995). Nonetheless, wild dogs fall near the extreme end of the eusociality continuum, relative to other vertebrates (Sherman et al. 1995). In this chapter, we use data on demography, dispersal, and relatedness (Chapters 7 and 8) to compare the direct and indirect fitness consequences of dispersal and nondispersal. In this context, indirect fitness refers to fitness accrued by increasing the production or survival of nondescendant kin, while direct fitness simply refers to reproduction (Brown 1987). We also use inclusive fitness calculations to examine patterns of reproductive suppression, testing whether cases in which subordinate females breed match predictions from a simple model of reproductive skew (Vehrencamp 1983; Creel & Waser 1991; Keller & Reeve 1994).
10.1 Age-specific Relatedness of Natal and Immigrant Subordinates to Breeders As a subordinate ages, it is likely to become less closely related to the dominant dogs in its pack, because it parents are increasingly likely to have died, been displaced from the pack, or, in the case of males, lost rank. Thus the indirect fitness benefits of philopatry and helping will tend to decline with age, but to what extent? Tables 10.1 and 10.2 show our calculations of agespecific relatedness between natal subordinates and the pack’s breeders. We do not assume that subordinates are unrelated to breeders following dispersal—those data are discussed in a moment (Table 10.3, also see Chapter 8). The predominant factor affecting relatedness of natal subordinates to breeders is the survival of the subordinate’s parents (Table 10.1). Although replacement breeders are often related to subordinates, the relationship tends to be distant, even when the new breeder comes from within the pack (Table 10.2). Consequently, the age-related decline in total relatedness of natal sub-
0.71 0.50 0.36 0.25 0.18 0.13 0.09 0.06
0.35 0.25 0.18 0.13 0.09 0.06 0.05 0.03
Average Relatedness via One parent B 0.29 0.50 0.64 0.75 0.82 0.87 0.91 0.94
Cumulative Probability that a Parent Has Died 0.029 0.029 0.029 0.029 0.029 0.029 0.029 0.029
Relatedness to One Replacement C 0.01 0.01 0.02 0.02 0.02 0.03 0.03 0.03
Average Relatedness via One Replacement D 0.36 0.27 0.20 0.15 0.11 0.09 0.07 0.06
Average Relatedness, via One Parent or Replacement E 0.73 0.53 0.39 0.30 0.23 0.18 0.14 0.12
Relatedness X 2F
For each sex, the annual probability of survival for an adult was raised to the power of the subordinate’s age. Survival probabilities for male and female parents were then averaged. Because survival of alphas did not differ significantly from that of subordinates, we used the entire sample of adults to calculate survival rates (N ⳱ 249 for males, 201 for females); see Table 7.2. B Mean relatedness to parents was assumed to be 0.5, as inbreeding is apparently not common (Chapter 8). Average relatedness via one parent ⳱ 0.5 ⳯ the age-specific probability that a parent survives. C See Table 10.2 for calculation. D Average relatedness via one replacement ⳱ the age-specific probability that a parent dies ⳯ mean relatedness to replacement. E Equal to the sum of average relatedness via one parent and average relatedness by one replacement. F Relatedness is multiplied by 2 to convert indirect fitness components to “offspring equivalents” (Brown 1987) for comparison to direct fitness components.
A
1 2 3 4 5 6 7 8
Age of Subordinate
Cumulative Probability that a Parent Survives A
Table 10.1 Calculation of age-specific relatedness of natal subordinates to dominant breeders
228 ▪ C H A P T E R 1 0 Table 10.2 Calculation of relatedness of natal subordinates to replacement breeders when a parent dies Average
Average
Relatedness via
Relatedness via
Proportion of
Proportion of
New Alphas
New Alphas
Mean
Mean
from Within
from
Relatedness
Relatedness
within Pack
Replacement
Pack
Outside pack
Within Pack
to Non-Pack
Replacement
from Outside
Males
0.43
0.56
0.11
0.03
0.0473
0.0168
Females
0.27
0.73
0.11
0.03
0.0297
0.0219 䉲
Average for both sexes and both types of replacement, accounting for frequency of each type ⳱ 0.029
ordinates to breeders closely parallels the decline in the probability that a subordinate’s parents will still be alive (Figure 7.1 and Table 10.1). From microsatellite data, our estimate of mean relatedness between natal subordinates and breeding replacements from within the pack is 0.11. Our estimate of mean relatedness to breeding replacements from outside the pack is substantially lower, 0.03. In an interesting asymmetry, this value of r is substantially lower than the relatedness between alphas and subordinates that have dispersed but failed to attain dominance (Table 10.3 and Chapter 8). This genetic asymmetry is expected, given patterns of dispersal by wild dogs. Natal subordinates are rarely joined by immigrants of the same sex. Though we did observe a few cases in which two lineages of the same sex fused (Chapter 8), same-sex fusion is rare enough that it has little effect on patterns of relatedness. It is common for natal subordinates to be joined by immigrants of the opposite sex, but these immigrants are distantly related or unrelated in most cases. These patterns of dispersal predict a low degree of relatedness between natal subordinates and replacement breeders from outside the pack, and genetic data confirm this prediction (Tables 10.1 and 10.2; Chapter 8). Dogs that remain subordinate after dispersing are in a different situation. Most dispersal occurs in groups. These groups are generally composed of littermates or litters of two consecutive years, which will also be close relatives in most cases, though not always. When a dog disperses but remains subordinate, another member of its dispersing unit is very likely to be the new breeder (because fusions of single-sex lineages are rare, but displacements of single-sex lineages by immigrants are common; Chapter 8). These patterns of dispersal predict that dogs that immigrate into subordinate roles should be relatively closely related to the pack’s breeder of the same sex,
0.58 0.58 0.58 0.58 0.58 0.58 0.58 0.58
0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
Males 0 1 2 3 4 5 6 7
Initial Relatedness to Same-sex Alpha
Females 0 1 2 3 4 5 6 7
Years Since Dispersal
1 0.72 0.52 0.37 0.27 0.19 0.14 0.10
1 0.69 0.48 0.33 0.23 0.16 0.11 0.07
Probability that Initial Alpha Remains Alive
0.10 0.07 0.05 0.04 0.03 0.02 0.01 0.01
0.58 0.40 0.28 0.19 0.13 0.09 0.06 0.04
Relatedness via Initial Alpha
0.00 0.28 0.48 0.63 0.73 0.81 0.86 0.90
0.00 0.31 0.52 0.67 0.77 0.84 0.89 0.93
Probability That Initial Alpha dies
0.029 0.029 0.029 0.029 0.029 0.029 0.029 0.029
0.029 0.029 0.029 0.029 0.029 0.029 0.029 0.029
Mean Relatedness to Replacement Breeder
0.00 0.01 0.01 0.02 0.02 0.02 0.02 0.03
0.00 0.01 0.02 0.02 0.02 0.02 0.03 0.03
Relatedness via Replacement Breeder
0.10 0.08 0.07 0.06 0.05 0.04 0.04 0.04
0.58 0.41 0.29 0.21 0.15 0.12 0.09 0.07
Total Relatedness
Table 10.3 Calculations of relatedness of immigrant subordinate to same-sexed alphas, as a function of the number of years since its immigration
230 ▪ C H A P T E R 1 0
Figure 10.1 Mean relatedness of natal subordinates to the breeders of their pack, as a function of the subordinate’s age.
and distantly related to the breeder of the opposite sex. Microsatellite data confirm these patterns. Upon entering their new pack, immigrant subordinates of both sexes are unrelated to breeders of the opposite sex, but related to breeders of their own sex (see below). In our sample for microsatellite analysis, immigrant subordinate females were particularly closely related to breeders of their own sex (r ⳱ 0.58 at immigration). Dispersing males were apparently less closely related upon immigration (r ⳱ 0.10 upon immigration) but the apparently large difference between the sexes may be an artifact of having relatively few individuals for genetic analysis (64 individuals from 10 packs). These relatedness values are for immigrant subordinates to same-sexed breeders, immediately upon immigration. It is important to recall that dispersal reduces relatedness to opposite-sexed breeders to zero, for both sexes (Chapter 8). Moreover, relatedness declines for immigrant subordinates as the period since immigration lengthens, due to breeder turnover (Figure 10.3 and Table 10.3), and this complicates the calculation of age-specific relatedness for immigrant subordinates. For immigrant subordinates, the decline in relatedness to same-sexed breeders is a function of time since immigration, and not a direct function of age. By combining the decline in r due to years since immigration with the frequency distribution of dispersal ages for each sex, we calculated expected r values between immigrant subordinates and same-sexed breeders as a function of age (Figure 10.4). Our data suggest that dispersal reduces relatedness to same-sexed breeders substantially for males, but has little effect for females. When we discuss inclusive fitness
PAT T E R N S O F R E L AT E D N E S S ▪ 231
Figure 10.2 The proportion of new breeders that come from within a pack or immigrate from outside the pack, following the death, eviction, or loss of rank of an alpha.
below, it should be borne in mind that these mean values do not give the relatedness of subordinates that have just immigrated; they give the expected relatedness for subordinate immigrants of a given age, including those that immigrated in previous years. From these data, it is clear that subordinates can accrue indirect fitness following dispersal. Weighted by the frequency of dispersal by each sex, mean relatedness to breeders for immigrant subordinates is r ⳱ 0.21 in the year of immigration (Figure 10.3 and Table 10.3). This is higher than the relatedness of natal subordinates to breeders at all ages beyond one year (Figure 10.1). These data highlight the fact that wild dog groups are not simply a breeding pair and their reproductively suppressed offspring (cf. Mech, 1999). Dispersal and breeder turnover can cause groups to develop social organization far more complex than a nuclear family, with ramifications for the indirect fitness of both natal and immigrant subordinates. By dispersing in groups, immigrants maintain relatedness to packmates of the same sex, and hence retain some potential to accrue indirect fitness. Simultaneously, dispersers reduce their relatedness to packmates of the opposite sex, which reduces the potential for inbreeding, but also reduces the potential to accrue indirect fitness.
10.2 Inclusive Fitness of Nondispersers Direct Fitness There are two ways for nondispersers to obtain direct fitness. First, they can inherit the breeding position if the dominant individual of their sex dies or is deposed. Second, they can attempt to breed even though they remain
232 ▪ C H A P T E R 1 0
Figure 10.3 Mean relatedness of immigrant subordinates to breeders in the pack they join, as a function of years since immigration. Note that the ordinate scales differ. (a) Females; (b) Males.
socially subordinate. For now, we will focus only on inheriting dominance. Dogs of both sexes do occasionally reproduce as subordinates, but we examine those data below (see “Incomplete Reproductive Suppression: Breeding by Subordinates”). Natal subordinate females only inherited the alpha position if the dominant female died. Dominance is strongly related to age (Chapters 5 and 7), and the age-specific probability of becoming dominant upon the death of the alpha female rises steadily from zero for yearlings to 0.67 for eight-year-olds (Table 10.4). The annual probability that the alpha female will die was 0.31. The product of these two probabilities gives the age-specific probability of inheriting the alpha female position, which is essentially zero for young females (less than 1% for ages three and younger), and rises to 20.7% for
PAT T E R N S O F R E L AT E D N E S S ▪ 233
Figure 10.4 The frequency distribution of dispersers that immigrate at each age, and expected age-specific relatedness to breeders, determined from this frequency distribution and the relationships between relatedness and years since immigration shown in Figure 10.3. (a) Females; (b) Males.
eight-year-olds. Multiplying the probability of inheriting by the mean annual reproductive success of a breeder yields the age-specific direct fitness of natal subordinate females that tolerate reproductive suppression (Table 10.4), which ranges from 0 to 0.96 offspring raised per year, increasing steadily with age (Figure 10.5). For males, the probability of inheriting the alpha position does not increase monotonically with age, as it does for females. As discussed in Chapter 7, old alpha males are occasionally deposed by prime-aged packmates. Because alpha males may be deposed (5 cases in 30-pack years) the chances that a turnover in the breeding male’s identity will occur is large in comparison to females. Alpha males are no more likely to die than alpha females (annual mortality ⳱ 0.28 and 0.31, respectively), but the alpha male turned over in 44% of packs annually. Upon the death or deposition of an alpha
0.73 0.53 0.39 0.30 0.23 0.18 0.14 0.12
Age
1 2 3 4 5 6 7 8
0.57 0.57 0.57 0.57 0.57 0.57 0.57 0.57
Effect of One Helper on RS B
0.41 0.30 0.23 0.17 0.13 0.10 0.08 0.07
Indirect Fitness
0.00 0.00 0.03 0.19 0.21 0.42 0.37 0.27
0.00 0.00 0.03 0.19 0.33 0.36 0.54 0.67
0.44 0.44 0.44 0.44 0.44 0.44 0.44 0.44
0.31 0.31 0.31 0.31 0.31 0.31 0.31 0.31
Females
Males
Males
Females
Annual Probability That Alpha Dies or is Deposed
Probability of Becoming Alpha if Alpha Dies
4.70 4.70 4.70 4.70 4.70 4.70 4.70 4.70
Mean Reproductive Success C
0.000 0.000 0.060 0.383 0.440 0.875 0.761 0.552
Males 0.000 0.000 0.047 0.269 0.477 0.512 0.771 0.956
Females
Direct Fitness
0.414 0.303 0.285 0.552 0.570 0.977 0.843 0.620
Males
B
0.414 0.303 0.272 0.439 0.607 0.614 0.853 1.024
Females
Inclusive Fitness
See Tables 10.1 and 10.2 for calculation. From regression of annual change in reproductive success on annual change in pack size, measured in units of offspring raised to one year. See Chapter 7. C Measured in units of offspring raised to one year, here and in Table 10.5.
A
Relatedness X 2A
Table 10.4 Calculation of age-specific inclusive fitness for nondispersers
PAT T E R N S O F R E L AT E D N E S S ▪ 235
Figure 10.5 Direct fitness accrued by subordinates via inheriting the alpha position in their natal pack, as a function of age. Differences between the sexes are driven by differences in the relationship between age and dominance.
male, the probability of inheriting the alpha position is highest for six-yearolds (42%), and subsequently declines. Overall, the age-specific probability of inheriting a breeding position shows a broad pattern similar to that of females (Table 10.4), peaking at 18.5% for six-year-old males. Multiplying the probability of inheriting by mean annual reproductive success yields mean age-specific direct fitness for natal subordinate males that tolerate reproductive suppression (Table 10.4), which ranges from 0 to 0.88 offspring raised per year, increasing steadily to age six and then declining (Figure 10.5). Indirect Fitness As discussed above, most natal subordinates are closely related to one or both of the breeders in their pack (Figure 10.1), and the reproductive success of a breeding pair is strongly influenced by the number of helpers in its pack (Chapter 7). Through cooperative hunting, pack size has a strong effect on foraging success (Chapters 4 and 6; Creel 2001). Subordinates of both sexes also guard the pups and regurgitate food to them while they are too young to follow the pack on hunts (Figure 10.6; Kuhme 1965; Malcolm & Marten 1982). Through these behavioral mechanisms, reproductive success increases as pack size increases, up to about 15 adults (Chapter 7). However, a significant regression of reproductive success on pack size does not guarantee a causal relationship. In the absence of other information, it is plausible that pack size and reproductive success are both affected by territory quality but have no direct causal relationship (Brown 1987). Experimental manipulations of pack size are needed to resolve this question beyond all doubt, as is true of any correlation. In the absence of manipulations of pack size, we
Figure 10.6 (A) A group of adults watches the surrounding long grass while allowing the pups to eat first. (B) A small pack cooperatively mobs a male lion. (C) Maintaining a safe distance, a wild dog taunts a male lion.
PAT T E R N S O F R E L AT E D N E S S ▪ 237
Figure 10.7 The relative contributions of direct and indirect fitness to the inclusive fitness of nondispersers, as a function of age.
used a natural experiment, regressing changes in pack size on changes in annual reproductive success, which showed that increasing (decreasing) pack size by one adult yielded an increase (decrease) of 0.57 pups raised to one year. From the standpoint of experimental design, one would like to systematically or randomly assign manipulations of pack size, to avoid the possibility that long-term changes in territory quality affect group size in one year and pups raised over the following year. We consider this unlikely for wild dogs, because most changes in pack size are abrupt shifts due to the immigration or emigration of single-sex groups, and dispersal events seem more closely related to demography than to environmental variables (McNutt 1995; Chapter 8). Moreover, the fact that no unaided pairs have been recorded to raise young makes it unlikely that the pack size effect is a statistical artifact. To calculate indirect fitness, we multiplied age-specific relatedness by the contribution of one helper to reproductive success. We multiplied relatedness by 2, to convert indirect fitness to units of offspring equivalents (Grafen 1982). These calculations (Table 10.4) show that indirect fitness is substantial for young dogs (0.41 offspring equivalents for one-year-olds), but drops to low levels for older dogs (0.07 offspring equivalents by eight years). For nondispersers of both sexes, indirect fitness is the dominant component of inclusive fitness from ages one to four, after which direct fitness is dominant (Figure 10.7). Thus, the benefits of nondispersal are primarily via helping for young subordinates and primarily via inheritance for old subordinates (as in dwarf mongooses: Creel & Waser 1994).
238 ▪ C H A P T E R 1 0
10.3 Inclusive Fitness of Dispersers Direct Fitness The direct fitness of dispersers depends on the probability of surviving to immigrate into a new group, the probability of becoming a breeder, and the reproductive success obtained as a breeder (Table 10.5). Males are slightly more likely to survive dispersal than females (Chapter 8). For females, the probability of becoming a breeder in the new pack increases steadily with age, while for males, the probability of becoming a breeder is highest in middle age (Figure 10.8). Because of these patterns, annual direct fitness increases steadily with age for female dispersers, to a substantial 3.3 offspring at ages six and older (Table 10.5). This is 70% of the mean reproductive success of established alpha females, suggesting that old subordinate females should be unlikely to accept reproductive suppression (see below). For males, expected direct fitness is substantially lower at all ages except five (Table 10.5). Indirect Fitness As for nondispersers, we calculated age-specific indirect fitness as the product of mean relatedness to the breeder, multiplied by the effect of one helper on the breeders’ reproductive success (0.57 offspring raised/year; see above). We measured relatedness using microsatellite data, which showed that dispersers that immigrate as subordinates are often close relatives of the samesexed alpha in their new pack, through co-immigration (see “Age-specific Relatedness of Natal and Immigrant Subordinates to Breeders,” above). Microsatellites also confirm that relatedness to opposite-sexed breeders is zero following dispersal, as in other populations (Girman et al. 1997). On average, dispersers are not as closely related to the breeders in their pack as nondispersers are, so they accrue less indirect fitness by helping. Recall that relatedness declines as a function of years since immigration, not strictly as a function of age (see above). In Table 10.5, we calculated agespecific relatedness using the empirical distribution of dispersal ages to determine the likelihood that an immigrant of a given age would have dispersed in that year or in previous years. This method gives the expected degree of relatedness for all immigrants of a given age, regardless of when they dispersed. These figures are of interest for comparisons of direct and indirect fitness. However, to examine age-specific dispersal decisions, it is more realistic to compare the inclusive fitness of a nondisperser with the inclusive fitness of an individual who disperses at that age. For this comparison, the decline in relatedness after immigration is not relevant, and the relatedness of subordinate immigrants to alphas holds constant across ages at 0.58 for females and 0.10 for males (rather than declining as shown in Table
Probability of Surviving Dispersal
0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71
0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8
Age
FEMALES 1 2 3 4 5 6 7 8
MALES 1 2 3 4 5 6 7 8 0.00 0.00 0.33 0.50 1.00 0.50 0.33 0.00
0.00 0.00 0.50 0.50 0.67 1.00 1.00 1.00
Probability of Becoming Alpha If Survive Dispersal
4.7 4.7 4.7 4.7 4.7 4.7 4.7 4.7
4.7 4.7 4.7 4.7 4.7 4.7 4.7 4.7
RS of Alpha
0.000 0.000 1.241 1.880 3.760 1.880 1.241 0.000
0.000 0.000 1.669 1.669 2.236 3.337 3.337 3.337
Direct Fitness
0.10 0.08 0.07 0.06 0.05 0.04 0.04 0.04
0.58 0.41 0.29 0.21 0.15 0.12 0.09 0.07
Relatedness to Same Sex Alpha (Table 10.5)
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
Relatedness to Opposite Sex Alpha
0.10 0.10 0.10 0.08 0.07 0.06 0.05 0.05
0.58 0.58 0.50 0.43 0.35 0.28 0.21 0.16
Mean Relatedness X2
Table 10.5 Calculation of age-specific inclusive fitness for dispersers
0.57 0.57 0.57 0.57 0.57 0.57 0.57 0.57
0.57 0.57 0.57 0.57 0.57 0.57 0.57 0.57
Effect of One Helper on RS
1.00 1.00 0.67 0.50 0.00 0.50 0.67 1.00
1.00 1.00 0.50 0.50 0.33 0.00 0.00 0.00
Probability of Staying Subordinate If Survive Dispersal
0.057 0.057 0.036 0.023 0.000 0.017 0.020 0.027
0.331 0.331 0.143 0.121 0.066 0.000 0.000 0.000
Indirect Fitness
0.057 0.057 1.277 1.903 3.760 1.897 1.261 0.027
0.331 0.331 1.811 1.790 2.302 3.337 3.337 3.337
Inclusive Fitness
240 ▪ C H A P T E R 1 0
Figure 10.8 The likelihood of becoming a dominant breeder for wild dogs of each sex that have successfully dispersed to join or form a new pack. The difference between the sexes is very similar to that seen in nondispersers.
10.5). This substitution has remarkably little effect on the comparison of dispersers versus nondispersers, or on the comparison of direct versus indirect components of inclusive fitness (Tables 10.4 and 10.5).
10.4 Incomplete Reproductive Suppression: Breeding by Subordinates In our analysis to this point, we have assumed that wild dogs never breed without first becoming dominant. In reality, reproductive suppression is not absolute. In all cooperatively breeding mammals and birds that have been studied in detail, there is some reproduction by social subordinates, and there is good evidence that these cases are not simply random failures of the normal mechanisms of reproductive suppression, at least in some species (Creel & Waser 1991; Keller & Reeve 1994). We suggest that reproductive suppression should not be viewed as a dichotomous trait that is present or absent in a species, but as a continuous trait. At the population level, the degree of suppression can be measured as the percentage of subordinates that do not breed, either annually (Creel & Waser 1991) or in their lifetime (Lacey & Sherman 1991). At the individual level, the degree of suppression can be measured as the age- and sex-specific probability of breeding. At the behavioral and endocrine levels, the degree of suppression can be quantified by comparing the mating rates and reproductive hormone levels of subordinates and dominants (Chapter 9). In this section we apply the reproductive skew model of Vehrencamp (1983), which uses inclusive fitness calculations to compare two measures:
PAT T E R N S O F R E L AT E D N E S S ▪ 241
(1) the inclusive fitness that a subordinate can expect to obtain if it does not disperse, does not breed, and helps to raise the offspring produced by the pack’s dominant pair, and (2) the inclusive fitness that the subordinate could expect if it dispersed. When the second measure exceeds the first, then the fitness benefits of remaining a reproductively suppressed helper are insufficient to make helping preferable to dispersal, and subordinates should require personal reproductive success equal to the difference. This personal reproductive success is what Keller & Reeve (1994) call the staying incentive, and Creel & Waser (1991) call an evolutionary compromise between dominant and subordinate. The dominant concedes part of its reproductive monopoly to retain valuable helpers. The subordinate concedes to remain and help, provided it begins to share reproductive privileges. Figure 10.9 shows the size of the staying incentive that subordinates should require in order for selection to favor nondispersal. For females, fully suppressed nondispersers have inclusive fitness similar to dispersers at ages one and two. At ages three and older, dispersing females have greater fitness than fully suppressed nondispersers, by a factor ranging from 3.3 to 6.7. For males, the inclusive fitness of suppressed nondispersers is substantially greater than that of dispersers at ages one and two. Among prime-aged males (three to five years), dispersers have greater fitness than suppressed nondispersers, by a factor of 3.4–6.6. Among older males, the fitness of dispersers and suppressed nondispersers is similar at ages six and seven, and by age eight the fitness of suppressed nondispersers again exceeds that of dispersers. These comparisons lead to several predictions about patterns of reproductive suppression in wild dogs. The most fundamental prediction is that reproductive suppression should not be complete in wild dogs, because the staying incentive is substantially greater than zero for subordinates of many ages. Observational and genetic data from several populations confirm that subordinate wild dogs do sometimes breed, and that their pups are sometimes raised (Chapter 9, and see below). The second prediction is that females should be the more dispersive sex, because the staying incentive is larger for females than for males in six of eight age classes, equal in one class, and larger for males in only one class Figure 10.9). In line with this prediction, female wild dogs in Selous are significantly more likely than males to disperse (Chapter 8). The third prediction is that the alpha female should share reproduction with old subordinates more often than with young ones, while middle-aged males should share paternity more than young or old males. We discuss these predictions below. Observations of mating behavior and endocrine data both suggest that subordinate males might occasionally sire young (Malcolm 1979; Reich 1981; Creel et al. 1997a). Multiple paternity within single litters is common in carnivores, including canids (badgers: Evans et al. 1989; dwarf mongooses: Keane et al. 1994; Ethiopian wolves: Gottelli et al. 1994; lions:
242 ▪ C H A P T E R 1 0
Figure 10.9 The staying incentive (measured in numbers of offspring raised to one year) that subordinates should require, as a function of age. The staying incentive is the direct fitness required to favor nondispersal, and is equal to the inclusive fitness of a disperser minus the inclusive fitness of a reproductively suppressed nondisperser.
C. Packer personal communication), so subordinate males might slip a few offspring into the mix, even when the alpha female is the only breeder. When a subordinate female also breeds, the odds of fathering offspring are probably better for subordinate males, because it is difficult for a single male to monopolize access to more than one female in estrus. Genetic data confirm that subordinate males occasionally father offspring, but the data are too sparse to test whether age affects patterns of paternity. Based on microsatellite data, 3 or 4 out of 29 pups (10–14%) in Kruger were fathered by subordinates (Girman et al. 1997). In Selous, only 9 pups in two litters could be accurately tested for paternity using microsatellites. For 2 pups (22%, one in each litter), the alpha male could be excluded as the father. In both of these cases, three subordinate males were all possible fathers, based on their genotypes. In one of the cases, the beta male was observed to mate and thus is the most likely candidate. In the other case, a subordinate male (who was not the beta) actively challenged the alpha male during the estrus period, became behaviorally dominant, and mated. Though he returned to subordinate status soon after the mating period, he is the most likely candidate for paternity. Wild dog packs occasionally hold more than one pregnant female (Malcolm 1979; Reich 1981). Of 39 litters in Selous, 8 (21%) were born to subordinate females, and 27% of all packs had more than one breeding female, compared to 38% in the Masai Mara (Fuller et al. 1992a) and 6% in Kruger (Reich 1981). Subordinates that become pregnant usually give birth soon after the dominant female (usually within a week but up to two months later). It is not always clear whether subordinate females’ litters are raised,
PAT T E R N S O F R E L AT E D N E S S ▪ 243
Figure 10.10 Pups at first count and pups raised to one year in single-female litters and in joint litters. The central line in each bar shows the mean. Bars show Ⳳ 1 SE and whiskers show ⳮ 1.96 SE.
and in some cases they are clearly killed (Malcolm 1979; personal observations). On the other hand, it is clear in some cases that subordinates contribute pups that are raised in a joint litter. One of our study packs was raising 34 pups when we first encountered them, too many to have been whelped by one female. There were three lactating females, and the pups were not all of the same age, based on distinct differences in body size. Overall, joint litters in Selous averaged 15.9 Ⳳ 3.4 pups at first count, twice as large as alphaonly litters (7.0 Ⳳ 0.8; F1,28 ⳱ 13.32, P ⳱ 0.001). Based on pups raised to one year, joint litters were also larger than alpha-only litters (16.8 Ⳳ 0.61 vs. 3.8 Ⳳ 0.38 pups raised; F1,23 ⳱ 68.41, P ⬍ 0.001. Note that the mean for yearlings raised is slightly larger than the mean for pups produced in joint litters. This artifact arises because we could not observe all litters for a full year). In summary, most of the offspring produced by subordinate females are apparently raised (Figure 10.10), at least under the ecological and demographic conditions of Selous. Under tighter ecological constraints, infanticide by the alpha female is probably more common (for example, in Serengeti: Malcolm 1979), and it should be borne in mind that most subordinates (⬎90%) do not become pregnant in all populations, including Selous. Inclusive fitness calculations suggest that one- and two-year-old subordinate females should not require a staying incentive, and old subordinate females should require a larger staying incentive than young ones (Figure 10.9). Figure 10.11 shows the pregnancy rate among subordinate females as
244 ▪ C H A P T E R 1 0
Figure 10.11 The percentage of subordinate females that become visibly pregnant and give birth in Selous, as a function of age.
a function of age, with a good qualitative match to the predictions of Figure 10.9. As a quantitative test, we regressed the observed pregnancy rate on the predicted staying incentive, with one point for each age class. This test also showed a good fit between the reproductive skew model and the data (t5 ⳱ 2.93, R 2 ⳱ 0.68, P ⳱ 0.048). Pregnancies among subordinates are not random with respect to age, and match expectations from Vehrencamp’s skew model. This result is almost identical to an analysis of reproductive skew in another social carnivore, the dwarf mongoose, for which more detailed tests were possible (Creel & Waser 1991). In general, the data are limited but suggest that patterns of reproductive suppression in female social carnivores match the predictions of skew models based on inclusive fitness theory. For males, the data are sparse and more equivocal (Creel & Waser 1997).
11
Interspecific Competition with Larger Carnivores
Early studies of African wild dogs and spotted hyenas noted that interference competition between the two species was common. Estes & Goddard (1967) noted that spotted hyenas “seriously compete with wild dogs for their kills” (pp. 66–67) in the Ngorongoro Crater. In Serengeti and Ngorongoro, Kruuk (1972) observed that hyenas were present at 74% of 62 wild dog kills, and ate at 60%, but wild dogs appropriated only 5 of 465 hyena kills. Kruuk (1972) summarized his observations by stating that “the relationship between wild dogs and hyenas is one of one-sided benefit for the hyenas” (p. 137). Similar observations led Malcolm (1979) to propose that competition with spotted hyenas might limit Serengeti wild dogs in number or distribution. Summarizing observations in Serengeti from 1967 to 1978, Lory Frame (1985) stated: Hyenas typically assembled behind wild dog packs as they hunted, and we recorded periods lasting weeks at a time in which hyenas stole almost all kills made by the dogs before the latter finished eating. Wild dogs consequently had to travel farther and make more kills. Their pups lost condition and succumbed to disease. . . . The plains population of wild dogs dwindled during our study from 110 adults in 1969 to about 30 in 1978. We attributed the decline mostly to the dry season increase in spotted hyenas on the plains. (p. 3) Recent research suggests that lions may also limit wild dog numbers, either through competitive exclusion from areas of high prey density or by direct predation (Mills & Biggs 1993). Mills and Gorman (1997) showed that wild dogs in Kruger National Park did not favor habitats with high impala density (although impala comprised 71% of their prey) because these habitats were also favored by lions. Analyses of habitat selection in Selous also suggest that wild dogs avoid areas heavily used by lions (Chapter 3). A wild dog is outweighed 7 to 1 by a female lion, and 2 to 1 by a hyena, so it is plausible that lions or hyenas may be dominant competitors. Lions and hyenas obtain food both by hunting and by scavenging; the frequency of scavenging varies among ecosystems (e.g., 26% of meals for Serengeti hyenas, but 50% for Kruger hyenas; Kruuk 1972; Mills & Biggs 1993). Although pack-living may help wild dogs defend kills or avoid predation, this benefit is equivocal because lions and hyenas also live in groups. Lion and hyena feeding group sizes vary considerably among ecosystems, but can be
246 ▪ C H A P T E R 1 1
large (e.g., the mean number of hyenas feeding at wildebeest kills was 26 in Ngorongoro and 15 in Serengeti; Kruuk 1972). In this chapter we test the hypothesis that wild dogs are limited (perhaps locally excluded) by competition with spotted hyenas and/or lions. First, we summarize data on population density for the three species, and test for negative correlations. Second, we test for avoidance in the spatial distributions of wild dogs, lions, and hyenas in Selous. Third, we examine dietary overlap among the species to assess the potential for exploitation competition. Fourth, we examine data on the frequency and outcome of interference competition at wild dog kills. Finally, we describe non–food-related interactions among the species. In this chapter our focus is exclusively on interspecific competition: Other ecological limiting factors are discussed elsewhere.
11.1 Specific Methods We tabulated data on population density, diet, and behavioral interactions between wild dogs, lions, and spotted hyenas, for Selous and other ecosystems where all three species have been studied (sources noted in Tables 11.2 to 11.4 and Table 11.6). The methods used to gather these data varied among studies, but, lacking an objective method of making adjustments, we simply used the population densities reported in the original studies. If more than one density was available for a given species/area, the mean was tabulated. All densities exclude juveniles but include subadults if they were distinguished. Densities of hyenas and lions were only taken from ecosystems with known wild dog density. Table 11.2 summarizes population densities for nine protected areas well scattered across the wild dogs’ range (excluding west Africa, for which data are not available). These areas are largely natural with the following exceptions. The Save Valley Conservancy in Zimbabwe was created within the last decade by reintroductions into a relatively small area (3,400 km2) of land formerly used for ranching (Alistair Pole, personal communication). Small villages are found inside Moremi, including livestock. Portions of the perimeter of Kruger is fenced. Animals are culled in Kruger, including a hyena cull in the 1970s. Sport hunting (including lions and hyenas) occurs six months annually in Selous. Pastoralists and cattle use Ngorongoro and occasionally enter Serengeti. The wild dog population of Hluhluwe was artificially reintroduced in 1981 but has increased, so that the density in 1993 is likely to be ecologically meaningful. Hwange and Serengeti are relatively undisturbed. Wiresnare poaching occurs in all areas. Densities for two periods were tabulated for Serengeti (Table 11.2) because carnivore densities have changed considerably during the 25 years for which data are available. For all parks except Serengeti, the densities for all
1.30 Ⳳ 0.13* 1.20 Ⳳ 0.27 1.07 Ⳳ 0.20 0.94 Ⳳ 0.28 not measured4
Preference Ratio2,3 422 370 235 297 28 1,352
(31%) (27%) (17%) (22%) (2%)
Prey Herds Encountered 5818 5908 5183 6462 199 23,570
(25%) (25%) (22%) (27%) (1%)
Prey Individuals Encountered 1552 825 458 393 43 3271
(47%) (25%) (14%) (12%) (1%)
Kilometers Traveled in Habitat
2
Entries give raw numbers, then percentage in brackets. Preference ratio ⳱ % of radiolocations within a habitat type ⳰ % of home range covered by that habitat type. 3 Mean Ⳳ SE. 4 Riverine thicket covered ⬍1% of the wild dog study area, making its preference ratio highly sensitive to measurement error. *P ⬍ 0.05 for single-point t-test comparing preference ratio with 1.
1
Deciduous Woodland Thorn Woodland Long Grass Short Grass Riverine Thicket Totals
Habitat Type
Wild Dog
Table 11.1 Encounters between wild dogs and prey as a function of habitat type1
3.75 7.16 11.31 16.4 4.63
Prey Individuals per km Traveled 4 2 5 3 1
Lion Preference Rank
248 ▪ C H A P T E R 1 1
species either come from the same study area, or apply to the entire park. Serengeti is more complex because it experiences an annual migration of prey that affects carnivore distributions. Wild dog densities came primarily from the southeastern short grass plains, but occasionally from the plainswoodland boundary to the northwest (Frame et al. 1979; Malcolm 1979; Burrows et al. 1994). Lion densities came from an overlapping area focused on the plains-woodland boundary (Schaller 1972; Packer et al. 1988; Packer 1990). Packer (1990) suggested that lion density for this area is applicable to much of the area in which wild dogs were studied. Hyena density for the earlier period is Kruuk’s (1972) estimate for the wildebeest wet season range, which encompassed the wild dog study area. Hyena density for the later period came from an estimate of the number of hyenas that either reside on the plains or frequently travel there to hunt (Hofer & East 1995). Hyena density in the wet season was used for both periods because this is the puprearing season for wild dogs in Serengeti. We determined correlations between the densities of species across ecosystems by ordinary least squares regressions, using residual plots to test assumptions. A negative exponential model (y ⳱ eaⳭbx) or piecewise linear regressions maximized the variance explained (r 2) for regressions of wild dog density on lion and hyena densities. A linear model (y ⳱ aⳭbx) maximized r 2 for the regression of lion density on hyena density. Within the Selous ecosystem, we mapped the spatial distributions of wild dogs and hyenas. For wild dogs, we used 1,330 satellite fixes to determine home ranges, as described in Chapter 3. With these home ranges, we used Idrisi (Eastman 1997) to map the number of packs known to use each point in the study area. The spatial distribution of hyenas was based on playback results, with the number of hyenas responding to each playback plotted on a 0.083⬚ by 0.083⬚ grid (5 minutes of arc, 9.26 km at the latitude of Selous). Mean values were used for cells with more than one playback (9 cells). We examined spatial correlations between the distributions of wild dogs, lions, and hyenas with Cramer’s V statistic, using the cross-tabulation module of Idrisi.
11.2 Carnivore Densities and Distributions in Selous Selous Wild Dogs From counts of known individuals, wild dog density in northern Selous was 0.038 adults per km2. Including young of the year, population density peaked in August, just after the birth season, at a mean density of 0.057 individuals per km2 (Chapter 2). Interestingly, this density matches reasonably well with an estimate of 0.052 individuals per km2 based on sightings of uncollared packs in 1991–1992. Figure 11.1 shows the spatial distribution of wild dogs
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 249
Figure 11.1 Spatial distribution of wild dogs in the northern Selous Game Reserve, shown by plotting the number of packs known to use each point.
in northern Selous. All of the study area was used, except for long grass areas in the northeast and northwest, which held little prey year-round. Chapter 3 describes space use and home range overlap in detail. Selous Spotted Hyenas A mean of 7.9 Ⳳ 1.0 adult hyenas responded to each tape playback (min. ⳱ 0; max. ⳱ 22; median ⳱ 8). To convert this number to an estimate of population density, we estimated the area from which responding hyenas were drawn. In response trials, 18 of 18 hyenas responded to playbacks at distances from 1.5 to 2.0 km. A two-kilometer response radius equates to a response area of 12.6 km2, and a density estimate of 0.62 adult hyenas/km2. No hyenas responded from distances greater than 3.7 km2. This response radius yields a density estimate of 0.18 adult hyenas/km2. The likelihood of responding dropped off sharply beyond 2.8 km; this response radius gives a density of 0.32 adult hyenas/km2, which we consider the best estimate. Mills (1985) obtained a similar response distance using the same method in a similar habitat. Figure 11.2 shows the spatial distribution of hyenas, derived from the playback data. Hyenas used the entire study area that we censused, and, like the wild dogs, their distribution was not even. The resolution of the hyena
250 ▪ C H A P T E R 1 1
Figure 11.2 Spatial distribution of spotted hyenas in the northern Selous Game Reserve, shown by plotting the number of hyenas that responded to playbacks in each grid cell. Values were averaged for cells with more than one playback. The plot frame is the same as in Figure 11.1, although the resolution is more coarse.
distribution is coarse, because the data are sparse. However, spatial crosstabulation detects a significant similarity between the wild dog and hyena distributions (Cramer’s V ⳱ 0.71, χ2 ⳱ 40.8 df ⳱ 13, P ⬍ 0.001). Within our study site, areas that were heavily used by wild dogs were also heavily used by hyenas. This contradicts our earlier conclusion (based on a less powerful method) that the spatial distributions of wild dogs and hyenas were uncorrelated (Creel & Creel 1996). Selous Lions Direct counts of recognizable lions in nine prides living in an area of 350 km2 gave an estimate of 0.13 adult lions/km2. During periods in which we recorded all carnivores seen, 273 lions and 762 hyenas were noted, giving a ratio of 0.36 lion:1 hyena. Multiplying this proportion by the density of hyenas gives an estimate of 0.11 adult lions/km2. Rodgers (1974) estimated a density of 0.08 lions/km2 for similar habitat in the east of Selous. Overall, our density estimates range from 0.08 to 0.13 adult lions/km2. Of these, we use the middle estimate of 0.11 adult lions/km2 for comparison with other populations.
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 251
Studies in Kruger and Selous suggest that, within an ecosystem, wild dogs actively avoid lions, but do not avoid hyenas. In Kruger National Park, impala comprised 73% of 79 wild dog kills (Mills & Gorman 1997). In accord with this prey preference, the most preferred habitat (out of six types) was the same for wild dogs and impala. However, the second- and third-mostpreferred habitats for wild dogs were the least preferred for impala, and the second-most-preferred habitat for impala was least preferred for wild dogs. Overall, the habitat preference ranks of wild dogs did not correlate with those of impala. This pattern was explained by avoidance of lions, with a significant negative correlation between the ranked habitat preferences of wild dogs and lions (Mills & Gorman 1997). The ranked habitat preferences of wild dogs and hyenas were not significantly correlated, but the three habitats least used by hyenas were also the three habitats most preferred by wild dogs, and the correlation between wild dog habitat preferences and an index of hyena density tended to be negative (r ⳱ ⳮ0.61). Wild dogs also avoided lions in the Selous Game Reserve. For an area of 2,600 km2, there was a significant negative correlation between the spatial distributions of wild dogs and lions (Figure 11.3: Cramer’s V ⳱ 0.30, df ⳱ 36, χ2 ⳱ 1310, P ⬍ 0.001). Wild dogs’ only significant habitat preference was for deciduous woodland, which held a low density of prey, but was little used by lions (Table 11.1). Data from 1,352 encounters with prey during 3,271 kilometers of wild dog follows showed that an average of only 3.75 prey were encountered per kilometer, significantly lower than prey encounter rates in other habitat types, which ranged from 4.6 to 16.4 prey/km (Table 11.1). The energetic costs of preferentially using habitats with low prey density are discussed below. Direct encounters between wild dogs and lions are rarely seen, perhaps because wild dogs can detect lions from a distance and avoid them. As mentioned previously, the strength of selection on avoidance behavior will depend on the cost of failing to avoid an encounter. For wild dogs, encounters with lions are clearly costly. Dogs were killed in 2 of 19 encounters, abandoned food in 4, and traveled several kilometers at a trot in some retreats (for further examples from Kruger, see Mills & Gorman 1997). Wild dogs exhibited clear avoidance in 17 of 19 natural encounters with lions in Selous. If dogs were resting when they detected lions, they moved off in single file, regardless of the time and temperature. If dogs encountered lions when traveling, their behavior changed immediately and markedly. From a loose group following several lines of travel, they congregated in single file, adopted a more linear travel route, moved at a steady trot, and frequently glanced back. The dogs sometimes mobbed a lion group before retreating, by approaching very cautiously to within 20 meters (standing on their hind legs if the view was obstructed) and alarm barking. In response to audiotape playbacks of lion roars broadcast from 300 m or less, wild dogs tried to locate the lions within one minute, by approaching
252 ▪ C H A P T E R 1 1
Lion Habitat Preference High
Low
Figure 11.3 Spatial distribution of lions in the northern Selous Game Reserve, shown by projecting habitat preferences onto a map of habitat types (see Chapter 2). The plot frame is the same as in Figure 11.1.
carefully or scanning in the direction of the roars. Failing to locate any lions, their next movement was directed at least 90⬚ away from the roars in 11 of 12 trials with six different packs. Collectively, natural encounters and playbacks showed clear-cut avoidance in 28 of 31 cases (χ 2 with Yates correction ⳱ 18.6, P ⳱ 0.001; Creel et al. 2001). Wild dogs do not obviously avoid spotted hyenas. As noted above, there was not a significant negative correlation between the habitat preferences of the two species in Kruger. In Selous, the spatial distributions of wild dogs and hyenas actually show a significant positive association (Cramer’s V ⳱ 0.71, χ 2 ⳱ 40.8, df ⳱ 13, P ⬍ 0.001: This contradicts our prior report that the distributions were uncorrelated, which used a less powerful method [Creel & Creel 1996]). It is interesting to speculate why wild dogs avoid lions but not hyenas. Perhaps wild dogs would avoid hyenas if they could, but they simply cannot. Hyena densities are typically two to four times higher than lion densities (Creel & Creel 1996), so hyenas cover the landscape more continuously than lions. Second, hyenas are well-adapted for scavenging. They are able to detect kills by smell from distances up to 4.2 km, and by hearing from distances up to 10.5 km (Mills 1990). Given spotted hyenas’ extraordinary mobility (Mills 1990; Hofer & East 1995) and ability to locate kills, there may be no “holes” in the hyena distribution, particularly for ecosystems with high hyena density. Finally, if hyenas also
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 253
avoid areas of high lion density, then the association between wild dogs and hyenas could be caused by a shared pattern of lion avoidance. It is difficult to quantify the costs of active spatial avoidance for subordinate competitors. In some examples, subordinates are displaced into habitats where their rates of encounter with prey are low. If hunting is energetically costly, then a reduction in prey encounter rates is likely to carry a cost in terms of reproduction or survival. A rough sketch of the costs of avoidance is possible for African wild dogs. By avoiding lions, wild dogs in Selous encountered only 3.75 prey individuals for each kilometer they traveled in deciduous woodland, while the average prey encounter rate for other habitats was 9.88 prey/km (Table 11.1). All else equal, wild dogs would have to increase their hunting effort by a factor of 2.6 to maintain average foraging success while restricting their movements to deciduous woodland. Gorman et al. (1998) estimated that hunting increases a wild dog’s metabolic rate up to 25 times the resting rate, with an average daily expenditure of 5.2X basal metabolic rate. Independent data suggest that the maximum sustainable longterm expenditure of energy is 6–7X BMR (Hammond & Diamond 1997). Consequently, it is not likely that wild dogs could maintain the 2.6-fold increases in foraging effort that would be required by a complete retreat to deciduous woodland.
11.3 Correlations between Species Densities Table 11.2 shows population densities for wild dogs, hyenas, and lions from eight ecosystems. Of the three species, hyenas were most abundant in every ecosystem. Lion densities were intermediate in every ecosystem. Wild dogs were far less common than lions and hyenas in every ecosystem except one, contradicting the general rule that small species tend to outnumber large ones (Blackburn & Gaston 1994). The one exception was the Save Valley Conservancy, where all three carnivores were reintroduced in 1991 to an area formerly depleted of carnivores (Alistair Pole, personal communication). In this case, densities recorded in the late 1990s may simply reflect numbers of each species that were reintroduced. There is a strong negative correlation (Figure 11.4) between wild dog and hyena densities (piecewise linear regression: r 2 ⳱ 98%, P ⬍ 0.05). Figure 11.5 shows that there is also a strong negative correlation between wild dog and lion densities (piecewise linear regression: r 2 ⳱ 99%, P ⬍ 0.05). The significance of these correlations is striking, given that the limited data permit us to detect only strong relationships. In contrast to their relationship with wild dogs, there was a strong positive correlation (Figure 11.6) between lion and hyena densities (t ⳱ 8.60, P ⳱ 0.003, r 2 ⳱ 96%). Both of the larger carnivores attain high densities in ecosystems with high prey densities (Stander 1991). These three correla-
254 ▪ C H A P T E R 1 1 Table 11.2 Population densities of African wild dogs (WD), spotted hyenas (HY), and lions (LI) Adults/km2 Wild Dogs
Spotted Hyenas
Selous, Tanzania
0.040
0.32
0.11
8:3:1
Kruger, RSA
0.017
0.135
0.10
8:6:1
Ngorongoro, Tanzania
0
1.43
0.16–0.24
70Ⳮ:10Ⳮ:1
Serengeti, TZ 1967–79
0.015
0.17
0.079–0.094
11:6:1
Serengeti, TZ 1985–91
0.0067
1.10
0.14
164:21:1
Hluhluwe, RSA Hwange NP, Zimbabwe Moremi NP, Botswana Save Conservancy, Zimbabwe
0.033
0.34
—
0.015
0.2
—
—
—
—
Population
0.040
—
0.040
0.008
Lions
0.002
Ratio HY:LI:WD
10:?:1
Recently introduced populations
Sourcea WD:1, 2 HY:1 LI:1, 24 WD:3, 4, 5 HY:6, 25 LI:4, 29 WD:7, 8 HY:9 LI:26, 27 WD:10–12 HY:9, 12 LI:26 WD:12–14 HY:15 LI:28 WD:17, 18 HY:19, 20 WD:21, 22 HY:16 WD:23 30
a
Sources: 1, Creel & Creel 1995a; 2, Creel & Creel 1996; 3, Maddock & Mills 1994; 4, Pienaar 1969; 5, Reich 1981; 6, Mills 1985; 7, Personal communication with Ngorongoro Conservation Area Authority staff; 8, Malcolm 1979; 9, Kruuk 1972; 10, Frame et al. 1979; 11, Malcolm 1979; 12, Burrows et al. 1994; 13, Laurenson et al. 1990; 14, Scott 1991; 15, Hofer & East 1995; 16, Woodroffe & Ginsberg, 1997; 17, Maddock 1993; 18, M.G.L. Mills report at IUCN Lycaon PVA, Arusha 1992; 19, Whateley & Brooks 1978; 20, Whateley 1981; 21, Childes 1988; 22, Ginsberg 1993; 23, McNutt 1995; 24, Rodgers 1974; 25, Mills & Briggs 1993; 26, Schaller 1972; 27, Rudnai 1970; 28, Packer 1990; 29, Mills 1995; 30, Alistair Pole, personal communication.
tions show that, at the ecosystem level, wild dog densities are low where the densities of larger carnivores are high, supporting the hypothesis that wild dogs are limited by competition. This hypothesis is also supported by changes in density over time in two areas, Serengeti and Ngorongoro. In the late 1960s, wild dog density in Serengeti was comparable to healthy populations in other ecosystems (0.026–0.032 adults/km2; Malcolm 1979; Burrows et al. 1994). A long decline in wild dog numbers (nearly to local extinction) was paralleled by
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 255
Figure 11.4 Relationship of wild dog population density to hyena population density, across ecosystems. Lines show the best fit from piecewise linear regression.
Figure 11.5 Relationship of wild dog population density to lion population density, across ecosystems. Lines show the best fit from piecewise linear regression.
256 ▪ C H A P T E R 1 1
Figure 11.6 Relationship of lion population density to hyena population density, across ecosystems. Dashed lines show 95% confidence limits for the linear regression.
steady increases in lion and hyena numbers, in response to increasing prey populations (Table 11.2; Dublin et al. 1990). Hyenas, in particular, increased, by a factor of 2.4 for the ecosystem as a whole (Kruuk 1972; Hanby & Bygott 1979; Hofer & East 1995). Because wild dogs and spotted hyenas have very similar diets in Serengeti (Tables 11.3 and 11.4), the failure of wild dogs to track the huge increase in wildebeest population requires an explanation. The declining population of Serengeti wild dogs was exposed both to increasing numbers of competitors and increasing problems with viral disease. Because lions and hyenas in Serengeti carry canine distemper virus (Haas et al. 1996; Roelke-Parker et al. 1996) and perhaps other viruses (e.g., Mills 1993) these problems are interrelated. As with most populations of most species, it is unlikely that a single factor drove the Serengeti wild dogs’ decline. For the purposes of this chapter, we note that the negative correlation between wild dog and larger carnivore densities remains strong if data for Serengeti from the latter period are excluded (for hyenas: r ⳱ ⳮ0.95, t ⳱ 5.03, P ⳱ 0.015; for lions: r ⳱ ⳮ0.93, t ⳱ 3.54, P ⳱ 0.07). In a simpler case study, wild dogs were found in the Ngorongoro Crater in the mid-1960s following a severe crash in the lion population (Estes & Goddard 1967; Packer et al. 1991). As lion numbers recovered, the wild dogs disappeared. Lion density increased more than 5-fold between 1965 and 1980, then stabilized; wild dogs have remained absent.
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 257 Table 11.3 Primary prey species of African wild dogs
Population Selous GR, Tanzania Kruger NP, RSA
Serengeti NP, Tanzania Ngorongoro, Tanzania Aitong, Kenya Moremi, Botswana Hwange, Zimbabwe
Rank of Prey 1: 2: 3: 1: 2: 3: 1: 2: 3: 1: 2: 3: 1: 2: 3: 1: 2: 1: 2: 3:
Wildebeest Impala Warthog Impala Kudu Reedbuck and Waterbuck Wildebeest Thomson’s gazlle Zebra Wildebeest Thomson’s gazlle Grant’s gazelle Thomson’s gazelle Wildebeest Impala Impala Gr. Kudu Impala Gr. Kudu Duiker
Measure of Importance a
Source b
Biomass
1
Percent of Kills
2, 3, 4
Biomass
5, 6, 7
Biomass
8
Biomass
9, 10
Biomass
11
Number of Kills
12
a
Original data converted to biomass where possible. Sources: 1, Creel & Creel 1995b; 2, Pienaar 1969; 3, Reich 1981; 4, Mills & Biggs 1993; 5, Schaller 1972; 6, Fanshawe & Fitzgibbon 1993; 7, Kruuk & Turner 1967; 8, Estes & Goddard, 1967; 9, Fuller & Kat 1990; 10, Fuller & Kat 1993; 11, McNutt 1995; 12, G. Rasmussen, unpublished newsletter.
b
11.4 Diet Overlap Negative correlations between the densities of wild dogs and larger carnivores could be due to competitive limitation. Negative correlations could also arise if wild dogs depend on different prey than the larger carnivores, and the densities of prey species vary among ecosystems. These alternatives can be disentangled using data on diet overlap; high overlap implicates competition. Table 11.3 shows that wildebeest (Connochaetes taurinus) were wild dogs’ most important food in three of six studies. Wildebeest were among the top two prey types in four of six studies. Other wild dog prey were usually smaller than wildebeest. Impala (Aepyceros melampus) were the
258 ▪ C H A P T E R 1 1 Table 11.4 Primary species eaten by spotted hyenas Population Selous, Tanzania Kruger, RSA Serengeti, Tanzania Ngorongoro, Tanzania Kalahari, RSA
Rank of Primary Foods 1: 2: 1: 2: 1: 2: 3: 1: 2: 3: 1: 2:
Wildebeest Impala Steenbok Wildebeest, impala, kudu Wildebeest Thomson’s gazelle Zebra Wildebeest Zebra Thomson’s gazelle Gemsbok Wildebeest
Measure of Importance a Opportunistic observations, % of kills
Source b 5 4
Biomass killed or scavenged
2, 3
Biomass killed or scavenged
2
Biomass killed or scavenged
1
a
Original data converted to biomass where possible. Sources: 1, Mills 1990; 2, Kruuk 1972; 3, Hofer & East 1993a; 4, Mills & Biggs 1993; 5, S. Creel & N. M. Creel, unpublished data.
b
most important prey for two populations, and were among the top two prey types in four of six studies. Gazelles were important prey for wild dogs in savannah, where impala are not available. The diet of spotted hyenas overlapped extensively with that of wild dogs (Table 11.4). As with wild dogs, wildebeest were hyenas’ most important food in three of five studies, and among the top two foods in four of five studies. In three of four places where both wild dogs and hyenas were studied, wildebeest were the most important food for both species. In Serengeti, the three most important food species of wild dogs and hyenas were the same. Like wild dogs, hyenas often took prey smaller than wildebeest (e.g., impala and gazelles) in four of five studies. However, hyenas also took large prey other than wildebeest (e.g., zebra Equus burchelli, gemsbok Oryx gazella) to a greater extent than wild dogs did. Generally, overlap between wild dog and hyena diets is substantial. Lions, like wild dogs and hyenas, dined often on wildebeest (Tables 11.5 and 11.6). Wildebeest were lions’ primary or secondary food in each of four studies. Unlike wild dogs, much of lions’ diet was made up of other species as large as wildebeest or larger (e.g., buffalo Syncerus caffer, zebra, and gemsbok). In Selous, buffalo were the most important prey in terms of meat obtained, though wildebeest were killed more often (Figure 11.7). In summary, the diet of wild dogs overlapped substantially with those of the larger carnivores, particularly hyenas. Two plains species (wildebeest and Thomson’s gazelle) and two woodland species (impala and greater kudu Tragelaphus strepsiceros) accounted for much of the overlap.
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 259 Table 11.5 Primary species eaten by lions Population Selous, Tanzania Kruger, S. Africa Serengeti, Tanzania Kalahari, S. Africa
Rank of Primary Foods 1: 2: 3: 1: 2: 3: 1: 2: 3: 1: 2:
Measure of Importance a
Source b
Biomass
4, 5
Biomass
2, 3
Biomass
1
Biomass
6
Buffalo Wildebeest Zebra & Eland Wildebeest Zebra Impala Buffalo Wildebeest Zebra/Warthogc Wildebeest Gemsbok
a
Original data sometimes converted to biomass from other units. Sources: 1, Scheel 1993; 2, Pienaar 1969; 3, Mills & Shenk 1992; 4, Rodgers 1974; 5, This study; 6, Mills 1990. c Warthogs were considered critical prey for small prides when migratory prey were absent. b
11.5 Direct Competition at Kills Competition at Wild Dog Kills The frequency and outcome of interference competition from hyenas at wild dog kills has been reported in five studies (Table 11.7). Hyenas rarely obTable 11.6 The diet of lions in the Selous Game Reserve Species
Number of meals a
Proportion of Meals
Proportion of Mass b
Buffalo Wildebeest Zebra Eland Warthog Waterbuck Greater Kudu Hartebeest Reedbuck Impala Sable
25 51 8 6 20 4 3 3 1 2 1
0.19 0.41 0.06 0.05 0.16 0.03 0.02 0.02 0.01 0.02 0.01
0.48 0.30 0.06 0.06 0.03 0.02 0.02 0.01 0.01 0.005 0.005
124
0.98
1.00
Totals a
Combined data, 50 meals that we observed and 74 reported by Rodgers (1974). Species-specific estimates of mass of individuals killed by lions come from Schaller (1972).
b
260 ▪ C H A P T E R 1 1
Figure 11.7 Rank of prey species in the diet of Selous lions.
tained food from wild dogs in two places: Selous (2% of kills) and Kruger (0% of kills). Both ecosystems have large and stable wild dog populations (Maddock & Mills 1994; Creel & Creel 1996). Both are also largely wooded, which makes kills difficult to locate for a would-be scavenger. Relative to savannah, woodland limits the direct line of sight, muffles the sound of dying prey, and reduces the ability of vultures to locate kills. Lions and hyenas often use the sight of descending vultures to detect a carcass, so fewer vultures at a kill translates to fewer large mammalian scavengers. For these reasons, hyenas were present at only 18% of wild dog kills in Selous (Table 11.7). When present, hyenas were most often solitary (Kruger: Mills & Biggs 1993; Selous: see below). Losses by wild dogs to hyenas were intermediate in Aitong (21% of kills), where wild dog density has fluctuated, including crashes to local extinction (Scott 1991). These fluctuations have been at least partially driven by disease (Kat et al. 1995). Fuller & Kat (1993) note that losses to hyenas may have been small in Aitong because “competitors were scarce” (p. 467). L. Frank (personal communication) suggests that hyenas were common in Aitong; perhaps interactions with wild dogs were infrequent because hyenas become more nocturnal in areas used by humans, while wild dogs do not, creating a niche divergence. In recent years, wild dogs have often been found in the Aitong area outside the Masai Mara National Reserve (Scott 1991), but not inside the reserve. The density of competitors (particularly lions: L. Frank, personal communication) is a plausible explanation for this pattern.
76 (18%) frequent 53 (86%) 46 (74%) 18 (41%)
404
52
62a
62
43
Number (%) with Hyenas
rarely
37 (60%)
53 (86%)
0 (0%)
14 (2%)
Number (%) at which Hyenas Ate
moderate
0.82–1.43
0.17
0.14
0.32
Hyena Population Density (adult/km2)
Intermediate
Open
Open
Wooded
Wooded
Thickness of Habitat
Moderate, varies with migration
Rapid, many
Rapid, many
Slow, few
Slow, few
Congregation of Vultures at Kills b
Gone
Gone
Large, stable Large, stable Gone
Status of Dog Population
Sources: 1, Creel & Creel 1995b; 2, Mills & Biggs 1993; 3, Fanshawe & Fitzgibbon 1993; 4, Kruuk 1972; 5, Fuller & Kat 1993. a Original data excluded kills of gazelle fawns. b Vulture congregation from personal observation; Mills personal communication; Holekamp, personal communication.
Selous, Tanzania1 Kruger, S. Africa2 Serengeti, Tanzania3 Ngorongoro/ Serengeti, TZ4 Aitong, Kenya5
Population
Number of Wild Dog kills
Table 11.7 The frequency and outcome of competition between wild dogs and hyenas at wild dog kills
262 ▪ C H A P T E R 1 1
Hyenas were common at wild dog kills in two areas: Serengeti and Ngorongoro (Table 11.7; Malcolm 1979; Kruuk 1972; Fanshawe & Fitzgibbon 1993). Hyenas ate at 60% of Ngorongoro wild dog kills (Kruuk 1972) and at 86% of Serengeti wild dog kills (data for Serengeti excluded gazelle fawns: Fanshawe & Fitzgibbon 1993). In these areas wild dogs have declined to local extinction (Ngorongoro: Malcolm 1979) or nearly to extinction (Serengeti: Burrows et al. 1994). Overall, there is a clear pattern: Wild dogs fare well where interference competition at their kills is rare, and fare poorly where interference competition is common. Scavenging and Food Intake Where scavenging is common, observers have varied in their assessment of its impact of on wild dogs’ food intake. Malcolm (1979) states that “[Serengeti] wild dogs regularly lose kills they have made to hyenas, which gather in large numbers and drive off the dogs” (p. 20). Fanshawe & Fitzgibbon (1993) state that “although hyenas ultimately took over all kills of adult prey [in Serengeti], their attempts to appropriate them frequently only succeeded when the dogs had nearly finished feeding” (p. 485). Estes & Goddard (1967) state that hyenas “seriously compete with wild dogs for their kills” (p. 67), and suggest that one function of pack living in wild dogs is “mutual protection against competitors (spotted hyenas)” (p. 69). As we noted at the beginning of the chapter, Frame’s (1985) observations in Serengeti included “periods lasting weeks at a time in which hyenas stole almost all kills made by the dogs before the latter finished eating” (p. 3). In general, authors who studied several packs over a long interval concur that scavenging by hyenas was substantial. Quantitative data on the impact of hyenas on wild dogs’ foraging time are available for two populations. In Serengeti (1985–1987, when hyena density was high; Table 11.2), wild dogs’ feeding time was reduced significantly when more than four hyenas were present (Fanshawe & Fitzgibbon 1993). Serengeti wild dogs’ feeding time decreased as the ratio of dogs to hyenas at the kill decreased (r ⳱ 0.63, P ⬍ 0.01; Fanshawe & Fitzgibbon 1993). When hyenas were present at dog kills, the median dog-to-hyena ratio was 2:1 (min. ⳱ ⬃0.1, max. ⳱ 9:1). Dogs were outnumbered by hyenas at 15 (30%) of 50 wild dog kills for which the ratio was reported (Fanshawe & Fitzgibbon 1993). Carbone et al. (1997) used Fanshawe & Fitzgibbon’s data in a quantitative model of the effects of hyenas on wild dogs’ food consumption, and concluded that an increase in pack size would not improve defense of carcasses well enough to increase the amount of meat available to each pack member. However, this conclusion is sensitive to the values of several variables that are not well estimated, particularly maximum gut capacity and time to fill the gut. Their estimate of gut capacity came from Reich (1981, p. 394), who simply weighed the gut contents (3.1 and 4.4 kg) of two wild dogs that were
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 263
shot. It is not likely that either of these dogs had a full stomach: Our observations show that a wild dog can consume 8–9 kg in one sitting. For example, four dogs can eat all but the spine, skull, pelvis, and gut contents of an adult male impala. Fuller et al. (1995) suggested that wild dogs might consume as much as 12 kg, but we never observed a meal that large. Carbone et al. (1997) also estimated that a wild dog gorges itself in 15 to 27 minutes, less than the gorging times for wild dogs undisturbed by scavengers in Selous. The model of Carbone et al. (1997) would show a stronger benefit to grouping if these values were adjusted. In Selous (where hyena density was moderate; Table 11.1) wild dogs’ feeding time was not reduced by the presence of hyenas: Mean feeding time was 34 minutes in the absence of (mammalian) scavengers, and 43 minutes when hyenas were present. When hyenas were present, the median dog-tohyena ratio was much higher than in Serengeti, 7:1 (min. ⳱ 0.63:1, max. ⳱ 20:1). When hyenas were present at wild dog kills, there was no detectable correlation between feeding time and the dog-to-hyena ratio (Figure 11.8: r ⳱ 0.17, N ⳱ 64, t ⳱ 1.41, P ⳱ 0.16), probably because the dogs almost always outnumbered hyenas by a wide margin, so few of our observations fell in the range of ratios observed by Fanshawe & Fitzgibbon (1993). Dogs were outnumbered by hyenas at only one of 336 Selous wild dog kills. Overall, these data suggest that where spotted hyena population density is high and visibility is good, hyenas accumulate at wild dog kills in sufficient numbers to have an impact on foraging success. Although wild dogs can defend their kills from small numbers of hyenas, large groups of hyenas can appropriate prey captured by wild dogs even before they have died. In Serengeti, it was not unusual to see hyenas loping at the back of a wild dog hunt before a prey animal was grabbed (Frame 1985; personal observations). Where hyena density is lower and visibility is poor, competition at wild dog kills has little impact. Wild dogs rarely scavenge. Kruuk (1972) reported that wild dogs appropriated 5 of 465 hyena kills in Serengeti and Ngorongoro. In Selous, wild dogs successfully scavenged once from lions, four times from hyenas and three times from leopards in 310 days of observation (Creel & Creel 1995b). In Kruger, wild dogs were not observed to scavenge from lions or hyenas (Mills & Biggs 1993). Kruuk’s (1972) summary that “on the whole, the relationship between wild dogs and hyenas is one of one-sided benefit for the hyenas” (p. 137) applies broadly, to lions as well as hyenas.
11.6 Interactions Away from Kills Wild dogs and lions interact rarely, and wild dogs avoid areas of high lion density (see above; Mills & Gorman 1997). Wild dogs usually move away from the sound of lions roaring nearby, even when they are resting at night.
264 ▪ C H A P T E R 1 1
Figure 11.8 Feeding time of wild dogs in Selous as a function of the ratio of dogs and hyenas present at the carcass. Dashed lines indicate 95% confidence limits for linear regression, which was not significant.
If wild dogs encounter lions by chance, they generally move away. Occasionally, the dogs mob the lions prior to retreating (Figure 10.6c), and on one occasion a pack of 13 dogs drove 2 lionesses with cubs away by mobbing them aggressively for half an hour. Despite infrequent interactions, direct predation on wild dogs by lions is common. Nineteen wild dog deaths in Kruger have been attributed to lions (van Heerden et al. 1995). These comprise 39% of pup deaths and 43% of adult deaths of known cause (Mills & Gorman 1997). In Moremi, 7 (50%) of 14 known-cause deaths were due to lions (J. W. McNutt, unpublished data; Ginsberg et al. 1995). In Selous, 4 (9%) of 45 known-cause deaths were due to lions. An attempted reintroduction of captive wild dogs into Etosha National Park failed when a pride of lions systematically hunted and killed the dogs over several weeks (Schee-
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 265
pers & Venzke 1995). Lions scavenging from wild dogs were observed to kill one pup in Serengeti (T. Caro, personal communication). Two possible cases of predation by wild dogs on lions have been recorded in Kafue National Park (Mitchell, Shenton, & Uys 1965, cited in Schaller 1972). Predation on wild dogs by hyenas is much less common than predation by lions. One case has been reported from Moremi (Ginsberg et al. 1995) and two from Selous (Creel et al. 1995). The two wild dogs killed in Selous were pups that were separated from their pack while suffering from anthrax. Malcolm (1979) reported that “two litters of pups that were left unattended at a time of food scarcity were almost certainly killed by hyenas” (p. 111). To our knowledge, wild dogs have not been observed to kill hyenas. Away from kills, wild dogs and hyenas often ignore one another, and sometimes rest near one another (Estes & Goddard 1967) Frame & Frame (1981) described a hyena grooming a resting wild dog. However, wild dogs frequently go out of their way to harass and mob hyenas away from food, pursuing them at full speed up to 300 meters (described in detail by Estes & Goddard 1967; Frame & Frame 1981). In Selous and Ngorongoro, wild dogs have been observed to attack hyenas at hyena dens, with up to 30 hyena adults and cubs present (personal observations; Estes & Goddard 1967). If they are outnumbered, hyenas mobbed by wild dogs are submissive, despite their size advantage (Estes & Goddard 1967). For example, five Selous wild dogs dragged an adult hyena over 10 meters and bit it more than 20 times, while it made no attempt to retaliate. Hyenas become much more aggressive at night, or when the numbers of hyenas present at a fight approaches the number of wild dogs. In these encounters, hyenas were occasionally wounded, but we never saw a wild dog bitten. The wounds of hyenas rarely appeared serious, but they occasionally had fist-sized bite wounds. Interestingly, hyena behavioral submission is similar to that of wild dogs, and hyena cubs that submitted were sometimes not attacked. In summary, interactions between hyenas and wild dogs away from kills are sometimes neutral, but are often highly aggressive. Hyenas and wild dogs interact frequently, and within Selous they have similar spatial distributions. Lions and wild dogs interact less often because wild dogs avoid lions, but predation by lions is still a major cause of mortality for wild dogs in most ecosystems (Ginsberg et al. 1995; Mills & Gorman 1997).
11.7 Impact of Interspecific Competition Wild dogs have apparently never been a common species (Selous 1908; Malcolm 1979). Currently, even the highest wild dog densities are low relative to those of lions and hyenas (Table 11.1). The biomass of wild dogs is invariably 1–2 orders of magnitude lower than the biomass of lions or hy-
266 ▪ C H A P T E R 1 1
enas. This suggests that there are basic ecological reasons for wild dogs’ rarity. Although conflict with expanding agriculture and human populations are clearly crucial factors on the continental scale, it is important to identify the factors that limit wild dogs within large, intact ecosystems. The fact that wild dogs do poorly where hyenas and lions thrive (Figures 11.4 and 11.5) suggests that wild dogs may be limited by competition with larger, more abundant carnivores. In longitudinal data for Serengeti, wild dogs declined while hyenas and lions increased in response to growing prey populations. Highly overlapping diets suggest that exploitation competition explains the negative correlation between the densities of wild dogs and hyenas. The extent of competition between wild dogs and lions is less obvious, but it is clear that lions commonly kill wild dogs. Aggressive interactions at wild dog kills confirm that the frequency and impact of interference competition increases where hyena density is high, and where visibility is good. Quantitative differences in direct competition from hyenas are particularly striking in the comparison of Selous (where competition is mild and wild dogs attain their highest density) with Serengeti (where competition is intense and wild dog density is low and declining). The effects of competition are largely one way; wild dogs rarely scavenge food from lions or hyenas. Antagonistic interactions unrelated to food also suggest competition. Hyenas are manifestly strong and dangerous predators—twice the weight of wild dogs. That wild dogs should go out of their way to attack hyenas suggests that their presence is in some way harmful to the dogs. Hyenas have rarely been observed to kill wild dogs, but predation by lions accounts for up to 50% of known causes of death. Together, these data suggest that competition with spotted hyenas and predation by lions are factors that commonly limit wild dog populations. Ngorongoro provides an example in which wild dogs may have been locally excluded by larger competitors. Serengeti provides a possible example of the process of exclusion in action. We do not imply that factors other than competition (e.g., diseases) are unimportant. Interspecific competition is probably one element among several that cause wild dogs’ universally low population density. These elements are likely to interact. For example, it is logical that vulnerability to disease may interact with the level of competition. As competition increases, food stress may increase susceptibility, while increased rates of interaction with other carnivores may increase transmission.
11.8 Adaptations to Interspecific Competition Competition might explain four traits of wild dogs that are unusual among large social carnivores. First, wild dogs move unpredictably over very large
I N T E R S P E C I F I C C O M P E T I T I O N ▪ 267
ranges, relative to hyenas or lions (Frame et al. 1979; Fuller & Kat 1990; Chapter 3). Perhaps this is an adaptation for finding patches of prey not occupied by competitors. Combining the data in this chapter with our analyses of space use (Chapter 3), we suggest that wild dogs are a fugitive species, whose large home ranges provide a degree of spatiotemporal isolation from dominant competitors. Second, wild dogs’ loudest call (the “hoo”) is substantially quieter than those of lions or hyenas, and it is rarely used, particularly during denning intervals. This might be explained by the need to avoid attracting competitors. Third, wild dogs rarely remain with their own kills after they have eaten their fill, and they almost never return to partially eaten carcasses they have left. In Selous, wild dogs returned to only 1 of 404 kills in 310 days. Fourth, wild dogs rarely scavenge from other species. Both of these traits would reduce the likelihood of encountering other carnivores at a resource likely to be contested. Limitation by larger carnivores may be an issue of general importance in the conservation of medium-sized carnivores. For example, cheetahs (Acinonyx jubatus) in Serengeti are limited by poor recruitment, mostly due to predation on cubs. Laurenson (1995) found that 95% of cubs died before reaching independence, and that predation accounted for 73% of these deaths. Lions were responsible for most of the predation (79%), with hyenas responsible for 12%. Cheetahs are sometimes found at higher density outside protected areas than within, perhaps because predation pressure is less intense (Caro 1994; Durant 1998, 2000). Like wild dogs, the density of cheetahs is negatively correlated with the density of lions across ecosystems (Laurenson 1995). In North America, there is a chain of competitive interaction between wolves, coyotes, and foxes, with larger species limiting or excluding smaller species. Wolves are widely reported to kill coyotes (Paquet 1992; Peterson 1995). On a large geographic scale, coyotes are rare or absent in places where wolves occupy contiguous home ranges (Peterson 1995). On a smaller scale, coyotes and wolves use nonoverlapping home ranges in places where wolf ranges are patchy (Fuller & Keith 1981). Paquet (1992) suggests that coyotes are excluded by wolves unless the wolves’ prey is sufficiently large that coyotes can scavenge. Relations similar to those of the wolf and coyote have been described for other pairs of canids, including coyotes and red foxes (Sargeant et al. 1987; Sargeant & Allen 1989), coyotes and kit foxes (Ralls & White 1995), and culpeo zorros and gray zorros (Johnson et al. 1996). Similar results have been found throughout the Carnivora (Palomares & Caro 1999; Creel et al. 2001). The findings of this chapter emphasize that very basic ecological data on density and diet are relevant to conservation. While generalized population viability modeling is valuable, it is of most value when it incorporates eco-
268 ▪ C H A P T E R 1 1
logical processes and constraints. Wild dogs may best be conserved in areas in which lions and hyenas attain only moderate densities. A clear policy recommendation is that reintroductions of wild dogs, and perhaps other medium-sized carnivores, should avoid areas with high densities of larger carnivores.
12
Infectious Diseases
Among the ecological processes that might regulate a population, predation and competition have received considerable attention, because they are relatively easy to observe and quantify. Theory and data show that infectious diseases can also limit or regulate natural populations, although their effects are more difficult to document (Anderson & May 1979; Hudson et al. 1998). It has been widely suggested that wild dogs are unusually vulnerable to diseases, and that diseases commonly limit their numbers (Bere 1956; Schaller 1972; Fanshawe et al. 1991; Kat et al. 1995). For diseases transmitted by saliva or feces, the frequent social interactions within a pack put all pack members at risk if one becomes infected. Counterbalancing this effect, direct interactions between packs are rare in wild dog populations (one encounter per 40 days in Selous; Chapter 3), so the risk of transmission among packs is relatively low for diseases that run their course quickly (Mills 1993). A surprising amount of research has focused on the incidence and pathology of diseases in wild dogs (van Heerden 1986; Gascoyne et al. 1993; van Heerden et al. 1995; Creel et al. 1995; Alexander et al. 1996). Until recently, most information came from a single population, in the Serengeti National Park (SNP), where diseases were a serious and recurring problem. The SNP population declined to extinction between 1970 and 1991, with recurrent outbreaks of disease. Schaller (1972) described a fatal illness among the members of one pack that caused anorexia and weight loss, mucopurulent discharge from the eyes, staggering, and myoclonal twitching. Based on these clinical signs and postmortem examination of one carcass, Schaller ascribed mortality in three packs in 1967–1968 to canine distemper. Malcolm (1979) attributed a decline in 1971–1973 to disease, though no serological or postmortem data were collected to identify the pathogen, and it has been suggested that some of these dogs were actually shot (Burrows et al. 1994). In the late 1980s and early 1990s, the last few wild dogs in SNP disappeared during a rabies epizootic. Rabies viral encephalitis was confirmed in four cases by brain histology and by isolation of the rabies virus, which was a variant identical to that of domestic dogs neighboring the ecosystem (Gascoyne et al. 1993; Kat et al. 1995). Collectively, data from SNP have shown that infectious diseases can cause substantial mortality in wild dogs, and can contribute to a local extinction in a small population. This has led to the widespread conviction that wild dogs are “particularly sensitive to disease” (Fanshawe et al. 1991, p. 140), and that disease has played “a main role in the numerical and distributional decline of African wild dogs” (Kat et al. 1995, p. 229). However, little is known about the regulatory role of diseases in other wild dog populations.
270 ▪ C H A P T E R 1 2
The Serengeti population was unusual in several ways. Population density was very low (Frame et al. 1979; Malcolm 1979; Burrows et al. 1994), averaging ⬍1 dog/200 km2 over a 20-year period. The population held fewer than 50 dogs in all but 2 years during the 23 years leading up to their disappearance, and often numbered less than 30. Consequently, the population was quite vulnerable to local extinction by chance or deterministic decline (see Chapter 13). Competition from larger carnivores was intense (Frame & Frame 1981; Fanshawe and Fitzgibbon 1993), and probably limited wild dog numbers (Creel and Creel 1996). Finally, many carnivores that could harbor and transmit viral infections were common in the SNP ecosystem. Common species with documented exposure to canine distemper virus (CDV) and/or rabies virus included bat-eared foxes, Otocyon megalotis (Maas 1993), domestic dogs (Alexander & Appel 1994), jackals, Canis aureus and C. mesomelas (Alexander et al. 1994; Roelke-Parker et al. 1996), spotted hyenas, Crocuta crocuta (Alexander et al. 1995; Haas et al. 1996), and lions, Panthera leo (Roelke-Parker et al. 1996). While it is clear that infectious diseases can deliver the knockout punch to a population such as the one in SNP, it may be misleading to generalize conclusions from SNP about the broader role of disease in wild dog population dynamics. Despite close monitoring, disease-related declines were not observed in Kruger National Park (KNP, South Africa) over a period of 22 years, or in the Selous Game Reserve (SGR, Tanzania) over a period of 6 years (Reich 1981; Maddock & Mills 1994). Combining serological and demographic data, van Heerden and colleagues (1995) concluded that “disease could not be incriminated as an important cause of death” in KNP (p. 18). One possible explanation for this is that these populations are vulnerable to viral diseases, but have not been exposed. In KNP, extensive serological screening showed no evidence of exposure to rabies virus, CDV, or canine parvovirus (CPV). Bacillus anthracis, the bacterium causing anthrax, has caused several deaths in KNP, SGR, and in Luangwa, Zambia (Turnbull et al. 1991; Creel et al. 1995; van Heerden et al. 1995). All in all, the role of disease in regulating wild dog populations is still poorly understood. Epidemiological models show that a microparasite can regulate a host if the pathogen is highly virulent, acquired immunity is weak, or the host’s intrinsic rate of increase is low (Anderson & May 1979). At least two of these criteria (high virulence, low host rate of increase) characterize interactions between wild dogs and rabies or canine distemper. Epidemiologic models of pathogens with multiple hosts (like rabies, CDV, and CPV) show that a small, low-density population of one host is vulnerable to spillover transmission from a high-density host that supports an enzootic infection (Grenfell & Dobson 1995). Because wild dogs invariably live at low density, in contact with other hosts that attain much higher densities, spillover transmission is a serious concern. Thus, there are theoretical models and empirical data suggesting that dis-
I N F E C T I O U S D I S E A S E S ▪ 271
ease may modulate wild dog numbers in some ecosystems. Canine distemper and rabies have caused severe mortality and population declines in other carnivore species (Williams et al. 1988; Osterhaus et al. 1995; Roelke-Parker et al. 1996), and some authors suggest that infectious diseases have been a major factor in the continentwide decline of wild dogs (Kat et al. 1995; Alexander et al. 1996). In this chapter we review serological data from several populations, including Selous. We discuss data on the effects of pathogens in captivity and the wild, with emphasis on three pathogens known to kill adult wild dogs in the wild (rabies, canine distemper, and anthrax), and one that may kill pups (parvovirus).
12.1 Canine Distemper Virus Canine distemper virus (CDV) is a morbillivirus that affects all carnivores and is found throughout the world (Appel 1987). The virus is transmitted by inhalation, and cannot survive outside of a host for more than a few hours. Domestic dogs with acute CDV infection shed the virus in all their secretions, beginning roughly one week after exposure and continuing for up to 90 days (Appel 1987). Although CDV can cause high mortality in naive carnivores (Williams et al. 1988; Roelke-Parker et al. 1996), domestic dogs can recover from infection, and subsequently show effective cellular and humoral immunity (Appel 1987). CDV has never been isolated from wild dogs. No unvaccinated wild dog has been found to carry antibodies to CDV until recently, probably because little serological screening was done. Nonetheless, population declines have been tentatively attributed to CDV on the basis of behavioral observations (Schaller 1972; Malcolm 1979) or concurrent CDV outbreaks in other species (Alexander & Appel 1994; Roelke-Parker et al. 1996). Vaccine-induced canine distemper has caused deaths in captivity, confirming that infection is potentially fatal for wild dogs (McCormick 1983; van Heerden et al. 1989; Durchfield et al. 1990). In 1996, canine distemper was rigorously confirmed as the cause of death for a free-living wild dog for the first time (Alexander et al. 1996). Ten wild dogs (from a pack of 12) died over two weeks in Botswana’s Chobe National Park. One carcass was recovered, and pathology and immunohistochemistry revealed pneumonia and meningoencephalitis due to CDV as the cause of death. Several entire packs disappeared in 1996 near Chobe, in Moremi National Park, perhaps due to CDV (J. W. McNutt, personal communication). The similarity of the Chobe outbreak with earlier outbreaks in other ecosystems supports the conclusion that these were related to CDV. In a serological survey of 22 wild dogs in Selous (Creel et al. 1997c), 59% were seropositive for CDV, with 95% confidence limits (CI) of 43– 76% (Table 12.1). Despite intensive monitoring, we observed no wild dogs
272 ▪ C H A P T E R 1 2 Table 12.1 The proportion of adult and yearling wild dogs in the Selous Game Reserve (Tanzania) that were positive for antibodies to canine distemper, canine parvovirus, and rabies between 1992 and 1995a Year
CDV
CPV
Rabies
1992
5/8 (0.63) 2/5 (0.40) 1/4 (0.25) 5/5 (1.00) 13/22 (0.59)
7/8 (0.88) 3/5 (0.60) 3/4 (0.75) 2/5 (0.40) 15/22 (0.68)
0/7b (0.00) 0/5b (0.00) 0/1b (0.00) 0/0b (0.00) 0/13b (0.00)
1993 1994 1995 Total
a
Number seropositive/number sampled, with proportion in parentheses. If titers ⬍ 11 are considered negative for hemolyzed samples, sample sizes from 1992 to 1995 were 8, 5, 4, and 4, totaling 21.
b
with signs of distemper, and annual mortality rates were low in comparison to other populations for all age classes (see Chapter 7). Seropositive individuals were found in 10 of 12 packs, indicating that exposure to CDV occurs throughout northern Selous. Of 13 seropositive individuals, 5 (38%) were born during the study, and thus were known to have seroconverted while under observation. Given this, low mortality rates and the absence of any clinical signs of disease, it appears likely that the dogs seroconverted without serious illness. The proportion of individuals with antibodies to CDV did not vary significantly across four years, though low samples sizes mean that these are weak tests (Fisher’s exact test: 92 vs. 93, z ⳱ 0.15, P ⳱ 0.88; 93 vs. 94, z ⳱ 0.01, P ⳱ 0.99; 94 vs. 95, z ⳱ 0.95, P ⳱ 0.34). CDV titers were positively related to age (Figure 12.1: polynomial regression, r 2 ⳱ 0.56, P ⬍ 0.02), with no antibodies in dogs less than 1.9 years old. This suggests two possible explanations, which are not mutually exclusive. First, individuals that contract canine distemper before 2 years old might be more likely to die. Second, exposure to CDV may be limited in dogs younger than 2 years. CDV has poor survival outside of a host; it is transmitted by inhalation, which requires fairly direct contact (Appel 1987). If spotted hyenas are a source of infection (Alexander et al. 1996; Haas et al. 1996), yearlings may be exposed less frequently than adults, because they rarely fight with spotted hyenas at carcasses. Regardless, the age-distribution of seropositivity for CDV suggests that vaccination programs, if considered, should focus on young individuals. For 10 serosampled individuals that died during the study, postsampling
I N F E C T I O U S D I S E A S E S ▪ 273
Figure 12.1 Among wild dogs in Selous Game Reserve, CDV titers increase sharply at 2 years of age, and thereafter do not covary significantly with age. Solid line shows polynomial (third order) regression; dashed lines show 95% confidence belts for the regression.
survival averaged 638 Ⳳ 92 (SE) days, and did not correlate with CDV titer (least squares regression, P ⳱ 0.37). Although the population was under frequent observation, no wild dogs or other species were observed with clinical signs of canine distemper. Because 5 of 13 seropositive dogs were born during the study, we can exclude the possibility that seropositive dogs were simply the survivors of a CDV epidemic prior to the study. Among all sampled individuals (seropositive or negative), at least 24% were exposed to CDV during the study. Recent screening in southern wild dog populations suggests that the high seroprevalence of antibodies to CDV seen in Selous might not be unusual. All four dogs sampled in Hluhluwe-Umfolozi were CDV positive (South Africa), where one recent death was attributed to CDV on the basis of clinical signs (J. van Heerden, cited in Woodroffe et al. 1997). In the Tsumkwe district of Namibia, four of six wild dogs were CDV positive (K. Laurenson, cited in Woodroffe et al. 1997). High seroprevalences make it clear that CDV is not necessarily fatal for wild dogs, but we do not know why some populations maintain a high proportion of immune individuals and show no apparent illness, while other populations experience epizootics. Possibly, some ecosystems do not hold a sufficient density of other carnivores to form a reservoir, or perhaps interaction rates between wild dogs and other carnivores vary among ecosystems,
274 ▪ C H A P T E R 1 2
so that a high proportion of wild dogs remain na¨ıve, allowing periodic epidemics. In Selous, other canids were rare: We saw black-backed and sidestriped jackals 0–5 times a year, and never saw golden jackals or bat-eared foxes, although civets, and dwarf and banded mongooses were common. The densities of lions and spotted hyenas were moderate (Chapter 11). Domestic dogs can be an important reservoir for CDV and other viral diseases that affect wild carnivores (Cleaveland & Dye 1995; Roelke-Parker et al. 1996). We never saw a domestic dog in the reserve in six years of study. When driving in and out of the reserve for resupply, we kept records of domestic dogs along two inhabited roads to the reserve, for the 100 km closest to the reserve. We saw no dogs to the east of the reserve, and interviews with villagers (mostly Muslims, who rarely keep dogs) confirmed that dogs were not kept. We noted dogs in all villages to the north, and government veterinary staff estimated that 900 dogs lived in the seven villages nearest the reserve (from 5 km to 25 km away). In 1995, 14% of these dogs were vaccinated for rabies. In 1996, veterinary staff killed five domestic dogs that showed behavioral signs of distemper (twitching, mucopurulent discharge), together with 23 domestic dogs suspected of being infected. Variation among populations in CDV epidemiology might also arise if variants of CDV with different pathogenicity are found in different ecosystems. The virus has never been isolated and typed from a wild dog, so differences in its effects may be due to differences in the virus itself, rather than differences among populations in terms of ecological or epidemiologic conditions.
12.2 Rabies Virus Rabies is a rhabdovirus, transmitted in saliva, mainly by biting. The virus has a cosmopolitan distribution and affects all mammals, although carnivores are often the hosts that maintain enzootic rabies. In temperate areas, where domestic dogs and cats are generally vaccinated, foxes and raccoons are common reservoirs (Macdonald 1980). In Africa, domestic dogs commonly serve as a reservoir for rabies (Kennedy 1988; Wandeler 1993; Swanepoel et al. 1993; Cleaveland & Dye 1995), though herpestids, viverrids, mustelids, and wild canids (jackals) may also play this role (Foggin 1988; Swanepoel et al. 1993). Epizootics of rabies are not rare among African carnivores, and have been documented for black-backed and side-striped jackals (Kennedy 1988), spotted hyenas (Mills 1993; Swanepoel et al. 1993), Simien wolves (Sillero-Zubiri et al. 1996), bat-eared foxes (Maas 1993) and wild dogs (Gascoyne et al. 1993; Kat et al. 1995). A well-documented rabies outbreak killed most or all of the wild dogs in the Serengeti ecosystem between 1989 and 1991. In one pack in the north of
I N F E C T I O U S D I S E A S E S ▪ 275
the ecosystem (Masai Mara), 21 of 23 pack members died over six weeks in 1989. Of these, 18 individuals showed a set of behavioral symptoms identical to those of dumb rabies in domestic dogs, including loss of appetite, a stiff gait, ataxia, progressive paralysis, restlessness, soil eating, dehydration, and emaciation (Kat et al. 1995). Brain samples from three of these dogs were positive for rabies on the basis of fluorescent antibody tests, amplification of rabies viral RNA with specific primers, and the presence of Negri bodies (eosinophilic inclusions in the cytoplasm). Rabies viral RNA was amplified from the salivary gland of a fourth dog (Kat et al. 1995). One of the two wild dogs that survived this outbreak and two of those that died had been vaccinated with a killed-virus vaccine (Kat et al. 1995). By sequencing the nucleocapsid (300 base pairs) and glycoprotein (290 b. p.) genes of rabies virus isolated from wild dogs and domestic dogs, Kat et al. (1995) showed that the rabies strain that killed Kenyan wild dogs was identical to the rabies virus isolated from Kenyan domestic dogs, with 1–4% sequence divergence from rabies isolated from domestic dogs in Tanzania, Nigeria, and South Africa. In 1990 and 1991, all of the remaining wild dogs in the Masai Mara (4 packs) disappeared, concurrent with an outbreak of canine distemper in local domestic dogs (Alexander et al. 1994), though rabies might also have been involved. Two packs were known to live in Serengeti National Park in 1986. In June, all 13 adults of one pack died over a two-week period, together with their puppies (Scott 1991). Behavioral observations suggested that rabies was the cause, though blood and tissue samples did not confirm this. At the same time, rabies was confirmed in brain samples from sympatric bat-eared foxes, in which rabies caused 90% of known mortality in the late 1980s (Maas 1993). In 1990, a pack of 20 dogs in Serengeti disappeared after 1 adult showed clinical signs of rabies. Brain samples from a partially eaten wild dog carcass found in this pack’s home range tested positive for rabies by several tests (Gascoyne et al. 1993). After this pack died, only two known packs lived in Serengeti National Park. Given the immediate threat of local extinction, they were vaccinated against rabies with an inactivated vaccine, after a trial program showed that four wild dogs vaccinated in captivity seroconverted without adverse side effects (Gascoyne et al. 1993). All of the dogs vaccinated in Serengeti subsequently disappeared, probably dying due to disease, but none of the deaths were observed and no samples were collected, so little can be concluded. Epizootics of both rabies and distemper occurred in other Serengeti carnivores just before and after these dogs disappeared (Maas 1993; Roelke-Parker et al. 1996), so either virus, or both, might have played a role. Three of the dogs vaccinated in Serengeti had rabies serum neutralizing antibody titers of ⬎0.5 IU/ml prior to vaccination, suggesting that they had previously been exposed to rabies, though these titers might have been due to nonspecific neutralization (Gascoyne et al.
276 ▪ C H A P T E R 1 2
1993). It is not known whether these dogs had survived prior exposure to rabies, whether they were incubating an active infection, or whether their acquired immunity was effective. A pack of 11 wild dogs was introduced into Etosha National Park in 1990, and despite vaccination against rabies with an inactivated vaccine (Rabisan: J. L. Scheepers, personal communication 1996), at least 4 died of rabies, perhaps transmitted by a black-backed jackal they ate. These dogs were in poor condition as a result of their poor hunting ability (they were released from captivity), and this undoubtedly increased their vulnerability to disease. At least one member of the pack had been seen to interact with domestic dogs, actually hunting with them (Scheepers & Venzke 1995). Six other members of the pack were killed by lions, and it is possible that disease increased their vulnerability to predation, as in Selous, where two pups were killed by hyenas while suffering from anthrax. There is no evidence that wild dogs in Selous have been exposed to rabies, as all were seronegative (Table 12.1). The same is true for wild dogs in Kruger National Park (van Heerden et al. 1995). No wild dogs or other species in Selous were observed with clinical signs of rabies, despite six years of observation. Because rabies is normally acute and fatal, it might go undetected by serological survey, though high rabies seroprevalence has been reported in wild dogs in Serengeti (Gascoyne et al. 1993). Rabies is known to occur in domestic dogs in and around the city of Morogoro, less than 70 km from our study area (Magembe 1985). As we mentioned above, we never saw domesticated dogs to the east of the reserve, but approximately 900 dogs lived in seven villages within 25 km to the north, with 14% of these dogs vaccinated for rabies in 1995. The villages to the north of the reserve had seen considerable immigration by members of the Maasai tribe, together with their cattle, during the 1980s and 1990s. This increase in stock keeping has caused a substantial increase in the domestic dog population bordering the reserve. Given that domestic dogs are a reservoir for enzootic rabies in Africa (Kennedy 1988; Gascoyne et al. 1993), these domestic dogs constitute a threat to wild dogs in Selous. Mills (1993) noted that transmission of rabies between wild dog packs is unlikely, because rabies kills its victims so quickly that they are unlikely to encounter another pack while infectious. However, it took two months for rabies to work its way through the pack of 23 studied by Kat et al. (1995). Given this time frame and the rates of encounter between packs in Selous (1 encounter per 40 days; Chapter 3), transmission among packs is possible, though unlikely. This risk probably differs among populations, because encounter rates between packs were lower in Kruger than in Selous (Mills & Gorman 1997). Encounter rates have not been reported for Serengeti, but home ranges overlapped considerably, and Frame et al. (1979) did not seem to consider encounters highly unusual, stating “we never recognized clear avoidance of contact between packs. When packs met, one chased the other
I N F E C T I O U S D I S E A S E S ▪ 277
from the vicinity” (p. 228). In Selous, we did see clear avoidance in some cases, though we also saw cases in which one pack tracked another and made intentional contact (Chapter 3).
12.3 Anthrax Anthrax is caused by a bacterium, Bacillus anthracis, that is distributed globally and can apparently cause disease in all homeotherms (Edelstein et al. 1990). The spores of B. anthracis are resistant to environmental extremes in the soil, and become infectious upon exposure to the air. Ungulates are commonly infected by exposure to spores in the soil or water, while carnivores are infected by these routes or by eating infected prey (Turnbull et al. 1991). Among domesticated animals, pigs and domestic dogs are known to have some resistance to anthrax (Cristoph 1973; Edelstein et al. 1990). Relatively little is known about anthrax in wild dogs. In the Luangwa Valley of Zambia, five wild dog carcasses were found during an anthrax epizootic, and four of these tested positive for anthrax (Turnbull et al. 1991). Opportunistic sightings of wild dogs became less common during and after the outbreak, leading Turnbull et al. (1991) to suggest that anthrax caused a decline in wild dog numbers. Thus, with heavy exposure, anthrax might have serious consequences for wild dogs. On the other hand, a reintroduced pack in Namibia were seen to eat anthrax-positive zebra carcasses during an epidemic, but none developed signs of the disease (Scheepers & Venke 1995). During a period with several anthrax epidemics among ungulates in Kruger National Park, wild dogs seemed little affected (van Heerden et al. 1995; de Vos & Bryden 1996). Although the wild dogs were monitored closely during anthrax outbreaks in 1990, 1991, and 1993, and although 1538 anthrax-positive carcasses were collected, only three wild dogs were known to have died, and no other individuals were seen with clinical signs of infection (van Heerden et al. 1995). Of 12 wild dog serum samples screened for B. anthracis toxin between 1990 and 1993, none were positive, suggesting either that exposure was not common, or that infection was quickly fatal (van Heerden et al. 1995). Observations from Selous show that anthrax infection is not necessarily fatal for wild dogs (Creel et al. 1995). We observed an anthrax outbreak in one wild dog pack, in which 3 of 18 adults and 8 of 24 pups showed one or more signs of the disease. Of these, all of the adults and half the pups recovered fully within 10 days. Of the 4 pups that died, 2 were killed by anthrax directly, while 2 were killed by spotted hyenas after becoming separated from the pack while suffering from the disease. Anthrax was confirmed by autopsy and microscopic examination of blood from the two pups that died. Both carcasses hemorrhaged heavily through the nose within hours of
278 ▪ C H A P T E R 1 2
Figure 12.2 Manifestations of anthrax in wild dogs. (a) Grossly swollen submandibular lymph nodes, and swollen gums oozing blood. Bleeding was more profuse after death. (b) Red, swollen ulcers on the skin.
I N F E C T I O U S D I S E A S E S ▪ 279
dying. Upon postmortem examination, we found extensive subcutaneous hematoma in the facial, submandibular, and thoracic regions, and the thoracic and abdominal cavities were partially filled with poorly clotted blood (Figure 12.2). The spleen was swollen and congested, and the lungs were edematous and congested. All the superficial lymph nodes were very markedly enlarged; the submandibular lymph nodes were so swollen that the swelling was obvious in the field, before the dogs died (Creel et al. 1995). In infected dogs, clinical signs included lethargy, loss of appetite, lying in direct sunlight despite temperatures of 33–35⬚C, swelling around the eye orbits, and creamy white ocular discharge. The most obvious sign was severe swelling of the lips, gums, and submandibular lymph nodes. Some dogs developed necrotic lesions on the gums, and some passed watery diarrhea with opaque white mucus. The two pups that died experienced difficulty breathing in their final hours. Many of those that recovered (and some dogs that showed no obvious signs of infection) developed raised lesions on their muzzles that superficially looked like warts. These lesions waxed and waned for several years afterward in some individuals, but disappeared completely for others. While our observations in Selous show that wild dogs (like domestic dogs) possess a degree of resistance to anthrax, the general threat posed by anthrax remains unknown. Observations in the Luangwa Valley suggest that anthrax might cause significant mortality if exposure is heavy. However, it seems notable that wild dog die-offs have not been noted during anthrax outbreaks among ungulates in other populations. Data from Kruger in the early 1990s clearly show that even under conditions likely to cause heavy exposure to anthrax, wild dogs do not necessarily die. Data from one pack in Selous confirm this, but also show that anthrax can kill wild dogs, especially pups.
12.4 Canine Parvovirus Canine parvovirus (CPV) emerged in the 1970s, but is now found in domestic dogs throughout the world. The virus is shed in the feces, and survives well outside a host. Infection occurs by eating contaminated feces or soil. In domestic dogs, CPV rarely kills adults but commonly kills pups, who die of the enteritis and diarrhea induced by viral replication in the intestinal epithelium. There is no direct evidence that CPV causes mortality in free-living wild dogs, although some populations are exposed (Fuller et al. 1992a), and it is reasonable to hypothesize that CPV is a factor in some pup deaths. CPV can cause substantial mortality in young wolves (Peterson & Krumenaker 1989; Johnson et al. 1994). For wolves, CPV has not been shown to cause population declines, but may hinder recovery after a decline (Mech & Goyal 1993).
280 ▪ C H A P T E R 1 2
Figure 12.3 Variation in mean CPV titer and mean litter size over four years for Selous wild dogs.
Because the carcasses of wild dog pups are virtually impossible to collect, it is almost certain that some early juvenile mortality goes undetected, and it is plausible that some juvenile mortality is due to undetected CPV (among other factors). Pup mortality can be substantial for wild dogs, with annual mortality as high as 70% in Kruger (van Heerden et al. 1995). No data exist to establish what proportion of these deaths is caused by CPV. Of 22 Selous wild dogs evaluated for antibodies to CPV, 68% were seropositive (Table 12.1: 95% CI ⳱ 54–81%). Seropositives came from 10 of 11 packs sampled, showing that exposure to CPV is widespread. CPV titers were not significantly related to age (least squares regression, F1,20 ⳱ 0.25, P ⳱ 0.62), and did not vary significantly among calendar years (Table 12.1: Fisher’s exact test, all interyear pairs not significant, but tests have low power). Pups occasionally had diarrhea, and three pups that were lethargic disappeared. The effect of parvovirus (if any) is most likely to be detected through high juvenile mortality or small litters at first count. Across years, mean litter size dropped by 38% in 1994, when the mean CPV titer peaked (Figure 12.3; regression of mean litter size on mean CPV titer, F1,3 ⳱ 2.06, P ⳱ 0.29, R 2 ⳱ 0.51). Comparisons with two other populations can test for effects of CPV on juvenile survival. In Kruger, 0% of 43 wild dogs were seropositive for CPV, litters were large at three months (11.9 pups), and juvenile survival was
I N F E C T I O U S D I S E A S E S ▪ 281
29.6% (Maddock & Mills 1994; van Heerden et al. 1995). In Serengeti, 67% of 6 individuals were seropositive for CPV, litters averaged 10.0 pups at first count, and juvenile survival was 39% (Fuller et al. 1992a; Burrows et al. 1994). In Selous, 68% of 22 wild dogs were seropositive for CPV, litter size was 8.6 Ⳳ 0.7 (n ⳱ 27), and juvenile survival was 69% (n ⳱ 127). These data suggest that exposure to CPV might reduce litters before their emergence from the den (Figure 12.4a) but that it does not cause significant mortality after the denning period (Figure 12.4b). Nonetheless, high CPV seroprevalence was not associated with low recruitment of yearlings (Figure 12.4c). Because factors other than CPV are likely to cause mortality during the denning period, these data are not conclusive, but suggest that CPV and pup survival should be examined further. Direct data on the cause of pup deaths would be useful, but difficult to obtain.
12.5 Other Pathogens Viruses Wild dogs are known from serological surveys to be exposed in some ecosystems to a range of viruses that apparently have little or no effect on their survival (van Heerden et al. 1995; Woodroffe et al. 1997). Based on their effects in domestic dogs, canine coronavirus, canine herpes virus, canine para-influenza virus (kennel cough), and canine adenovirus (canine hepatitis) might cause some deaths among wild dog puppies, acting alone or in tandem with other pathogens. Wild dogs are sometimes seropositive for viruses that probably have no effect on their heath, namely reovirus type 3, rotavirus, African horse sickness, and bluetongue (van Heerden et al. 1995; Alexander et al. 1994; 1995). Bacteria Neitz & Thomas (1938) described an outbreak of rickettsial disease, due to Ehrlichia canis, among domestic dogs living in Kruger National Park in the late 1920s and early 1930s. Wild dog numbers declined in Kruger at this time, and sick wild dogs were observed, but the disease was not rigorously confirmed in wild dogs (Stevenson-Hamilton 1939; van Heerden et al. 1995). Experimental infections in captive wild dogs show that Ehrlichia induces severe diarrhea, weakness, and anorexia, and reduces leukocyte, thrombocyte, and hemoglobin levels over 2–3 weeks (van Heerden 1979a). Experimental infections were self-limiting, and symptoms were less severe in wild dogs than in domestic dogs, but it seems likely that a wild dog suffering from ehrlichiosis would be more vulnerable to other threats. The infections described by van Heerden (1979a) were induced by injecting wild dogs intravenously with 2 ml of blood from an infected domestic dog, and it
282 ▪ C H A P T E R 1 2
Figure 12.4 Litter size at first count and juvenile survival in relation to seroprevalence of canine parvovirus in three wild dog populations: (a) litter size, (b) juvenile survival, (c) litter size at one year.
I N F E C T I O U S D I S E A S E S ▪ 283
remains unknown whether wild dogs contract ehrlichiosis from its normal tick vector in the wild. Screening in Kruger (n ⳱ 29) and the Masai Mara (n ⳱ 16) showed that no wild dogs were seropositive for E. canis. As van Heerden et al. (1995) point out, the only known vector for E. canis is Rhipicephalus sanguineus, and this tick has not been found on free-living wild dogs. The occurrence of ehrlichiosis in wild populations remains moot. Serological screening in Kruger showed that most wild dogs (27 of 29) were exposed to spotted fever, transmitted by the tick-borne Rickettsia conori/africae. Domestic dogs develop no clinical symptoms of spotted fever, and there is no known effect on wild dogs. Eight wild dogs (of 29) in Kruger were seropositive for exposure to Coxiella burnetti, which causes Q fever in humans but rarely causes illness in other species (van Heerden et al. 1995). Protozoa Wild dogs have been found to carry infections of Hepatozoon spp. and Babesia canis (van Heerden et al. 1995; Pierce et al. 1995). Babesia generally induces no overt illness in captive wild dogs, although most are infected (van Heerden 1980), and one juvenile wild dog died of acute babesiosis in captivity (Colly & Nesbit 1992). Hepatozoon usually induces no clinical signs of disease in domestic dogs, and no effects on wild dogs have been observed, although 82% of wild dogs in Serengeti and 93% of wild dogs in Kruger were exposed. Van Heerden et al. (1995) suggest that Hepatozoon may act opportunistically to cause localized inflammation. All in all, it appears unlikely that babesiosis or hepatozoonosis pose serious threats to the health of wild dogs. There is some evidence that Toxoplasma gondii has an important but intermittent effect on the survival of pups (van Heerden et al. 1995). Of 16 wild dogs screened for exposure in Kruger, all were seropositive, and 5 individuals are known to have died of infection by Toxoplasma or Neospora caninum (a closely related protozoan). An entire litter of 20 pups in Kruger disappeared simultaneously, and 4 of these were confirmed to have died of Toxoplasma and/or Neospora (Mills & van Heerden, cited by Woodroffe et al. 1997). Macroparasites endoparasites Wild dogs in Selous were commonly infected with cestodes and trematodes. While collecting fecal samples for other purposes, we noticed that trematodes and cestode proglottids were more common in the wet season, though we have no systematic data to demonstrate this. The dogs passing these worms were normally vigorous and apparently in good health. In Kruger, Taenia was found in all 14 fecal samples, Toxascaris canis were
284 ▪ C H A P T E R 1 2
found in 1, and Ancylostoma caninum hookworms were found in 11. Ancylostoma caninum has also been found in wild dogs from Hwange, the Masai Mara, and Moremi (Spangenberg & Ginsberg cited by Woodroffe et al. 1997). In captivity, wild dogs have been diagnosed with the cestode Diplydium caninum and the microfilarial worm Dipetalonema reconditum (van Heerden 1986). ectoparasites Most of the wild dogs we handled in Selous carried at least a few ticks. We systematically counted tick loads for 14 dogs, which carried between 0 and 248 ticks. The upper limit was from a dog with a mass of pepper ticks in one ear. The mean was 37 Ⳳ 17.8 ticks, with no obvious seasonal pattern in our limited data. In Kruger, van Heerden et al. (1995) found that all dogs were infested with ticks of the following species: Haemaphysalis leachi, Amblyomma hebraeum, A. marmoreum, Boophilus decoloratus, Rhipicephalus simus, R. evertsi, R. appendicularis, and R. zambesiensis. Flea infestations are apparently not common for wild dogs. In captivity, mild infestations of Ctenocephalides felis damarensis and Hippobosca longipennis and heavy infestations of Echidnophaga larina have been recorded (van Heerden 1986). Ctenocephalides and Echidnophaga larina were also found on wild dogs in Kruger. Echidnophaga larina caused hair loss in captivity. Bare patches were seen on 16% to 29% of wild dogs in Kruger (van Heerden et al. 1995), and were also fairly common in Selous. Most of the dogs in one Selous pack showed extensive hair loss and weepy abscesses that resembled mange in domestic dogs. None were obviously affected by the condition.
12.6 Behavior and Epidemiology Four facets of wild dogs’ behavior are likely to affect the likelihood of disease transmission within and among packs. First, wild dogs engage in a social rally just before hunting or when rejoining packmates after a separation (Figure 4.3a). During rallies, licking at each other’s mouths is common. Puppies and babysitters also lick the mouths of returning hunters, to induce them to regurgitate meat. These behaviors obviously increase the risk that a saliva-transmissible disease such as rabies or canine distemper will pass among packmates. Wild dogs also commonly sniff their own feces and that of other carnivores, increasing the risk of contracting a fecally transmitted disease such as parvovirus or worms. Second, wild dogs maintain very large home ranges, so that they are unlikely to encounter other packs, even though ranges overlap substantially (Chapter 3; Mills & Gorman 1997). In Selous, we observed one interpack encounter per 40 days (Chapter 3). In comparison, spotted hyenas can en-
I N F E C T I O U S D I S E A S E S ▪ 285
counter individuals from other clans on a daily basis (Kruuk 1972; Hofer & East 1995). Low encounter rates decrease the risk of a disease sweeping through an entire population (Mills 1993), but only if intraspecific transmission is the predominant route of infection. If spillover infection from other species is the major route of infection, then low rates of interaction between wild dog packs would have little effect on the spread of a disease. Third, wild dogs sometimes care for injured packmates by returning to them after a hunt, feeding them, or defending their access to a carcass (Estes & Goddard 1967; personal observations). If the same solicitude were shown to sick packmates, it would increase the risk that diseases would pass from sick to healthy packmates. In Selous, we found that sick pack members were treated much like any other individual, though healthy dogs seemed nervous when approaching packmates that were nearing death, for example bobbing their heads and whining. During the anthrax outbreak described above, the pack did not decrease its daily travel distance and did not double back for sick individuals, which were abandoned without apparent distress (Creel et al. 1995). The pack simply stood up and moved away without them. These observations were surprising, because their willingness to abandon packmates that were too sick to follow is in marked contrast to the behavior of a pack that has been scattered while hunting. Scattered hunters invariably search and hoo-call until all pack members are reunited; we never observed a pack that failed to reunite within an hour after a hunt, usually much less. In Etosha, Scheepers & Venzke (1995) noted that “injured dogs had to keep up with the pack, or they were abandoned” (p. 140). In the Masai Mara, Kat et al. (1995) noted that wild dogs attacked packmates that were sick with rabies, driving one from the pack shortly before his death. They hypothesized that the inability of sick dogs to respond normally in social interactions provoked aggression. Although rabid dogs did not initiate fights, they did bite packmates once attacked, presumably transmitting the virus. Speculatively, perhaps wild dogs distinguish between injury and illness, providing care to injured packmates but not to sick ones. Because hunting success and reproductive success are positively related to group size, it is not surprising that wild dogs would seek to help injured packmates, but it is interesting to note that they seem not to care for the sick. Fourth, wild dogs occasionally interact with domestic dogs, which are a major reservoir for viral diseases in Africa (Kennedy 1988; Swanepoel et al. 1993). Stevenson-Hamilton (1947) reported that packs of domestic dogs attacked and killed wild dogs in “plenty of instances,” but also suggested that “a party of hunting dogs would assuredly chase and immolate any single domestic canine” (p. 234). While this might be correct, we are not aware of any observations of wild dogs attacking domestic dogs. Grzimek (1970) reported an interaction between a pack of wild dogs and a pack of domestic dogs, in which the two groups approached and sniffed one another then walked away without aggression. Butynski (1974) described a similarly non-
286 ▪ C H A P T E R 1 2
aggressive encounter between a single wild dog and two domestic dogs. Scheepers & Venzke (1995) noted that a captive-reared wild dog escaped and formed a pack with two domestic dogs and hunted with them until it was retrapped. When the wild dog was reintroduced to its packmates, they killed it. Whether the interactions are aggressive or not, interactions between wild and domestic dogs increase the risk of disease for wild dogs living in areas bordering those where domestic dogs are common.
12.7 Impact of Diseases on Population Dynamics and Density The extinction of wild dogs in the Serengeti–Maasai Mara ecosystem shows that rabies or canine distemper can play a role in the final decline of a small population (Gascoyne et al. 1993; Kat et al. 1995). It is also clear that diseases were a recurring problem over the 25 years preceding the Serengeti dogs’ final crash (Schaller 1972; Malcolm 1979; Scott 1991; Gascoyne et al. 1993). It is less clear that diseases alone would have driven the Serengeti dogs to extinction, in the absence of interspecific competition. Studies of Serengeti wild dogs consistently noted intense competition from spotted hyenas at kills, over a 25-year period in which hyena numbers more than doubled (Frame 1985; Fanshawe & Fitzgibbon 1993; Hofer & East 1995). Lion numbers also increased substantially in this period (Hanby et al. 1995). More than one ecological factor or an interaction between factors (e.g., disease and competition) may have caused the decline, and the force that gave the final push need not be the same force that kept numbers low in the preceding decades. Rabies and canine distemper epizootics affected many carnivores in the ecosystem, and it is not surprising that the combination of low numbers and frequent exposure to viral diseases eventually caused a local extinction (Sinclair 1995). Perhaps the safest conclusion is that interspecific competition and disease (and perhaps other unidentified factors) can act together to cause a local extinction. There are other possible examples of disease-driven population declines. Wild dog sightings began to decline in Kruger around 1920, then dipped dramatically in 1928, coincident with an outbreak of ehrlichiosis in domestic dogs (Neitz & Thomas 1938; Stevenson-Hamilton 1939; 1947). Numbers remained low until a recovery began in the 1940s (Stevenson-Hamilton 1947). While it is possible that ehrlichiosis caused a decline, it is possible that another disease was involved, particularly because no free-living wild dog has been found to carry antibodies to Ehrlichia canis. Perhaps poisoning and shooting also played a role in the decline (see descriptions of culling by Stevenson-Hamilton [1947]). Woodroffe et al. (1997) note that wild dog numbers may have declined during a rabies epidemic in northwestern Zimbabwe. This is certainly possible, as wild dog numbers were estimated to decline by 34% between 1976
I N F E C T I O U S D I S E A S E S ▪ 287
and 1985 (Childes 1988), and an epidemic of rabies passed through jackals and domestic dogs between 1980 and 1983 (Kennedy 1988). However, the period over which the decline was measured is considerably longer than the period of the rabies outbreak, and it is not known what portion of the wild dog decline occurred between 1980 and 1983. More important, the estimated population in 1976 (600 wild dogs) was based largely on an estimate of 400 individuals in Hwange/Matetsi/Victoria Falls (Rushworth 1973). Of this estimate, Childes (1988) states: “Rushworth’s estimates are very questionable as I was unable to trace more than 8 sighting records for the years 1973–76. The present population in Hwange is estimated at 100–150 animals. . . . The apparent decrease in Lycaon numbers from 1976 to 1985 may not be a true reflection of the population trend, although the decrease in mean group size from 1980–85 does suggest a decline” (p. 310). Overall, the evidence for a disease-driven decline in Zimbabwe is weak. Wild dogs in Luangwa probably declined during a general outbreak of anthrax, but there are no data to directly document a reduction in wild dog numbers (Turnbull et al. 1991). In 1996, canine distemper was confirmed as the cause of death for one pack in Chobe National Park in northern Botswana, and several packs disappeared nearby in Moremi National Park, with CDV suspected as the cause (J. W. McNutt, personal communication). Considerable evidence suggests that disease causes occasional declines in wild dog numbers, but there are few well-documented examples. Sometimes the evidence that a decline occurred is weak; sometimes, the evidence that disease caused the decline is weak. This certainly does not mean that diseases can be ignored. Theory suggests that small populations vulnerable to diseases with multiple hosts will be vulnerable to extinction (Anderson & May 1979; Grenfell & Dobson 1995). On the other hand, current data from Selous and Kruger do not show that wild dogs are unusually affected by diseases. Wild dogs do not seem to be unusually vulnerable to epidemics in any mechanistic sense, but they may be unusually vulnerable to local extinction by epidemics simply due to their small population sizes. In this regard, wild dogs’ vulnerability to disease does not differ from their vulnerability to other limiting factors, such as human impacts or interspecific competition.
13
Extinction Risk and Conservation
The ecology of large carnivores presents difficult problems for conservation. Almost by definition, conflicts with humans and livestock are a serious issue for animals that are large and carnivorous. Real or potential conflicts with human activities restrict large carnivores to reserves and adjacent areas in much of the world. Such reserves must be large and ecologically intact to accommodate large carnivores, because they attain very low population densities relative to smaller species or those at lower trophic levels (Blackburn & Gaston 1994), and because some large carnivores range widely (Woodroffe & Ginsberg 1998). Wild dogs illustrate these problems clearly. They formerly inhabited most of sub-Saharan Africa with the exception of desert and rainforest, but now have a patchy distribution and are largely confined to protected areas (Fanshawe et al. 1991, 1997). Historically, wild dogs were never known to attain high densities (Selous 1908), and population densities now vary from about 5 adults/1,000 km2 (Serengeti National Park) to 40 adults/1,000 km2 (Selous Game Reserve). Densities of 17–20 adults/1,000 km2 are typical for healthy populations in large parks dominated by open woodland (Fuller et al. 1992a; Maddock & Mills 1994). A park of 10,000 km2 is likely to hold less than 150 wild dogs, and field studies have documented only a handful of populations larger than 100 adults (Maddock & Mills 1994; McNutt 1995; Fanshawe et al. 1997). The largest remaining population, in Selous, is estimated to hold 880 adults and yearlings (Creel & Creel 1998). Certainly, human impacts are among the prime reasons that wild dogs are experiencing range contraction. However, humans have similar impacts on other sympatric large carnivores such as lions and spotted hyenas, and these species often persist in the face of human impacts where wild dogs do not. This observation suggests that some aspect of wild dog ecology plays an important role in their vulnerability to local extinction. A well-studied example illustrates the point. In Serengeti National Park, lions (Panthera leo) attained densities of 140 adults/1,000 km2 and spotted hyenas (Crocuta crocuta) attained densities greater than 1,000 adults/1,000 km2, but African wild dogs (Lycaon pictus) never exceeded 15 adults/1,000 km2 (Packer 1990; Hofer & East 1993a, b; Frame et al. 1979; Burrows 1995). Several lines of evidence make it unlikely that these differences in density are due to differences in the impact of humans. Hyenas, lions, and wild dogs were all killed under predator control operations prior to the establishment of the National Park in 1951. No data tell how many of each spe-
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 289
cies were shot in Serengeti, but data from Kruger illustrate the general point that all carnivores were commonly killed by game wardens in the first half of the 20th century. Between 1903 and 1927 (when Kruger National Park was established), wardens shot a minimum of 1,272 lions, 1,142 wild dogs, 660 leopards, 521 hyenas, and 269 cheetahs (Smuts 1982). In the Serengeti region, lions were also hunted by the Maasai tribe, and were hunted by tourists over most of the current park’s area until 1959, when the park was enlarged (Turner 1987). Lions and hyenas are still hunted by tourists in game reserves bordering Serengeti, but wild dogs are not (Leader-Williams et al. 1996). Overall, the intentional impact of humans on wild dogs in Serengeti appears similar to their impact on lions and hyenas. Wild dogs have large home ranges, and thus might be incidentally killed when they leave the park (Frame et al. 1979; Woodroffe & Ginsberg 1998). However, hyenas with home ranges near the center of the park also leave the park frequently (22% of 50 wet-season fixes; Hofer & East 1993a, b), where they are frequently killed by snares (Hofer & East 1995). Lions at the center of the park are rarely snared, but snaring is thought to be common for lions with ranges near the park border (C. Packer, personal communication; M. Turner in Schaller 1972). In contrast, no wild dogs are reported to have died by snaring in Serengeti (Arcese et al. 1995; Burrows 1995), though they are snared elsewhere. Wild dogs rarely scavenge (Kruuk 1972; Creel & Creel 1995a), so they are weakly attracted to ungulate carcasses or the remains of slaughtering at snare lines. By contrast, spotted hyenas scavenge 20–50% of their meals and lions scavenge 10–30% of their meals (Gasaway et al. 1991; Schaller 1972; Packer 1986), so their attraction to snare lines may be stronger. While snares are dangerous for all carnivores and can be a serious problem in some locations (Rasmussen 1998), their impact on wild dogs is not unusually strong. Wild dogs are killed by viral diseases, including rabies and canine distemper, and domestic dogs are thought to be a significant reservoir from which viral diseases spill over to African wildlife (Roelke-Parker et al. 1996), so viral epidemics could be considered an indirect human impact on protected carnivore populations. In some wild dog populations, mortality due to infectious diseases is severe (Gascoyne et al. 1993; Alexander & Appel 1994). In other populations, infectious diseases have little impact on mortality (van Heerden et al. 1995; Creel et al. 1997c). As with other human impacts, spotted hyenas and lions are also vulnerable to viral diseases, and mortality can be severe. In a 15-year study of spotted hyenas in Kalahari, 43% of mortality was attributed to rabies (Mills 1990). In Serengeti, canine distemper virus killed approximately 30% of the lion population in a period of nine months, with substantial mortality in the hyena population (Alexander et al. 1996; Roelke-Parker et al. 1996). To return to our original point, measurable human impacts on wild dogs in Serengeti were similar to human impacts on the other carnivores, but wild
290 ▪ C H A P T E R 1 3
dogs were always 10 to 100 times less common than lions and hyenas, and ultimately disappeared. Similar patterns of population density have been found everywhere that the three species have been studied: Lions outnumber wild dogs 3-fold to 21-fold, and spotted hyenas outnumber wild dogs 8-fold to 122-fold (Creel & Creel 1996). For reasons that have to do with ecology, wild dogs are always less common than sympatric large carnivores, and this makes wild dogs prone to local extinction. A focus of this chapter is to quantify the effect on extinction risk of ecological factors known to limit wild dog numbers. Because wild dog conservation depends on the likelihood that populations of a few hundred individuals will persist, quantitative assessments of factors affecting extinction risk are important. Field studies have identified factors that affect the density and demography of wild dogs, including habitat loss, human-caused mortality, infectious diseases, and competition with lions and hyenas (Kruuk 1972; Malcolm 1979; Gascoyne et al. 1993; Creel et al. 1995; van Heerden et al. 1995; Creel & Creel 1996; Mills & Gorman 1997; Gorman et al. 1998). Some aspects of human impact on wild dog populations are at least conceptually well understood. For example, habitat loss reduces population size and human-caused mortality is likely to alter survivorship, both of which would increase the risk of extinction. In contrast, the relationship between extinction risk and ecological factors, such as interspecific competition, is not well understood (Vucetich et al. 1997). In this chapter, we use demographic data from Selous to construct a Leslie matrix and determine the deterministic population growth rate via Eigen analysis. We then use the Leslie matrix to make stochastic population projections for an assessment of extinction risk due to demographic stochasticity alone. We also use the Leslie matrix to examine effects on population growth of potential changes in demography. We then change approaches and use a stochastic, individual-based model developed in collaboration with John Vucetich (who programmed the model in CⳭⳭ) to investigate how ecological factors affect extinction risk for a wild dog population with the demography seen in Selous (Chapter 7).
13.1 Analysis of Extinction Risk with Leslie Matrix Projections With the demographic data from Chapter 7, we constructed a Leslie matrix and used it in deterministic and stochastic population projections to assess extinction risks (Caswell 2000; de Kroon et al. 2000). The population growth rate, λ, estimated by the dominant eigenvalue of the Leslie matrix, ranged from 1.038 to 1.045, depending on the method we used to measure age-specific fecundity. We conducted stochastic population projections of density-independent growth, using the fecundity schedule that gave the
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 291
lower growth rate. In the stochastic projections, reproduction by dogs in each age class at annual time steps was determined by random draws of integers from a normal distribution with a mean set equal to the observed mean number of pups born per female in that age class, and variance set equal to the observed variance in litter size pooled across all ages. The survival of dogs in each age class was determined by a draw from a binomial distribution with a rate equal to the mean annual survival for females in that age class, and a sample size determined by tracking the number of dogs in each age class at each time step. We set the initial number of dogs in each age class using the age distribution observed in Selous, rather than the stable age distribution derived from the Leslie matrix. With these procedures, our projections include demographic stochasticity but exclude environmental stochasticity. That is, age-specific survival and reproduction have variances, but their means do not change through time. Our focus here is to estimate the risk of extinction due to demographic stochasticity alone and to evaluate the likelihood that the population will persist if conditions stay as good as they are. To assess the effects of variation in the environment (diseases, competitors), we use individual-based simulations described later in the chapter. As expected (Dennis et al. 1991), the population growth rate with demographic stochasticity was lower than the deterministic growth rate. With demographic stochasticity and an initial population size of 880, one thousand projections over 200 years gave a mean stochastic growth rate of λ ⳱ 1.001. Examples of population trajectories are shown in Figure 13.1, and trajectories for each age class in a typical projection are shown in Figure 13.2. Changes in mean population size with 50%, 75%, 90%, and 95% bootstrap confidence limits from 100 projections are shown in Figure 13.3. With the demography of wild dogs in Selous (λ ⳱ 1.034), the projections show no tendency for mean population size to change, but there is substantial variation in population size through time. At 100 years, the 95% confidence interval for the population’s size runs from 586 to 1,216 dogs. These results have several implications for long-term monitoring and active conservation. First, despite environmental conditions that yield fairly strong deterministic growth (3.4% annually), the variation in demographic rates is great enough that the population is not predicted to grow. (See Dennis et al. 1991 for an explanation of the reason that variance causes the stochastic growth rate to be less than the deterministic rate.) This result suggests that sustained, strongly positive deterministic growth is needed for a wild dog population to expand over the long term. Second, changes in population size up to 28% fall within the 95% confidence interval, which means that it will often be difficult to distinguish a deterministic decline from demographic stochasticity. This is important because active management can reverse a deterministic decline, but cannot realistically eliminate demographic stochasticity. Of course, when managing an endangered species, 95% might not be ap-
292 ▪ C H A P T E R 1 3
Figure 13.1 Examples of population trajectories from stochastic projections of a population of 880 dogs with means and variances in age-specific rates of survival and reproduction from Selous. The means for demographic variables do not change, so the variance observed here is purely from demographic stochasticity.
propriate as a level of confidence for accepting that a population is in deterministic decline: One could make the argument that we should assume that a decline is deterministic until proven otherwise. Based on this argument, we show less stringent confidence limits in Figure 13.2. To have 50% confidence that a deterministic decline is occurring, the population must drop by 13%. Smaller Populations With the demography of the Selous population, stochastic projections over 200 years suggest that populations of 500 or 100 dogs have a very low probability of extinction due to demographic stochasticity alone. In 100 iterations for each of these population sizes, the probability of extinction was zero. Altered Demography: Projected Effects of a Decline in Pack Size Through cooperative hunting, cooperative defense and communal pup rearing, both survival and reproduction are affected by pack size (Chapter 7). Because pack size has such wide-reaching effects on wild dogs’ fitness
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 293
Figure 13.2 Means and bootstrap 95% confidence limits from 1,000 iterations of a stochastic population projection for a population of 880 dogs with means and variances in age-specific rates of survival and reproduction from Selous.
(Chapters 4–6, 9–10), we used the Leslie matrix to examine the consequences for population growth of changes in pack size. The age-specific rates of survival and reproduction that we observed are dependent on the distribution of pack sizes in the population during our study, but we can use the regressions of reproductive success and annual survival on pack size to explore how a shift in mean pack size would alter λ. To model changes in fecundity, we used the linear regression of reproductive success on pack size (Chapter 7), which shows that breeding females’ fecundity increased 7.87% with the addition of one adult pack member. To model changes in survival, we used the logistic regressions of annual survival rates on pack size, with separate regressions for pups, yearlings, and adults (all age classes from two years and up). These regressions showed that pup survival improved substantially as pack size increased (Ⳮ0.19), yearlings were essentially unaffected (Ⳮ0.001), and adult survival declined (ⳮ0.05). Using the amounts just explained, we calculated the deterministic population growth rate expected for a population with mean pack size greater or less than the mean pack size in Selous (8.3 adults). Interestingly, plotting population growth against pack size shows an intermediate optimum with its peak at a pack size one larger than the observed mean (Figure 13.4). This
294 ▪ C H A P T E R 1 3
Figure 13.3 Example of the age-distribution in a stochastic projection for a population of 100 dogs with means and variances in age-specific rates of survival and reproduction from Selous. The means for demographic variables do not change, so the variance observed here is purely from demographic stochasticity.
suggests that the observed distribution of pack sizes is nearly optimal from the perspective of maximizing population growth, thus maximizing individual fitness (Fisher 1954). It is not surprising that a decrease in pack size would reduce the population’s growth rate, but a priori it is less obvious that an increase in pack size would have the same effect. It should be noted,
Figure 13.4 The effect of changes in pack size on the deterministic rate of population growth, , from Eigen analysis of attendant changes in the Leslie matrix.
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 295
however, that our individual-based model makes different predictions about the effect of pack size (see below).
13.2 Stochastic Individual-Based Modeling of Extinction Risk With John Vucetich, we developed an individual-based model of wild dog population dynamics, which John programmed in CⳭⳭ (Vucetich & Creel 1999). The model’s logical structure (e.g., pack membership, rules for dispersal), causal relationships (e.g., pack size and reproduction), and parameter values (e.g., age- and sex-specific survival) are based on demographic and ecological data from our study population. Lion density correlates negatively with wild dog population density (Creel & Creel 1996), and lion competition and predation are likely to be important components of extinction risk for wild dogs (Mills & Gorman 1997), so we structured the model to focus on the effects of interspecific competition with lions. Because infectious diseases may also affect extinction risk (Malcolm 1979; Gascoyne et al. 1993; Creel et al. 1995; Woodroffe et al. 1997), we also analyzed the sensitivity of extinction risk to scenarios that mimic the effects of infectious disease. Parameter Estimation We obtained means and variances for model parameters from the demographic data presented in Chapter 7, which in turn was based on observations of 366 wild dogs in 11 packs on our 2,600 km2 study site, from 1991 to 1996. Population density varied among years, from 49 dogs/1,000 km2 to 63 dogs/1,000 km2. We typically had about 140 individuals under study at any one time, including pups. From field notes, we compiled a database with one record per individual for every month in which that individual was seen or inferred to be alive from later observations (⳱ 7,317 dog-months). This database contained information on age, reproduction, dispersal status, and social status. From this database, we derived a second database summarizing information at the level of packs (numbers of adults of each sex, number of yearlings of each sex, number of breeding females, number of pups of each sex born, number of pups of each sex surviving at one year). With user-written programs, we determined means and variances for age- and sex-specific survival, dispersal risk (method of Waser et al. 1994), pack size and composition, sex ratios, and the parameters for a regression of reproductive success on pack size. We constructed a table summarizing 58 dispersal events (Chapter 8) and used the table to derive dispersal rules for the simulation as described below.
296 ▪ C H A P T E R 1 3
Model Structure The fate of each wild dog in the population was traced from birth to death, and extinction occurred when either sex fell to zero individuals. Each individual was characterized by its pack membership, sex, and age. During each year of the simulation, each wild dog had an age-dependent probability of survival. A litter was added annually to each pack that held at least one male and one female (i.e., packs comprising only a single sex did not produce litters until joined by members of the opposite sex). Litter size was selected randomly from a Poisson distribution whose mean value was determined by pack size. Specifically, the average litter size for the distribution (µ) was calculated as 1.87 Ⳮ (0.51 ⳯ pack size), based on least squares regression of litter size on pack size in Selous (t33 ⳱ 2.24; P-value ⳱ 0.03). To produce agreement in age structure between simulated populations and the Selous population, we adjusted the intercept to 2.30 and the slope to 0.63. This adjustment was necessary to make the model internally consistent (i.e., to produce simulated age structures and pack sizes similar to those observed in the study population). Although the regression is significant, our data on age structure are more precise than the regression coefficients: The adjusted slope and intercept are within the 95% confidence intervals of the least squares estimates. The sex ratio of each litter was selected from a binomial distribution, where the average proportion of males was 0.57. This proportion is similar in other free-ranging and captive populations (Malcolm 1979; Fuller et al. 1992a). Although the mechanisms that determine the number of packs in a population are not fully understood, two patterns suggest a possible mechanism for determining the number of packs. First, lion density has a strongly negative correlation with wild dog population density: dogs/km2 ⳱ exp(0.65–45.99 [lions/km2]) (Creel & Creel 1996). Second, pack size is not detectably related to lion density (P-value ⳱ 0.45; n ⳱ 6 populations; simple correlation). Based on these patterns, we predicted the number of packs each year of the simulation (except the first) by dividing the predicted number of wild dogs (based on lion density according to the immediately preceding expression) by the mean pack size in the Selous population (⳱ 4.7 yearlings Ⳮ 9.0 adults). Note that mean pack size was only used to determine whether new packs formed or existing packs failed—the rules governing pack size were independent of this process. As described above, pack size increased according to the relationship between pack size and fecundity, and decreased according to age- and sex-specific mortality rates. If the predicted number of packs (rounded to the nearest integer) exceeded the current number of packs in the simulated population, then a new pack was established by one male and one female from the pack(s) with the largest number of males and females, respectively. This approach allows pack size to be influenced by the fecundity and mortality of the individuals within each pack, while the
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 297
number of packs is influenced by ecological factors thought to affect population density. If the predicted number of packs was less than the actual number, then the smallest pack in the simulated population failed, and each wild dog in that pack dispersed into randomly selected packs. If any pack’s size exceeded 12, all the yearlings of one sex (chosen randomly) dispersed from that pack into the smallest pack. This pattern of dispersal is similar to that observed in the Selous population and in Botswana (McNutt 1996; Chapter 8). Dispersal was accompanied by a 0.5 chance of mortality. Subjecting dispersers to an additional risk of mortality is supported by data from Selous, where the estimated mortality due to dispersal fell between 0.38 and 0.83 (N ⳱ 127 dispersers; method of Waser et al. 1994). Collectively, the rules for recruitment, mortality, and dispersal determined the size and composition of each simulated pack in each year. We checked the appropriateness of these rules by confirming that simulated packs maintained a distribution of sizes and compositions that resembled real packs throughout the simulation (Vucetich & Creel 1999). Because calculating the number of packs required simulated information about lion density, we modeled lion population dynamics according to a diffusion process (Dennis et al. 1991), where the average lion population size, total range of lion population size, and annual lion population growth rate were directly specified as independent variables, but year-to-year variability in population size was estimated using data from a naturally regulated lion population (from the Ngorongoro Crater, Tanzania [C. Packer, personal communication; Packer et al. 1991]). We used this population because it provides the best information available for estimating year-to-year variability in a lion population. The initial number of wild dog packs in each simulation depended on the initial lion density. Each pack initially consisted of 9 adults, 5 yearlings, and 8 pups (mimicking the Selous population), but at annual time steps the composition of each pack changed according to the births, deaths, immigration, and emigration of individuals. If initial lion density was 85 individuals/1,000 km2, the simulation began with a population of 98 adult and yearling wild dogs (7 packs). We chose 98 wild dogs as a baseline model because most wild dog populations are about 100 or smaller, and initial simulations suggested that, in the absence of ecological or demographic problems, a population of ⬃100 dogs will persist for 100 years with a probability near one. Thus, a population of 98 wild dogs in seven packs provides a reasonable standard of comparison for evaluating extinction risk in simulated populations subjected to various challenges (e.g., increased interspecific competition or disease). The initial sex ratio was determined by randomly assigning each wild dog a sex, where the probability of being male was 0.57 (Malcolm 1979). Like pack size and composition, the population sex ratio then varied according to the birth and death of individuals. With all parameter values
298 ▪ C H A P T E R 1 3
fixed, we calculated the probability of persisting to 20, 50, and 100 years (denoted PP20, PP50, and PP100), based on the trajectories of 20,000 simulated populations for each scenario. We then evaluated the sensitivity of population persistence by varying one parameter at a time.
13.3 Sensitivity Analysis and Results Lion Density We modeled probabilities of persistence in wild dog populations for average lion densities ranging from 50 lions/1,000 km2 to 200 lions/1,000 km2. This covers the range of lion densities typically observed in nature (see Figure 13.5 for specific examples). When average lion densities were below 100 lions/1,000 km2, the probability of persistence (PPx) of wild dog populations was near 1.0 for periods of 20 to 100 years. As average lion density increased from 100 to 140 lions/1,000 km2, PPx dropped precipitously (Figure
Figure 13.5 The influence of lion density on the probability of a population of 98 wild dogs persisting 20, 50, and 100 years. Arrows indicate lion densities measured in several ecosystems. Heavy arrows indicate ecosystems in which wild dogs have disappeared.
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 299
Figure 13.6 The influence of temporal variation in lion density on the probability of a population of 98 wild dogs persisting 20, 50, and 100 years in environments with moderate and high lion density. Moderate lion density is 100 adult lions/1,000 km2, and high lion density is 131 adult lions/1,000 km2.
13.5). For average lion densities exceeding 140 lions/1,000 km2, PPx was virtually zero, even for periods as short as 20 years. We held lion density within 7% of the average density for these simulations, to examine the effect of mean lion density with minimal variation in lion density. To examine the effect of variation in lion density on wild dog population viability, we calculated PPx for constant average lion densities, but allowed the range (i.e., maximum–minimum) to vary (Figure 13.6). These simulations were repeated for moderate (100 lions/1,000 km2) and high (131 lions/1,000 km2) lion densities. At moderate lion densities, PPx was insensitive to changes in lion variation with a range ⬍30 lions/1,000 km2 and was only modestly affected by variation with a range ⬎30 lions/1,000 km2. At high lion densities, PPx dropped precipitously as the range increased beyond 20 lions/1,000 km2. With a large mean and high variance, lion density occasionally reached levels that quickly decimated wild dog populations. Although the lion population may subsequently drop, a bottleneck in wild dog numbers can be severe enough that recovery is unlikely, regardless of lion numbers.
300 ▪ C H A P T E R 1 3
Disease We modeled the influence of three diseases that are likely to affect wild dog populations: rabies (Gascoyne et al. 1993; Kat et al. 1995), canine distemper virus (CDV; Alexander et al. 1996; Roelke-Parker et al. 1996), and canine parvovirus (CPV; Creel et al. 1997c). These simulations are characterized by an annual probability (ranging from 0–100%) that the disease would affect the population via reduced survivorship (in that year only). The pattern of reduction in survivorship was as follows: Rabies affects all age classes equally, CDV primarily affects pups and yearlings, and to a lesser extent adults, and CPV affects only pups. We modeled each disease by increasing mortality for the age classes described above by specified (but arbitrary) values (Figure 13.7). We varied effects on survival over wide ranges to account for poor estimates of true disease-induced mortality. Because wild dogs may be more frequently exposed or more vulnerable to these diseases when the density of interspecific competitors is high (Grenfell & Dobson 1995), we modeled the influence of disease at moderate and high lion densities. The qualitative effect of disease on wild dog populations was similar under moderate and high lion densities (Figure 13.7), but extinction risk could be very high in the presence of disease and high lion density (provided the effects of disease and high lion density are additive, as we have modeled them). As modeled, rabies generally had a much greater effect on extinction risk than CDV or CPV. When CPV or CDV reduced survival by 0.20 or less, extinction risk increased little, unless outbreaks occurred frequently. In general, the simulations suggest that the effect of CPV is substantially less important than the effect of rabies, and that better data are needed on the effect of CDV on wild dogs’ survival before conclusions can be drawn (see Chapter 12). Immigration Because the risk of extinction was high when lion density was high or when infectious disease was severe, we investigated the degree to which typical patterns of immigration might mitigate extinction risk. Rather than modeling the dynamics of a neighboring source of immigrants, we simply allowed immigrants to appear under the following rules. Consistent with the dispersal behavior of wild dogs (Frame et al. 1979; McNutt 1995; Waser 1996), each immigrating group consisted of same-sex individuals, and the sex of each group was randomly chosen according to the observed primary sex ratio (i.e., the probability of being a male is 0.57). From observations in Selous, the number of dogs in each immigrating group was selected randomly from a Poisson distribution with a mean of four. We determined the number of groups immigrating each year randomly by drawing from a Poisson distribu-
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 301
Figure 13.7 The influence of viral diseases on the probability of a population of 98 wild dogs persisting 50 years in environments with moderate and high lion density. Moderate lion density is 100 adult lions/1,000 km2, and high lion density is 131 adult lions/1,000 km2.
tion with a mean of one. Immigration occurred only during randomly chosen years to vary immigration rates from an average of 0.1 group per year to 1 group per year (Figure 13.8). Immigrating groups were incorporated into the population in three ways. The first option was to join a group that only held dogs of the opposite sex, if such a group existed, to form a new breeding pack. The second option was to join a pack with both sexes, but fewer members of the same sex, if such a group existed. In this case, the same-sexed pack members were displaced by the immigrants. The displaced dogs then dispersed as a group into another
302 ▪ C H A P T E R 1 3
Figure 13.8 The influence of immigration on the probability of a population of 98 wild dogs persisting 50 years in environments with moderate (100 adult lions/1,000 km2) and high (131 adult lions/1,000 km2) lion density. Each immigrant group consisted, on average, of 4 same-sexed individuals.
pack (following the same rules for immigration), with a 0.5 risk of death added to normal age- and sex-specific mortality. The third option was to establish a new single-sex group, if options one and two were not available. The new pack (like other packs) was subject to disbanding if lion density was too high to allow an additional pack. When lion density was moderate (100 lions/1,000 km2), the probability of persistence to 50 years was high in the absence of immigration, so immigrants had little effect on PP50. When lion density was high (131 lions/1,000 km2), however, an average of one immigrant group every other year increased PP50 from 0.60 to 0.80, and an average of one immigrant group per year further increased PP50 to 0.98. These results strongly suggest that a modest amount of connectedness among populations can substantially increase the probability of persistence for a wild dog population that faces unfavorable ecological and demographic conditions (compare Figures 13.5 and 13.8). Litter Size Because the number of pups raised is correlated with pack size, we modeled the sensitivity of extinction risk to changes in average litter size by altering the y-intercept of the regression of litter size on pack size. In this manner, we estimated extinction risk when litter size varied from approximately 5 to 12 pups per litter. (Mean litter size in Selous was 8.6 Ⳳ 0.7 SE, compared to means of 10.0 and 11.9 in Serengeti and Kruger National Parks: Creel et al. 1997c; Maddock & Mills 1994; van Heerden et al. 1995; Fuller et al. 1992a). Under moderate lion densities, reducing mean litter size had little effect on
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 303
Figure 13.9 The influence of average litter size on the probability of a population of 98 wild dogs persisting 50 years in environments with moderate (100 adult lions/1,000 km2) and high (131 adult lions/1,000 km2) lion density.
extinction risk until average litter size dropped below 6–7 pups (Figure 13.9). Likewise, an increase in mean litter size from 8 to 12 pups/litter produced only a small reduction in extinction risk with high lion density. As mean litter size fell below 6 pups/litter, extinction risk increased substantially for both moderate and high lion densities. Pack Size Because reproductive success is strongly correlated with pack size, we directly examined the influence of pack size on extinction risk by changing the threshold size of the pack at which individuals dispersed. Our model best mimicked the Selous population when the dispersal threshold was 12 adults. Under moderate lion density, 12 adults was slightly above the pack size that yielded nearly a 100% chance of persistence to 50 years, so an increase in mean pack size would do little to decrease extinction risk (Figure 13.10). By contrast, extinction risk increased sharply as the dispersal threshold dropped below 10 adults per pack. To illustrate, PP50 dropped from 0.96 to 0.74 as the dispersal threshold decreased from 10 to 6 adults. Under high lion density, 12 wild dogs is near (slightly above) the inflection point in the PP50 curve, so an increase or a decrease in mean pack size would substantially increase or a decrease the population’s probability of
304 ▪ C H A P T E R 1 3
Figure 13.10 The influence of average pack size on the probability of a population of 98 wild dogs persisting 50 years in environments with moderate (100 adult lions/1,000 km2) and high (131 adult lions/1,000 km2) lion density.
persisting. Above approximately 20 dogs per pack, the curve flattens and population persistence increases little. These results suggest that there is a fairly strong interaction between lion density and pack size in their effects on the persistence of wild dog populations. Our model translates recent data on ecological factors that affect wild dogs into estimates of extinction risk. Although the simulations are as realistic as current data permit, the quantitative estimates of risk should be treated with caution. Extinction risk may be higher or lower if important population processes were not incorporated. The model’s parameter values are based on 6 field-years of observation, but these years may not be representative of population trends over the next 20 to 100 years. Nevertheless, these simulations provide a means to explore the relative risk of extinction under different ecological and demographic scenarios. Previous simulations of extinction risk in wild dogs have used the program VORTEX (Lacy & Kreeger 1992). Ginsberg et al. (1995) and Burrows et al. (1995) used VORTEX to debate the role of chance in the decline of a population of 20–30 wild dogs in Serengeti National Park. The message to be drawn from this debate regarding wild dog conservation in other contexts
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 305
is limited, because luck plays so large a role in the dynamics of populations that are this small. In a broader context, Ginsberg & Woodroffe (1997) used VORTEX to examine the effects of population size, fragmentation, inbreeding, and changes in mortality on extinction risk in wild dogs. Their simulations suggested that a population’s size (which varied from 20 to 100 adults) was the most important variable affecting persistence. Changes in adult mortality had a larger effect on extinction risk than changes in juvenile mortality. Fragmentation reduced the persistence of small populations (but not populations larger than 50 dogs), unless the population was linked to others by dispersal. These conclusions seem reasonable, but results from generalized population viability analysis (PVA) programs such as VORTEX can be misleading. Mills et al. (1996) discussed this problem after obtaining widely divergent estimates of extinction risk from four generalized population viability programs (including VORTEX) for a single grizzly bear population, despite efforts to implement each program as identically as possible. Inconsistencies arise because all population viability analyses depend on assumptions about demography, ecological interactions, and social structure, and these assumptions vary among PVA programs. Because structure in population models has important consequences for dynamics (Caswell et al. 1997; p. 9), the match between PVA assumptions and a given species’ population biology can have strong effects on the plausibility of population viability estimates (Mills et al. 1996; Pascual et al. 1997). Using a generalized PVA model to predict population viability when the population’s biology does not fit the assumptions of the PVA model is analogous to drawing inferences from a statistical test when the data do not meet its assumptions. Although generalized PVA programs provide valuable insight when implemented on populations whose biology match the assumptions of the model, generalized programs are not appropriate when population dynamics are strongly affected by social structure or complex ecological interactions (Vucetich et al. 1997; Boyce 1992; Caughley 1994). Socially, wild dog population dynamics are influenced by the formation and maintenance of stable packs, and pack size is related to foraging success (Creel 1997) and reproductive success (Creel et al. 1998). Ecologically, interspecific competition with lions and spotted hyenas affects foraging success, energy balance, habitat selection and population density (Fanshawe & Fitzgibbon 1993; Creel & Creel 1996; Mills & Gorman 1997; Gorman et al. 1998) Individually-based viability models (DeAngelis & Gross 1992) allow for analyses that are tailored to these ecological and demographic peculiarities. Our simulations corroborate empirical observations from Kruger (Mills & Gorman 1997), Serengeti (Kruuk 1972; Malcolm 1979), and Selous (Creel & Creel 1996) that interspecific competition and predation by lions increase extinction risk for wild dog populations. Also consistent with our predictions is the disappearance of wild dogs in Serengeti and Ngorongoro (Estes &
306 ▪ C H A P T E R 1 3
Goddard 1967; Packer et al. 1991; Creel & Creel 1996) within a decade of lion densities increasing from low (⬍100 lions/1,000 km2) to high (⬎140 lions/1,000 km2) (Fig. 13.5). Our simulations suggest that competition with lions is critically important to wild dog conservation because the probability of persistence (PPx) declines steeply as lion density increases, well within the observed range of lion density. Given the sensitivity of wild dog population viability to lion density, the conservation of wild dogs will require extreme caution, because valid population estimates for wild dogs and their competitors (lions and hyenas) are not easily obtained (Rodgers 1974; Creel & Creel 1996; Mills 1996; Ogutu & Dublin 1998). Like lions, spotted hyenas compete with wild dogs (Kruuk 1972; Creel & Creel 1996; Gorman et al. 1998), and it would be desirable to model extinction risk for a wild dog population in competition with spotted hyenas as well as lions, but information on the population dynamics of hyenas is so limited that it prevents us from modeling their population dynamics as we did for lions. Stander (1991), however, showed that there is a positive correlation between the densities of lions and hyenas across ecosystems. This correlation makes it difficult to disentangle the effects of lions and hyenas on wild dogs (Creel & Creel 1996), and a model that treated these effects as independent would probably overestimate the true risk of extinction. This problem highlights the general difficulty of estimating extinction risk with a model that explicitly incorporates multiple ecological interactions (Boyce 1992). The effect of lions is registered primarily through predation on dogs, whereas the effect of hyenas is registered mainly through food stealing. It is difficult to incorporate either of these effects into a demographic model because predation (like most other causes of death) is rarely observed and food loss affects demography indirectly. In our model, the effect of competition with hyenas could be considered in a crude manner by converting hyenas into “lion-equivalents,” in terms of their effect on wild dogs. We modeled the impact of viral diseases by increasing age-specific mortality. Woodroffe et al. (1997) used a similar approach to model the effect of disease on wild dogs and reached some conclusions that are generally similar to ours. Woodroffe et al. (1997) used VORTEX to model CDV and rabies as catastrophes that occurred in 3% of years and reduced persistence by 50%. With this approach, they found essentially no effect of disease on 50year extinction risk for populations larger than 20 dogs, whereas we conclude that rabies and CDV might have substantial effects on 50-year extinction risk for a population of ⬃100 (Woodroffe et al. 1997, their Figure 5.2; cf. our Figure 13.7). We modeled the effects of disease for annual probabilities of outbreak ranging from 0 to 1, whereas Woodroffe et al. (1997) used a single probability of 0.03/year. Some wild dog populations have experienced viral disease outbreaks at rates substantially higher than 0.03/year (e.g., Serengeti, with outbreaks of rabies and/or CDV at intervals ⬍10 years: Schaller 1972;
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 307
Malcolm 1979; Gascoyne et al. 1993; Roelke-Parker et al. 1996). In our simulations, outbreaks of rabies and CDV decreased persistence when disease-induced adult mortality exceeded 0.3, and outbreaks occurred at intervals shorter than 10 years. Serengeti, where wild dogs disappeared, met these conditions (Malcolm 1979; Gascoyne et al. 1993; Burrows et al. 1995). It is not yet clear whether CDV is normally fatal for adult wild dogs (Alexander et al. 1996; Roelke-Parker et al. 1996; Creel et al. 1997c), so we modeled CDV as fatal for juveniles and yearlings, but not adults. If more extensive data show that CDV is typically fatal for adult wild dogs, then our CDV simulation will underestimate its impact, and our rabies simulation would provide a better model. For now, we consider it useful to model three classes of viral diseases: those that kill all age classes (rabies); those that kill yearlings and pups (CDV, with the above caveat); and those that kill only pups (CPV). To mimic CPV, Woodroffe et al. (1997) reduced juvenile mortality in all years, and concluded that “in all but the smallest populations [20 dogs], varying juvenile mortality in the region 50–70% has little effect on population persistence (p. 85). Above 70%, however, small increases in juvenile mortality generate large changes in population persistence”. Our simulation results closely match this conclusion, despite a different modeling approach. We modeled CPV by increasing juvenile mortality in a proportion of years, so Woodroffe et al.’s scenario corresponds to Figure 13.7, with the annual probability of an epidemic fixed at one. Under these conditions, our simulation shows little effect of disease-induced mortality in pups until the mortality due to disease reaches 0.30 or total pup mortality of 0.71 (baseline annual mortality for pups was 0.41). Disease-induced pup mortality as high as 0.3– 0.4 had little effect when the annual probability of an epidemic dropped below 0.5. Collectively, our simulation results and those of Woodroffe et al. (1997) suggest that population persistence is relatively insensitive to juvenile mortality in wild dogs, unless it is severe and persistent. However, simulations of variation in litter size reveal rapid declines in population persistence if mean litter size drops below six pups, and this threshold should be considered when evaluating risks due to pup mortality. Finally, sensitivities and elasticities from the Leslie matrix indicate that variation in juvenile survival has a greater effect on population growth than does any other demographic variable (Chapter 7), and variation in density among populations relates more closely to juvenile survival than to adult survival. This set of results is somewhat contradictory, but the difference is partially reconciled by recognizing that the individual-based model examines changes in mortality that are intermittent, while the Leslie analysis examines changes in mortality that are constant. Increases in mean litter size above the range seen in the wild (8.5–11.9) had little effect on extinction risk. This is not surprising: Wild dogs produce unusually large litters, so pup production is not likely to be demographically
308 ▪ C H A P T E R 1 3
limiting. If mean litter size dropped below six pups/litter, however, extinction risk would increase substantially. Mean litter size in Selous is only 18% larger than this threshold, so any factor that reduces fecundity or the survival of newborns is a serious threat. In other populations, mean litter size is large enough that this threshold is perhaps of less concern. Based on the simulation model, we have previously suggested (Vucetich & Creel 1999) that populations comprising larger packs (up to about 20 adults) would have substantially lower extinction risk at high lion density (Figure 13.10). Wild dogs in large packs attain a higher net rate of energy intake than dogs in small packs—given this, why are packs not larger? One possible explanation is that social suppression of reproduction sets an upper limit on pack size because subordinates are unwilling to wait for breeding opportunities when the odds of attaining dominance fall too low (Frame et al. 1979; Malcolm & Marten 1982; Creel et al. 1997a). As pack size increases, subordinates are less likely to attain dominance and breed. If this explanation is correct, individual selection and group selection would act in opposition, because behavior that appears to increase individual fitness would reduce the population’s likelihood of persisting. Moreover, Eigen analysis of Leslie matrices casts doubt on our conclusion that larger packs would fare better. As shown in Figure 13.4, this analysis suggests that the observed mean pack size in Selous comes close to maximizing the population’s growth rate. We did not directly model the effects of humans on population persistence, though mortality due to humans in Selous is included in the agespecific mortality values used as inputs. Human impacts on wild dogs are not strong in Selous, but are important elsewhere (Rasmussen 1998). Wild dogs range very widely, and therefore are likely to move out of small protected areas, relative to more sedentary species. However, human effects do not provide a strong explanation as to why wild dogs are extinction-prone. In the 1800s, prior to widespread effects of humans on the distributions of carnivores in eastern and southern Africa, Selous (1908) reported that wild dogs were widely distributed but nowhere common. Also, wild dogs attain much lower densities than sympatric large carnivores, even when human impacts on the other species are known to be strong. For example, snaring imposes heavy mortality on spotted hyenas in Serengeti National Park (Hofer et al. 1993a, b), but spotted hyenas outnumbered wild dogs by at least an order of magnitude (Creel & Creel 1996) for several decades. These patterns suggest that wild dogs are extinction-prone for reasons of ecology.
13.4 Summary and Recommendations Our simulations suggest that wild dog populations smaller than 100 individuals (most populations) face a substantial risk of extinction within 100 years
E X T I N C T I O N R I S K A N D C O N S E R VAT I O N ▪ 309
under many realistic combinations of ecological and demographic conditions. As field studies suggest, competition with larger carnivores and viral diseases both pose serious problems for the conservation of wild dogs, particularly when paired with small pack size or small litters. Our conclusions are based on demographic data collected from a single population. To the extent that other populations differ in their demography, the conclusions may change. Future attempts to model extinction risk for wild dogs would benefit from better data on (1) the degree to which human-caused mortality is additive or compensatory, (2) age-specific mortality due to infectious diseases, (3) transmission rates among species, (4) the demography of dominant competitors, and (5) the mechanisms by which competitors affect wild dogs’ demography. Although our quantitative estimates of extinction risk should be evaluated cautiously, they suggest that active management will be needed to maintain many wild dog populations. Simulations of immigration suggest that management could greatly reduce extinction risks by maintaining dispersal among populations or by augmenting populations with single-sex immigrant groups of three to four individuals at intervals of one to two years. This amount of translocation is probably feasible for many populations. In very small populations that hold only one or two packs (e.g., Hluhluwe), translocations of single-sex groups are risky because fights between potential immigrants and residents can be fatal (Frame et al. 1979; S. & N. Creel, unpublished data). In very small populations, introductions of socially bonded breeding groups might be preferable. A serious difficulty for translocation programs is to identify a source population that is not put at risk by the removal of dogs for translocation. The release of dogs bred in captivity has had mixed success due to their poor abilities to hunt and to avoid lions (Scheepers & Venzke 1995; Woodroffe et al. 1997), so at least some wildborn dogs are desirable for translocations. Source populations might include the Selous Game Reserve (Tanzania), Kruger National Park (South Africa), and possibly the population of northern Botswana.
References
Abbott, D. H. 1984. Behavioral and physiological suppression of fertility in subordinate marmoset monkeys. American Journal of Primatology 6: 169–186. . 1987. Behaviourally mediated suppression of reproduction in female primates. Journal of Zoology, London 213: 455–470. . 1993. Social conflict and reproductive suppression in marmoset and tamarin monkeys. In: Primate Social Conflict. W. A. Mason & S. P. Mendoza, eds. pp. 331–373. Albany: State University of New York Press. Abbott, D. H., Saltzman, W., Schultz-Darken, N. J., & Smith, T. E. 1997. Specific neuroendocrine mechanisms not involving generalized stress mediate social regulation of female reproduction in cooperatively breeding marmoset monkeys (Callithrix jacchus jacchus). Annals of the New York Academy of Sciences 807: 219–238. Abramsky, Z., Bowers, M. A., & Rosenzweig, M. L. 1986. Detecting interspecific competition in the field: Testing the regression method. Oikos 47: 199–204. Akaike, H. 1983. Information measures and model selection. International Statistical Institute 44: 277–291. Alberts, S. C., Sapolsky, R. M., & Altmann, J. A. 1992. Behavioral, endocrine and immunological correlates of immigration by an aggressive male into a natural primate group. Hormones & Behavior 26: 167–178. Alexander, K. A., & Appel, M.J.G. 1994. African wild dogs (Lycaon pictus) endangered by a canine distemper epidemic among domestic dogs near the Masai Mara National Reserve, Kenya. Journal of Wildlife Diseases 30: 481–485. Alexander, K. A., Richardson, J. D., & P. W. Kat. 1992. Disease and conservation of African wild dogs. Swara 15: 13–14. Alexander, K. A., Kat, P. W., Wayne, R. K., & Fuller, T. K. 1994. Serologic survey of selected canine pathogens among free-ranging jackals in Kenya. Journal of Wildlife Diseases 30: 486–491. Alexander, K. A., Kat, P. W., Frank, L. G., Holekamp, K. E., Smale, L., House, C., & Appel, M.J.G. 1995. Evidence of canine distemper virus infection among freeranging spotted hyenas (Crocuta crocuta) in the Masai Mara, Kenya. Journal of Zoo and Wildlife Medicine 26: 201–206. Alexander, K. A., P. W. Kat, Munson, L. A., Kalake, A., & M.J.G. Appel. 1996. Canine distemper–related mortality among wild dogs (Lycaon pictus) in Chobe National Park, Botswana. Journal of Zoo and Wildlife Medicine 27: 426–427. Anderson, R. M., & May, R. M. 1979. Population biology of infectious diseases. Nature 280: 455–461. Angier, N. 1996. The most social of wild canids struggles for survival. The New York Times, December 3, 1996: C1,C6. Anonymous. 1967. Results of killing “vermin.” Oryx 9: 175–176. Appel, M.J.G. 1987. Canine distemper virus. In: Virus Infections of Carnivores, vol. 1. M.J.G. Appel, ed., pp. 133–159. Amsterdam: Elsevier Science Publishers. Appel, M.J.G., & Robson, D. S. 1973. A microneutralization test for canine distemper virus. American Journal of Veterinary Research 34: 1459–1463.
312 ▪ R E F E R E N C E S
Appel, M.J.G., & Summers, B. A. 1995. Pathogenicity of morbilliviruses for terrestrial carnivores. Veterinary Microbiology 44: 187–191. Arcese, P., Hando, J. & Campbell, K. 1995. Historical and present-day anti-poaching efforts in Serengeti. In: Serengeti II: Dynamics, Conservation and Management of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 506–533. Chicago: University of Chicago Press. Arnold, W., & Dittami, J. 1997. Reproductive suppression in male alpine marmots. Animal Behaviour 53: 53–66. Attwell, R.I.G. 1959. The African hunting dog: A wildlife management incident. Oryx 4: 326–328. Bailey, T. N. 1993. The African Leopard: Ecology and Behavior of a Solitary Felid. New York: Columbia University Press. Baird, R. W., & Dill, L. M. 1996. Ecological and social determinants of group size in transient killer whales. Behavioral Ecology 7: 408–416. Ballard, W. B., Whitman, J. S., & Gardner, C. L. 1987. Ecology of an exploited wolf population in south-central Alaska. Wildlife Monographs 98: 1–53. Bateman, A. A., Singh, T., Kral, T., & Solomon, S. 1989. The immune-hypothalamicpituitary-adrenal axis. Endocrine Reviews 10: 92–112. Baum, M. 1992. Neuroendocrinology of sexual behavior in the male. In: Behavioral Endocrinology. J. Becker, M. Breedlove, & D. Crews, eds., pp. 97–130. Cambridge, MA: Massachusetts Institute of Technology Press. Bednarz, J. C. 1988. Cooperative hunting in Harris’ hawks (Parabuteo unicinctus) Science 239: 1525–1527. Bere, R. M. 1956. The African wild dog. Oryx 3: 180–182. . 1975. Mammals of East and Central Africa. London: Longman. Berkovitch F. B., & Clarke, A. S. 1995. Dominance rank, cortisol concentrations, and reproductive maturation in male rhesus macaques. Physiology & Behavior 58: 215–221. Blackburn, T. M., & Gaston, K. J. 1994. Animal body size distributions: Patterns, mechanisms and implications. Trends in Ecology and Evolution 9: 471–474. Blanchard, D. C., Spencer, R. L., Weiss, S. M., Blanchard, R. J., McEwen, B., & Sakai, R. R. 1995. Visible burrow system as a model of chronic social stress: Behavioral and neuroendocrine correlates. Psychoneuroendocrinology 20: 117–134. Blumenschine, R. J. & Caro, T. M. 1986. Unit flesh weights of some East African bovids. African Journal of Ecology 24: 273–286. Boesch, C. 1994. Cooperative hunting in wild chimpanzees. Animal Behaviour 48: 653–667. Bothma, J. du P., & LeRiche, E.A.N. 1990. The influence of increasing hunger on the hunting behaviour of southern Kalahari leopards. Journal of Arid Environments 18: 79–84. Bourke, A.F.G., & Franks, N. R. 1995. Social Evolution in Ants. Princeton, NJ: Princeton University Press. Boyce, M. S. 1992. Population viability analysis. Annual Review of Ecology and Systematics 23: 481–506. Boyce, M. S., & Anderson, E. M. 1999. Evaluating the role of carnivores in the Greater Yellowstone Ecosystem. In: Carnivores in Ecosystems: The Yellowstone Experience. T. W. Clark, A. P. Curlee, S. C. Minta, & P. M. Kareiva, eds., pp. 265– 283. New Haven, CT: Yale University Press.
R E F E R E N C E S ▪ 313
Brain, P. F. 1983. Pituitary and gonadal influences on social aggression. In: Hormones and Aggressive Behavior. B. B. Svare, ed., pp. 3–26. New York: Plenum Press. Brault, S., & Caswell, H. 1993. Pod-specific demography of killer whales (Orcinus orca). Ecology 74: 1444–1454. Brockman, H. J. 1997. Cooperative breeding in wasps and vertebrates: The role of ecological constraints. In: The Evolution of Social Behavior in Insects and Arachnids. J. C. Choe & B. J. Crespi, eds., pp. 347–371. Cambridge, UK: Cambridge University Press. Bronson, F. H., & Eleftheriou, B. E. 1964. Chronic physiological effects of fighting in mice. General and Comparative Endocrinology 4: 9–14. Brown, C. R., & Brown, M. B. 1996. Coloniality in the Cliff Swallow. Chicago: University of Chicago Press. Brown, J. L. 1978. Avian communal breeding systems. Annual Review of Ecology and Systematics 9: 123–155. . 1987. Helping and Communal Breeding in Birds. Princeton, NJ: Princeton University Press. Brown, J. L., & Brown, E. R. 1981. Kin selection and individual selection in babblers. In: Natural Selection and Social Behavior: Recent Research and New Theory. R. D. Alexander & D. W. Tinkle, eds., pp. 244–256. New York: Chiron Press. Brown, J. L., Wasser, S. K., Wildt, D. E., & Graham, L. H. 1994. Comparative aspects of steroid hormone metabolism and ovarian activity in felids, measured noninvasively in feces. Biology of Reproduction 51: 776–786. Burnham, K. P., & Anderson, D. R. 1998. Model Selection and Inference: A Practical Information-Theoretic Approach. New York: Springer. Burrows, R. 1992. Rabies in wild dogs. Nature 359: 277. . 1995. Demographic changes and social consequences in wild dogs, 1964– 1992. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 400–420. Chicago: University of Chicago Press. Burrows, R., Hofer, H., & East, M. 1994. Demography, extinction and intervention in a small population: The case of the Serengeti wild dogs. Proceedings of the Royal Society of London, Series B 256: 281–292. . 1995. Population dynamics, intervention and survival in African wild dogs (Lycaon pictus). Proceedings of the Royal Society of London, Series B. 262: 235– 245. Butynski, T. M. 1974. An encounter between African wild dog and domestic dog. East African Wildlife Journal 12: 243. Cade, C. E. 1967. Notes on breeding the cape hunting dog Lycaon pictus at Nairobi zoo. International Zoo Yearbook 7: 122–123. Calder, W. A. 1984. Size, Function, and Life-History. Cambridge, MA: Harvard University Press. Calvert, W. H., Hendrick, L. E., & Brower, L. P. 1979. Mortality of the monarch butterfly (Danaus plexippus L.): Avian predation at five overwintering sites in Mexico. Science 204: 847–851. Campbell, K., & Borner, M. 1995. Population trends and distribution of Serengeti herbivores: Implications for management. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 117–145. Chicago: University of Chicago Press.
314 ▪ R E F E R E N C E S
Caraco, T., & Wolf, L. L. 1975. Ecological determinants of group sizes of foraging lions. American Naturalist 109: 343–352. Carbone, C., du Toit, J. T., & Gordon, I. J. 1997. Feeding success in African wild dogs: Does kleptoparasitism by spotted hyaenas influence hunting group size? Journal of Animal Ecology 66: 318–326. Carbyn, L. N. 1982. Coyote population fluctuations and spatial distribution in relation to wolf territories in Riding Mountain National Park, Manitoba. Canadian FieldNaturalist 96, 176–183. Carbyn, L. N., Armbruster, H. J., & Mamo, C. 1994. The swift fox reintroduction program in Canada from 1983 to 1992. In: Restoration of Endangered Species. M. Bowles & C. Whelan, eds., pp. 247–271. Cambridge, UK: Cambridge University Press. Carmichael, L. E., Joubert, J. C., & Pollock, R. V. 1980. Hemagglutination by canine parvovirus: Serologic studies and diagnostic applications. American Journal of Veterinary Research 41: 784–791. Caro, T. M. 1993. Ungulate antipredatory behaviour: Preliminary and comparative evidence from African bovids. Behaviour 128:189–228. . 1994. Cheetahs of the Serengeti Plains. Chicago: University of Chicago Press. . 1995. The importance of behavioural ecology for conservation biology: Examples from Serengeti carnivores. In: Serengeti II: Dynamics, Conservation and Management of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 451–472. Chicago: University of Chicago Press. Caro, T. M., & Fitzgibbon, C. D. 1992. Large carnivores and their prey: The quick and the dead. In: Natural Enemies: The Population Biology of Predators, Parasites and Diseases. M. J. Crowley, ed., pp 117–142. Oxford: Blackwell Scientific. Caro, T. M., & Laurenson, M. K. 1994. Ecological and genetic factors in conservation: A cautionary tale. Science 263: 485–486. Caswell, H. 1997. Methods of matrix population analysis. In: Structured Populations in Marine, Terrestrial and Freshwater Systems. S. Tuljapurkar & H. Caswell, eds., pp. 19–58. New York: Chapman & Hall. . 2000. Prospective and retrospective perturbation analyses: Their roles in conservation biology. Ecology 81: 619–627. . 2001. Matrix Population Models: Construction, Analysis and Interpretation. Sunderland, MA: Sinauer. Caswell, H., Nisbet, R., M., de Roos, A. M., & Tuljapurkar, S. 1997. Structuredpopulation models: Many methods, few basic concepts. In: Structured-Population Models in Marine, Terrestrial and Freshwater Systems. S. Tuljapurkar & H. Caswell, eds., pp. 3–18. New York: Chapman and Hall. Caughley, G. 1977. Analysis of Vertebrate Populations. London: John Wiley & Sons. . 1994. Directions in conservation biology. Journal of Animal Ecology 63: 215–244. Cekan, Z. 1975. Assessment of reliability of steroid radioimmunoassays. Journal of Steroid Biochemistry 6: 271–275. Charnov, E. L., & Orians, G. H. 1973. Optimal foraging: Some theoretical explorations. Unpublished manuscript, cited by Stephens & Krebs (1986). Charnov, E. L., Orians, G. H., & Hyatt, K. 1976. The ecological implications of resource depression. The American Naturalist 110: 247–259.
R E F E R E N C E S ▪ 315
Chepko-Sade, B. D., & Halpin, Z. T. 1987. Mammalian Dispersal Patterns: The Effects of Social Structure on Population Genetics. Chicago: University of Chicago Press. Childes, S. L. 1988. The past history, present status and distribution of the hunting dog Lycaon pictus in Zimbabwe. Biological Conservation 44: 301–316. Choe, J. C., & Crespi, B. J. 1997. The Evolution of Social Behavior in Insects and Arachnids. Princeton: Princeton University Press. Christian, J. J., & Davis, D. E. 1964. Endocrines, behavior and population. Science 146: 1550–1560. Clark, T. W., Curlee, A. P., & Reading, R. P. 1996. Crafting effective solutions to the large carnivore conservation problem. Conservation Biology 10: 940–948. Cleaveland, S., & Dye, C. 1995. Maintenance of a microparasite infecting several host species: Rabies in the Serengeti. Parasitology 111: S33–S47. Clutton-Brock, T. H., Brotherton, P.N.M., Russell, A. F., O’Riain, M. J., Gaynor, D., Kansky, R., Griffin, A., Manser, M., Sharpe, L., McIlrath, G. M., Small, T., Moss, A., & Monfort, S. 2001. Cooperation, control and concession in meerkat groups. Science 291: 478– 481. Cohen, R. 1971. Breastfeeding without pregnancy. Pediatrics 48: 996–997. Colly, L. P., & Nesbit, J. W. 1992. Fatal acute babesiosis in a juvenile wild dog (Lycaon pictus). Journal of the South African Veterinary Medical Association 63: 36–38. Concannon, P. W., Hansel, W., & Visek, B. 1975. The ovarian cycle of the bitch: Plasma estrogen, LH and progesterone. Biology of Reproduction 13: 112–119. Cooper, S. M. 1991. Optimal hunting group size: The need for lions to defend their kills against loss to spotted hyaenas. African Journal of Ecology 29: 130–136. Creel, S. 1992. Cause of wild dog deaths. Nature 360: 633. . 1997. Cooperative hunting and group size: Assumptions and currencies. Animal Behaviour 54: 1319–1324. . 1998a. Conservation of carnivores: The impact of social organization on effective population size. In: Behavioral Ecology and Conservation Biology. T. M. Caro, ed., pp. 246–265, Oxford, UK: Oxford University Press. . 1998b. Birth order, estrogens and sex-ratio adaptation in African wild dogs (Lycaon pictus). Animal Reproduction Science 53: 315–320. . 2001. Cooperative hunting and sociality in African wild dogs, Lycaon pictus. In: Model Systems in Behavioral Ecology. L. A. Dugatkin, ed. Princeton, NJ: Princeton University Press. Creel, S. R., & Creel, N. M. 1991. Energetics, reproductive suppression and obligate communal breeding in carnivores. Behavioral Ecology and Sociobiology 28: 263– 271. . 1995a. African wild dogs in the Selous Game Reserve. IUCN/SSC Canid News 3: 34–35. . 1995b. Communal hunting and pack size in African wild dogs, Lycaon pictus. Animal Behaviour 50: 1325–1339. . 1996. Limitation of African wild dogs by competition with larger carnivores. Conservation Biology 10: 526–538. . 1997a. Status of African wild dogs in the Selous Game Reserve and Mikumi National Park. IUCN/CSG Canid News 5.
316 ▪ R E F E R E N C E S
Creel, S. R., & Creel, N. M. 1997b. Lion density and population structure in the Selous Game Reserve: Evaluation of hunting quotas and offtake. African Journal of Ecology 35: 83–93. . 1998. Six ecological factors that may limit African wild dogs. Animal Conservation 1: 1–9. Creel, S., & Waser, P. M. 1991. Failures of reproductive suppression in dwarf mongooses (Helogale parvula): Accident or adaptation? Behavioral Ecology 2: 7–15. . 1994. Inclusive fitness and reproductive strategies in dwarf mongooses. Behavioral Ecology 5: 333–348. . 1997. Variation in reproductive suppression among dwarf mongooses: Interplay between mechanisms and evolution. In: Cooperative Breeding in Mammals. N. Solomon & J. French, eds., pp. 150–170, Cambridge, UK: Cambridge University Press. Creel, S., Monfort, S. L., Wildt, D. E., & Waser, P. M. 1991. Spontaneous lactation is an adaptive result of pseudopregnancy. Nature 351: 660–662. Creel, S. R., Creel, N. M., Wildt, D. E., & Monfort, S. L. 1992. Behavioral and endocrine mechanisms of reproductive suppression in Serengeti dwarf mongooses. Animal Behaviour 43: 231–245. Creel, S. R., Wildt, D. E., & Monfort, S. L. 1993. Androgens, aggression and reproduction in wild dwarf mongooses: a test of the challenge hypothesis. The American Naturalist 141: 816–825. Creel, S., Creel, N. M., Matovelo, J. A., Mtambo, M.M.A., Batamuzi, E. K., & Cooper, J. E. 1995. The effects of anthrax on endangered African wild dogs (Lycaon pictus). Journal of Zoology, London 236: 199–209. Creel, S., Creel, N. M., & Monfort, S. L. 1996. Social stress and dominance. Nature 379: 212. Creel, S., Creel, N. M., Mills, M.G.L., & Monfort, S. L. 1997a. Rank and reproduction in cooperatively breeding African wild dogs: Behavioral and endocrine correlates. Behavioral Ecology 8: 298–306. Creel, S., Creel, N. M., & Monfort, S. L. 1997b. Radiocollaring and stress hormones in African wild dogs. Conservation Biology 11: 544–548. Creel, S. Creel, N. M., Munson, L., Sanderlin, D., & Appel, M.J.G. 1997c. Serosurvey for selected viral diseases and demography of African wild dogs in Tanzania. Journal of Wildlife Diseases 33: 823–832. Creel, S., Creel, N. M., & Monfort, S. L. 1998. Birth order, estrogens and sex-ratio adaptation in African wild dogs (Lycaon pictus). Animal Reproduction Science 53: 315–320. Creel, S., Spong, G. & Creel, N. M. 2001. Interspecific competition and the population biology of extinction-prone carnivores. In: Conservation of Carnivores, J. Gittleman, S. Funk, D. Macdonald, & R. Wayne, eds., pp. 35–60, Cambridge, UK: Cambridge University Press. Crespi, B. J., & Yanega, D. 1995. The definition of eusociality. Behavioral Ecology 6: 109–115. Cresswell, W. 1994. Flocking is an effective anti-predation strategy in redshanks, Tringa totanus. Animal Behavior 47: 433–442. Cristoph, H-J, 1973. Klinik der Hundekrankheiten. Jena, East Germany: G. Fischer. Crooks, K. R., Sanjayan, M. A., & Doak, D. A. 1998. New insights on cheetah conservation through demographic modelling. Conservation Biology 12: 889–195.
R E F E R E N C E S ▪ 317
Crow, J. F. & Kimura, M. 1970. An Introduction to Population Genetics Theory. Minneapolis: Burgess Publishing. DeAngelis, D. L., & Gross, L. J. 1992. Individual-based Models and Approaches in Ecology. New York: Chapman and Hall. Dekker, D. 1968. Breeding the cape hunting dog at the Amsterdam Zoo. International Zoo Yearbook 8: 27–30. de Kroon, H., van Groenendael, J., & Ehrlen, J. 2000. Elasticities: A review of methods and model limitations. Ecology 81: 607–618. Dennis, B., Munholland, P. L., & Scott, J. M. 1991. Estimation of growth and extinction parameters for endangered species. Ecological Monographs 61: 115–142. Department of Wildlife, 1995. Selous Game Reserve General Management Plan. Tanzania Department of Wildlife, Selous Conservation Programme, Dar es Salaam. de Villiers, M. S., Meltzer, D.G.A., van Heerden, J., Mills, M.G.L., Richardson, P.R.K., & van Jaarsveld, A. S. 1995. Handling-induced stress and mortalities in African wild dogs (Lycaon pictus). Proceedings of the Royal Society of London, Series B. 262: 215–220. de Vos, V. & Bryden, H. B. 1996. Anthrax in the Kruger National Park: Temporal and spatial patterns of disease occurrence. Salisbury Medical Bulletin 87: 26–30. Diamond, J. M. 1997. Aaaaaaaaaaaargh, no! Nature 385: 295–296. Doak, D. Karieva, P., & Klepetka, B. 1994. Modelling population viability for the desert tortoise in the western Mojave desert. Ecological Applications 4: 446–460. Dobson, A. 1995. The ecology and epidemiology of rinderpest virus in Serengeti and Ngorongoro Conservation Area. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 485–505. Chicago: University of Chicago Press. Dorst, J. & Dandelot, P. 1986. A Field Guide to the Larger Mammals of Africa. London: Collins. Drews, C. 1995. Road kills by public traffic in Mikumi National Park, Tanzania, with a note on baboon mortality. African Journal of Ecology 33:89–100. Drickamer, L. C. 1974. Social rank, observability, and sexual behaviour of rhesus monkeys (Macaca mulatta). Journal of Reproduction and Fertility 37: 117–120. Dublin, H. T., Sinclair, A.R.E., Boutin, S., Anderson, E., Jago, M., & Arcese, P. 1990. Does competition regulate ungulate populations? Further evidence from Serengeti. Oecologia 82: 283–288. Dunlop, D. 1963. Eighty-six cases of Addison’s disease. British Medical Journal 2: 887–899. Durant, S. 1998. Competition refuges and coexistence: An example from Serengeti carnivores. Journal of Animal Ecology 67: 370–386. . 2000. Predator avoidance, breeding experience and reproductive success in endangered cheetahs, Acinonyx jubatus. Animal Behaviour 60: 121–130. Durchfield, B., Baumgartner, W., Herbst, W., & Brahm, R. 1990. Vaccine-associated canine distemper infection in a litter of African hunting dogs (Lycaon pictus). Journal of Veterinary Medicine, Series B 37: 203–212. East, R. 1984. Rainfall, soil nutrient status and biomass of large African savanna mammals. African Journal of Ecology 22: 245–270. Eastman, J. R. 1997. Idrisi for Windows version 2.0 User’s Guide. Worcester, MA: Clark University.
318 ▪ R E F E R E N C E S
Eaton, R. L. 1974. The Cheetah: The Biology, Ecology and Behavior of an Endangered Species. New York: Van Nostrand Reinhold. Edelstein, R. M., Gourlay, R. N., Lawson, G.H.K., Morrow, A. N., & Ramachandran, S. 1990. Diseases caused by bacteria. In: Handbook on Animal Diseases in the Tropics. M.M.H. Sewell & D. W. Brocklesby, eds., pp. 33–103. London: BalliereTindall. Emlen, S. T. 1982. The evolution of helping. I. An ecological constraints model. The American Naturalist 119: 29–39. . 1995. An evolutionary theory of the family. Proceedings of the National Academy of Sciences 92: 8092–8099. Estes, R. D. 1991. The Behavior Guide to African Mammals. Berkeley: University of California Press. Estes, R. D., & Goddard, J. 1967. Prey selection and hunting behavior of the African wild dog. Journal of Wildlife Management 31: 52–70. Estrada, A., & Patterson, J. D. 1979. A case of adoption in a captive group of Mexican spider monkeys. Primates 21: 128–129. Evans, P.G.H., Macdonald, D. W., & Cheeseman, C. L. 1989. Social structure of the Eurasian badger (Meles meles): Genetic evidence. Journal of Zoology, London 218: 587–595. Fanshawe, J. H., & Fitzgibbon, C. D. 1993. Factors influencing the hunting success of an African wild dog pack. Animal Behaviour 45: 479–490. Fanshawe, J. H., Frame, L. H., & Ginsberg, J. R. 1991. The wild dog—Africa’s vanishing carnivore. Oryx 25: 137–146. Fanshawe, J. H., Ginsberg, J. R., Sillero-Zubiri, C., & Woodroffe, R. 1997. The status and distribution of remaining wild dog populations. In: The African Wild Dog: Status Survey and Conservation Action Plan. R. Woodroffe, J. Ginsberg, & D. Macdonald, eds., pp. 11–57. Gland, Switzerland: International Union for the Conservation of Nature. FAO. 1961. Rufiji River Survey Report. Rome: Food and Agriculture Organization. Faulkes, C. G., Abbott, D. H., Liddell, C. E., George, L. M. & Jarvis, J.U.M. 1991. Hormonal and behavioral aspects of reproductive suppression in female naked mole-rats. In: The Biology of the Naked Mole-Rat. P. W. Sherman, J.U.M. Jarvis, & R. Alexander, eds., pp. 426–445. Princeton, NJ: Princeton University Press. Feder, H. 1981. Essentials of steroid structure, nomenclature, reactions, biosynthesis, and measurements. In: Neurobiology of Reproduction. N. T. Adler, ed., pp. 19–63. New York: Plenum Press. Fisher, R. A. 1954. Statistical Methods for Research Workers, 12th ed. Edinburgh: Oliver & Boyd. Fox, M. W. 1984. The Whistling Hunters: Field Studies of the Asiatic Wild Dog (Cuon alpinus). Albany: State University of New York Press. Fitzgibbon, C. D. 1988. The antipredator behaviour of Thomson’s gazelles. Ph.D. thesis, University of Cambridge, Cambridge, UK. . 1990. Mixed-species grouping in Thomson’s and Grant’s gazelles: The antipredator benefits. Animal Behaviour 39: 1116–1126. Fitzgibbon, C. D., & Lazarus, J. 1995. Antipredator behavior of Serengeti ungulates. In: Serengeti II: Dynamics, Conservation and Management of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 274–296. Chicago: University of Chicago Press.
R E F E R E N C E S ▪ 319
Foggin, C. M. 1988. Rabies and rabies-related viruses in Zimbabwe: Historical, virological and ecological aspects. Ph.D. thesis, University of Zimbabwe, Harare. Foster, W. A., & Trehorne, J. E. 1981. Evidence for the dilution effect in the selfish herd from fish predation on a marine insect. Nature 293: 466–467. Frame L. H. 1985. Communal breeding and pup development in African wild dogs. M.S. Thesis. Utah State University, Logan, Utah. Frame, G. W., & Frame, L. H. 1976. Female African wild dogs emigrate. Nature 263: 227–229. . 1981. Swift and Enduring: Cheetahs and Wild Dogs of the Serengeti. New York: E. P. Dutton. Frame, L. H., Malcolm, J. R., Frame, G. W., & van Lawick, H. 1979. Social organization of African wild dogs (Lycaon pictus) on the Serengeti Plains, Tanzania, 1967–1978. Zeitschrift fur Tierpsychologie 50: 225–249. Francis, C. M., Anthony, E.L.P., Brunton, J. A., & Kunz, T. H. 1994. Lactation in male fruit bats. Nature 367: 691–692. Frank, L. G. 1986. Social organization of the spotted hyaena Crocuta crocuta. II. Dominance and reproduction. Animal Behaviour 34: 1510–1527. Frank, L. G., Holekamp, K. E., & Smale, L. E. 1995. Dominance, demography and reproductive success of female spotted hyenas. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 364–384. Chicago: University of Chicago Press. Franklin, I. R. 1980. Evolutionary changes in small populations. In: Conservation Biology: An Evolutionary-Ecology Perspective. M. E. Soule & B. A. Wilcox, eds., pp. 135–149. Sunderland, MA: Sinauer. French, J. A. 1997. Proximate regulation of singular breeding in callitrichid primates. In: Cooperative Breeding in Mammals. N. Solomon & J. French, eds., pp. 34–75, Cambridge, UK: Cambridge University Press. French, J. A., Abbott, D. H., Scheffler, G., Robinson, J. A., & Goy, R. W. 1984. Cyclic excretion of urinary oestrogens in female tamarins (Sanguinus oedipus). Journal of Reproduction and Fertility 68: 177–184. French, J. A., Inglett, B. J., & Dethlefs, T. M. 1989. The reproductive status of nonbreeding group members in golden lion tamarin social groups. American Journal of Primatology 18: 73–86. Fritts, S. H., & Mech, L. D. 1981. Dynamics, movements and feeding ecology of a newly protected wolf population in Northwestern Minnesota. Wildlife Monographs 80: 1–79. Fuller, T. K., & Kat, P. W. 1990. Movements, activity and prey relationships of African wild dogs (Lycaon pictus) near Aitong, southwestern Kenya. African Journal of Ecology 28: 330–350. Fuller, T. K., & Kat, P. W. 1993. Hunting success of African wild dogs in southwestern Kenya. Journal of Mammalogy 7: 464–467. Fuller, T. K., & Keith, L. B. 1981. Non-overlapping ranges of coyotes and wolves in Northeastern Alberta. Journal of Mammalogy 62: 403–405. Fuller, T. K., Kat, P. W., Bulger, J. B., Maddock, A. H., Ginsberg, J. R., Burrows, R., McNutt, J. W., & Mills, M.G.L. 1992a. Population dynamics of African wild dogs. In: Wildlife 2001: Populations. D. R. McCullough & R. H. Barret, eds., pp. 1125– 1139. London: Elsevier Applied Science. Fuller, T. K., Mills, M.G.L., Borner, M., Laurenson, M. K., & Kat, P. W. 1992b. Long
320 ▪ R E F E R E N C E S
distance dispersal by African wild dogs in East and South Africa. Journal of African Zoology 106: 535–537. Fuller, T. K., Nicholls, T. H., & Kat, P. W. 1995. Prey and estimated food consumption of African wild dogs in Kenya. South African Journal of Wildlife Research 25: 106–110. Gasaway, W. C., Mossestad, K. T., & Stander, P. E. 1991. Food acquisition by spotted hyaenas in Etosha National Park, Namibia: Predation versus scavenging. African Journal of Ecology 29: 64–75. Gascoyne, S. C., Laurenson, M. K., Lelo, S., & Borner, M. 1993. Rabies in African wild dogs Lycaon pictus in the Serengeti region, Tanzania. Journal of Wildlife Diseases 29: 396–402. Gese, E. M., Ruff, R. L., & Crabtree, R. L. 1996. Foraging ecology of coyotes (Canis latrans): The influence of extrinsic factors and a dominance hierarchy. Canadian Journal of Zoology 74: 769–783. Ginsberg, J. R. 1993. Mapping wild dogs. IUCN/CSG Canid News 1: 2–6. Ginsberg, J. R., & Macdonald, D. W. 1990. Foxes, Wolves, Jackals and Dogs: An Action Plan for the Conservation of Canids. Gland, Switzerland: International Union for the Conservation of Nature. Ginsberg, J. R. & Woodroffe, R. 1997. Extinction risks faced by remaining wild dog populations. In: The African Wild Dog: Status Survey and Conservation Action Plan. R. Woodroffe, J. R. Ginsberg, & D. W. Macdonald, eds., pp. 11–57. Gland, Switzerland: International Union for the Conservation of Nature. Ginsberg, J. R., Alexander, K. A., Creel, S., Kat, P. W., McNutt, J. W., & Mills, M.G.L. 1995a. Handling and survivorship in the African wild dog: A survey of five ecosystems. Conservation Biology 9: 665–674. Ginsberg, J. R., Mace, G. M., & Albon, S. 1995b. Local extinction in a small and declining population: Serengeti wild dogs. Proceedings of the Royal Society of London, Series B. 262: 221–228. Giraldeau, L-A., & Caraco, T. 1993. Genetic relatedness and group size in an aggregation economy. Evolutionary Ecology 7: 429–438. Giraldeau, L-A., & Gillis, D. 1988. Do lions hunt in group sizes that maximize hunters’ daily food returns? Animal Behaviour 36: 611–613. Girman, D. J. 1996. Molecular genetic and morphometric analysis of the African wild dog, Lycaon pictus: Systematics: population genetics, and social structure. Ph.D. dissertation, University of California, Los Angeles. Girman, D. J., Kat, P. W., Mills, M.G.L., Ginsberg, J. R., Borner, M., Wilson, V., Fanshawe, J. H., Fitzgibbon, C., Lau, L., & Wayne, R. K. 1993. Molecular genetic and morphological analyses of the African wild dog (Lycaon pictus). Journal of Heredity 84: 450–459. Girman, D. J., Mills, M.G.L., Geffen, E., & Wayne, R. K. 1997. A molecular genetic analysis of social structure, dispersal, and interpack relationships of the African wild dog (Lycaon pictus). Behavioral Ecology and Sociobiology 40: 187–198. Girman, D. J., & Wayne, R. K. 1997. Genetic perspectives on wild dog conservation. In: The African Wild Dog: Status Survey and Conservation Action Plan. R. Woodroffe, J. R. Ginsberg & D. W. Macdonald, eds., pp. 7–10. Gland, Switzerland: International Union for the Conservation of Nature. Gittleman, J. L. 1989. Carnivore group living: Comparative trends. In: Carnivore
R E F E R E N C E S ▪ 321
Behavior, Ecology and Evolution, J. L. Gittleman, ed., pp. 183–207. Ithaca, NY: Cornell University Press. Goodall, J., & van Lawick, H. 1971. Innocent killers. Boston: Houghton Mifflin. Gorman, M. L., Mills, M.G.L., & French, J. 1992. Satellite tracking of the African wild dog Lycaon pictus. In: Wildlife Telemetry: Remote Monitoring and Tracking of Animals. I. G. Priede & S. M. Swift, eds., pp. 219–228. London: Ellis Horwood. Gorman, M. L., Mills, M.G.L., Raath, J., & Speakman, J. R. 1998. High hunting costs make African wild dogs vulnerable to kleptoparasitism by hyaenas. Nature 391: 479–481. Gottelli, D., Sillero-Zubiri, C., Applebaum, G. D., Roy, M. S., Girman, D. J., GarciaMoreno, J., Ostrander, E. A., & Wayne, R. K. 1994. Molecular genetics of the most endangered canid: The Ethiopian wolf Canis simensis. Molecular Ecology 3: 301– 312. Grafen, A. 1982. How not to measure inclusive fitness. Nature 298: 425–426. Green, B., J. Anderson, M., & Whateley, T. 1984. Water and sodium turnover and estimated food consumption in free-living lions (Panthera leo) and spotted hyaenas (Crocuta crocuta). Journal of Mammalogy 65: 593–599. Greene, C. H. 1986. Patterns of prey selection: Implications for predator foraging tactics. The American Naturalist 128: 824–839. Grenfell, B. T., & Dobson, A. P. 1995. Ecology of Infectious Diseases in Natural Populations. Cambridge, UK: Cambridge University Press. Grinnell, J., Packer, C., & Pusey, A. E. 1995. Cooperation in male lions: Kinship, reciprocity or mutualism? Animal Behaviour 49: 95–105. Grzimek, B. 1970. Among the Animals of Africa. London: Collins. Guo, W., Evermann, J. F., Foreyt, W. F., Knowlton, F. F., & Windberg, L. A. 1986. Canine distemper virus in coyotes: A serologic survey. Journal of the American Veterinary Medical Association 189: 1099–1100. Haas, L., Hofer, H., East, M., Wohlsein, P., Liess, B., & Barret, T. 1996. Canine distemper virus infection in Serengeti spotted hyaenas. Veterinary Microbiology 49: 147–152. Haldeman, E. G. 1962. Geology of the Rufiji Basin. Dodoma, Tanganyika: Geological Survey of Tanganyika. Hamilton, W. D. 1964. The genetical evolution of social behavior, I & II. Journal of Theoretical Biology 7: 1–52. . 1971. Geometry for the selfish herd. Journal of Theoretical Biology 31: 295– 311. Hammond, K. A., & Diamond, J. 1997. Maximal sustained energy budgets in humans and animals. Nature 386: 457–462. Hanby, J. P., & Bygott, J. D. 1979. Population changes in lions and other predators. In: Serengeti: Dynamics of an Ecosystem. A.R.E. Sinclair and M. Norton-Griffiths, eds., pp. 249–262. Chicago: University of Chicago Press. Hanby, J. P., Bygott, J. D., & Packer, C. 1995. Ecology, demography, and behavior of lions in two contrasting habitats: Ngorongoro crater and the Serengeti plains. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 315–331. Chicago: University of Chicago Press. Harcourt, A. H. 1987. Dominance and fertility among female primates. Journal of Zoology 213: 471–487.
322 ▪ R E F E R E N C E S
Hatchwell, B. J., & Komdeur, J. 2000. Ecological constraints, life history traits and the evolution of cooperative breeding. Animal Behaviour 59: 1079–1086. Hausfater, G. 1975. Dominance and reproduction in baboons (Papio cyanocephalus): A quantitative analysis. Contributions in Primatolology, Volume 7. Hegner, R., & Wingfield, J. C. 1987. Social status and circulating levels of hormones in flocks of house sparrows, Passer domesticus. Ethology 76: 1–14. Heinsohn, R. 1992. When conservation goes to the dogs. Trends in Ecology and Evolution. 7: 214–215. Heinsohn, R., & Packer, C. 1995. Complex cooperative strategies in group-territorial African lions. Science 269: 1260–1262. Henschel, J. R., & Tilson, R. L. 1988. How much does a spotted hyaena eat? Perspective from the Namib desert. African Journal of Ecology 26: 247–255. Hofer, H., & East, M. L. 1993a. The commuting system of Serengeti spotted hyaenas: How a predator copes with migratory prey. I. Social organization. Animal Behaviour 46: 547–557. . 1993b. The commuting system of Serengeti spotted hyaenas: How a predator copes with migratory prey. II. Intrusion pressure and commuters’ space use. Animal Behaviour 46: 559–574. . 1995. Population dynamics, population size, and the commuting system of Serengeti spotted hyenas. In Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 332–363. Chicago: University of Chicago Press. Hofer, H., East, M., & Campbell, K.L.I. 1993. Snares, commuting hyenas and migratory herbivores: Humans as predators in the Serengeti. Symposia of the Zoological Society of London 65: 347–366. Holekamp, K. E., & Smale, L. 1993. Ontogeny of dominance in free-living spotted hyaenas: Juvenile rank relations with other immature individuals. Animal Behaviour 46: 451–466. Holekamp, K. E., Smale, L. Berg, R., & Cooper, S. M. 1997. Hunting rates and hunting success in the spotted hyena (Crocuta crocuta). Journal of Zoology, London 242: 1–15. Holman, S. D., & Goy, R. W. 1980. Behavioral and mammary responses of adult female rhesus to strange infants. Hormones and Behavior 14: 348–357. Holt, R. D., & Polis, G. A. 1997. A theoretical framework for intraguild predation. American Naturalist 149: 745–764. Hoogland, J. L. 1995. The Black-Tailed Prairie Dog: Social Life of a Burrowing Mammal. Chicago: University of Chicago Press. Horvitz, C. C., Schemske, D. W., & Caswell, H. 1997. The “importance” of life history stages to population growth. In: Structured Population Models in Marine, Terrestrial and Freshwater Systems. S. Tuljapurkar & H. Caswell, eds., pp. 247– 272. New York: Chapman & Hall. Houston, A., Clark, C. W., McNamara, J., & Mangel, M. 1988. Dynamic models in behavioural and evolutionary ecology. Nature 332: 29–34. Houston, D. 1969. Hunting dogs of East Africa. Animals 11: 506–509. Hrdy, S. B. 1987. Sex-biased parental investment among primates and other mammals: A critical evaluation of the Trivers-Willard hypothesis. In: Child Abuse and Neglect: Biosocial Dimensions. R. Gelles & J. Lancaster, eds., pp. 97–147. New York: Aldine Publishers.
R E F E R E N C E S ▪ 323
Hudson, P. J., Dobson, A. P., & Newborn, D. 1998. Prevention of population cycles by parasite removal. Science 282: 2256–2259. Hubbard, W. D. 1954. Wild dogs of Africa. Natural History 12/54: 454–455. Huggard, D. J. 1993. Prey selectivity of wolves in Banff National Park, I. Prey species. Canadian Journal of Zoology 71: 130–139. Hunter, J. A. 1960. The rapacious wild dog. Journal of the Kenya Wildlife Society 2: 42. Hurlbert, S. H. 1984. Pseudoreplication and the design of ecological field experiments. Ecological Monographs 54: 187–211. Inman, A. J., & Krebs, J. R. 1987. Predation and group living. Trends in Ecology and Evolution. 2: 31–32. Isbell, L. A., Cheney, D. L., & Seyfarth, R. M. 1990. Costs and benefits of home range shifts among vervet monkeys (Cercopithecus aethiops) in Amboseli National Park, Kenya. Behavioral Ecology and Sociobiology 27: 351–358. Jennions, M. J., & Macdonald, D. W. 1994. Cooperative breeding in mammals. Trends in Ecology and Evolution 9: 89–93. Johnsingh, A.J.T. 1982. Reproductive and social behavior of the dhole, Cuon alpinus (Canidae). Journal of Zoology, London 198: 443–463. Johnson, M. L., & Gaines, M. S. 1990. Evolution of dispersal: Theoretical models and empirical tests using birds and mammals. Annual Review of Ecology and Systematics 21: 449–480. Johnson, M. R., Boyd, D. K., & Pletscher, D. H. 1994. Serologic investigations of canine parvovirus and canine distemper in relation to wolf (Canis lupus) pup mortalities. Journal of Wildlife Diseases 30: 270–273. Johnson, W. E., & Franklin, W. L. 1994. Spatial resource partitioning by sympatric grey fox (Dusicyon griseus) and culpeo fox (Dusicyon culpaeus) in southern Chile. Canadian Journal of Zoology 72: 1788–1793. Johnson, W. E., Fuller, T. K., & Franklin, W. L. 1996. Sympatry in canids: a review and assessment. In: Carnivore Behavior, Ecology and Evolution. J. L. Gittleman, ed., pp. 189–218. Ithaca, NY: Cornell University Press. Karanth, K. U., & Sunquist, M. E. 1995. Prey selection by tiger, leopard and dhole in tropical forests. Journal of Animal Ecology 64: 439–450. Kat, P. W., Alexander, K. A., Smith, J. S., & Munson, L. 1995. Rabies and wild dogs in Kenya. Proceedings of the Royal Society of London, Series B 262: 229–233. Keane, B., Waser, P. M., Creel, S., Creel, N. M., Elliott, L. F., & Minchella, D. J. 1994. Subordinate reproduction in dwarf mongooses. Animal Behaviour 47:65–75. Keane, B., Waser, P. M., & Creel, S. 1997. No evidence of inbreeding avoidance or inbreeding depression in a social carnivore. Behavioral Ecology Vol. : 480–489. Keller, L., & Reeve, H. K. 1994. Partitioning of reproduction in animal societies. Trends in Ecology and Evolution 9: 98–102. Kennedy, D. J. 1988. An outbreak of rabies in northwestern Zimbabwe Veterinary Record 122: 129–133. Kenward, R. E. 1978. Hawks and doves: Factors affecting success and selection in goshawk attacks on woodpigeons. Journal of Animal Ecology 47: 449–460. Keverne, E., Meller, R., & Eberhart, J. 1982. Dominance and subordination: Concepts or physiological states? In: Advanced Views in Primate Biology. V. Chiarelli & J. Corrucinni, eds., pp. 213–247. Berlin: Springer-Verlag. Kie, J. G., Baldwin, J. A., & Evans, C. J. 1994. CALHOME Home Range Analysis
324 ▪ R E F E R E N C E S
Program User’s Manual. Fresno, CA: United States Forest Service Pacific Southwest Research Station. Kleiman, D. G. 1980. The socioecology of captive propagation. In: Conservation Biology: An Evolutionary-Ecological Perspective. M. E. Soule & B. A. Wilcox, eds., pp. 243–261. Sunderland, MA: Sinauer. Koehler, G. M., & Hornocker, M. G. 1991. Seasonal resource use among mountain lions, bobcats, and coyotes. Journal of Mammalogy 72: 391–396. Krause, J., & Godin, J-G. J. 1995. Predator preferences for attacking particular prey group sizes: consequences for predator hunting success and prey predation risk. Animal Behaviour 50: 465–473. Kruuk, H. 1972. The Spotted Hyaena: A Study of Predation and Social Behavior. Chicago: University of Chicago Press. . 1975. Functional aspects of social hunting in carnivores. In: Function and Evolution in Behavior. G. Baerends, C. Beer, & A. Manning, eds., pp. 119–141. Oxford, UK: Oxford University Press. Kruuk, H., & Macdonald, D. W. 1985. Group territories of carnivores: Empires or enclaves? In: Behavioural Ecology: Ecological Consequences of Adaptive Behaviour. R. M. Sibly & R. H. Smith, eds., pp. 521–536. Oxford, UK: Blackwell Scientific Publications. Kruuk, H., & Turner, M. 1967. Comparative notes on predation by lion, leopard, cheetah and wild dog in the Serengeti area, Tanzania. Mammalia 31: 1–27. Kuhme, W. D. 1965. Communal food distribution and division of labour in African hunting dogs. Nature 205: 443–444. Kurten, B. 1968. Pleistocene Mammals of Europe. Chicago: Aldine. Lacey, E. A., & Sherman, P. W. 1991. Social organization of naked mole-rat colonies: Evidence for divisions of labor. In: The Biology of the Naked Mole-Rat. P. W. Sherman, J.U.M. Jarvis, & R. D. Alexander, eds., pp. 275–336. Princeton, NJ: Princeton University Press. Lack, D. 1954. The Natural Regulation of Animal Numbers. Oxford, UK: Oxford University Press. . 1971. Ecological Isolation in Birds. Oxford, UK: Blackwell Scientific Publications. Lacy, R. C., & Kreeger, T. 1992. VORTEX Users Manual. A stochastic simulation of the extinction process. Apple Valley, MN: IUCN SSC Captive Breeding Specialist Group. Lande, R. 1988. Genetics and demography in biological conservation. Science 241: 1455–1460. . 1995. Mutation and conservation. Conservation Biology 9: 782–791. Lande, R., & Barrowclough, G. F. 1987. Effective population size, genetic variation, and their use in population management. In: Viable Populations for Conservation. M. E. Soule, ed., pp. 87–123. Cambridge, UK: Cambridge University Press. Laurenson, M. K. 1995. Implications of high offspring mortality for cheetah population dynamics. In: Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 385–399. Chicago: University of Chicago Press. Laurenson, M. K., Borner, M., Lelo, S., & Burrows, R. 1990. Conservation status of the wild dog (Lycaon pictus) in the Serengeti ecosystem. Serengeti Wildlife Research Center, Seronera, Tanzania.
R E F E R E N C E S ▪ 325
Laurenson, M. K., Wielebnowski, N., & Caro, T. M. 1995. Extrinsic factors and juvenile mortality in cheetahs. Conservation Biology. 9: 1329–1331. Lawton, J. H., & Hassell, M. P. 1981. Asymmetrical competition in insects. Nature 289, 7: 93–795. Leader-Williams, N., Kayera, J. A., & Overton, G. L. 1996. Tourist Hunting in Tanzania. Gland, Switzerland: International Union for the Conservation of Nature. Leakey, R. 1983. One Life. Salem: Salem House. Le Boeuf, B. J., & Reiter, J. 1988. Lifetime reproductive success in northern elephant seals. In: Reproductive Success. T. H. Clutton-Brock, ed., pp. 344–362. Chicago: University of Chicago Press. Legendre, S., & Clobert, J. 1995. ULM, a software for conservation and evolutionary biologists. Journal of Applied Statistics 22: 817–834. Lima, S. L. 1995a. Back to the basics of anti-predatory vigilance: The group-size effect. Animal Behaviour 49: 11–20. . 1995b. Collective detection of predatory attack by social foragers: Fraught with ambiguity? Animal Behavior 50: 1097–1108. Lima, S. L., & Dill, L. M. 1990. Behavioral decisions made under the risk of predation: A review and prospectus. Canadian Journal of Zoology 68: 619–640. Lima, S. L., & Zollner, P. A. 1996. Anti-predatory vigilance and the limits to collective detection: Visual and spatial separation between foragers. Behavioral Ecology and Sociobiology 38: 355–363. Lombardi, C. M., & Hurlbert, S. H. 1996. Sunfish cognition and pseudoreplication. Animal Behaviour 52: 419–422. Maas, B. 1993. Bat-eared fox behavioural ecology and the incidence of rabies in the Serengeti National Park. Onderstepoort Journal of Veterinary Research 60: 389–393. MacArthur, R. H. 1972. Geographical Ecology: Patterns in the Distribution of Species. Princeton, NJ: Princeton University Press. Macdonald, D. W. 1980. Rabies and Wildlife: A Biologist’s Perspective. Oxford, UK: Oxford University Press. . 1983. The ecology of carnivore social behaviour. Nature 301: 379–384. . 1992. Cause of wild dog deaths. Nature 360: 633–634. Macdonald, D. W., & Moehlman, P. D. 1982. Cooperation, altruism and restraint in the reproduction of carnivores. Perspectives in Ethology 5: 433–467. Maddock, A. H. 1993. Wild dog research in Hluhluwe-Umfolozi Park, South Africa. Licaone 2: 10–11 Maddock, A. H., & Mills, M.G.L. 1994. Population characteristics of African wild dogs Lycaon pictus in the Eastern Transvaal lowveld, South Africa, as revealed through photographic records. Biological Conservation 67: 57–62. Maddock, L. 1979. The “migration” and grazing succession. In: Serengeti: Dynamics of an Ecosystem. A.R.E. Sinclair & M. Norton-Griffiths, eds., pp. 104–129. Chicago: University of Chicago Press. Magembe, S. R. 1985. Epidemiology of rabies in the United Republic of Tanzania. In: Rabies in the Tropics. E. Kuwert, C. Merieux, H. Koprowski, and K. Bagel, eds., pp. 392–398. Berlin: Springer-Verlag. Malcolm, J. R. 1979. Social organisation and communal rearing in African wild dogs. Ph.D., Dissertation, Harvard University, Cambridge, Massachusetts. . 1980. Food caching by African wild dogs (Lycaon pictus). Journal of Mammalogy 61: 743–744.
326 ▪ R E F E R E N C E S
Malcolm, J. R., & van Lawick, H. 1975. Notes on wild dogs (Lycaon pictus) hunting zebras. Mammalia 39: 231–240. Malcolm, J. R., & Marten, K. 1982. Natural selection and the communal rearing of pups in African wild dogs (Lycaon pictus). Behavioral Ecology and Sociobiology 10:1–13. Mangel, M., & Clark, C. W. 1988. Dynamic Modeling in Behavioral Ecology. Princeton, NJ: Princeton University Press. Manogue, K. R. 1975. Dominance status and adrenocortical reactivity to stress in squirrel monkeys (Saimiri sciureus). Primates 14: 457–463. Matthew, W. D. 1930. The phylogeny of dogs. Journal of Mammalogy 11: 117–138. Maugham, R.C.F. 1914. Wild Game in Zambesia. London: John Murray. May, R. M. 1985. Host-parasite associations: Their population biology and population genetics. In: Ecology and Genetics of Host-Parasite Interactions. R. M. Anderson, ed., pp. 243– 262. London: Linnean Society of London. Mays, N. A., Vleck, C. M., & Dawson, J. 1991. Plasma lutienizing hormone, steroid hormones, behavioral role and nesting stage in the cooperatively breeding Harris’s Hawk (Parabuteo unicinctus). Auk 108: 619–637. McCormick, A. E. 1983. Canine distemper in African cape hunting dogs, Lycaon pictus—possibly vaccine induced. Journal of Zoo Animal Medicine 14: 66–71. McEwen, B. S., & Schmeck, H. M. 1994. The Hostage Brain. New York: Rockefeller University Press. McNutt, J. W. 1995. Sociality and dispersal in African wild dogs, Lycaon pictus. Ph.D. dissertation, University of California, Davis. . 1996. Sex-biased dispersal in African wild dogs, Lycaon pictus. Animal Behaviour 52: 1067–1077. Mech, L. D. 1970. The Wolf: The Ecology and Behavior of an Endangered Species. Minneapolis: University of Minnesota Press. . 1977. Wolf pack buffer zones as prey reservoirs. Science 198: 320–321. . 1999. Alpha status, dominance, and division of labor in wolf packs. Canadian Journal of Zoology 77: 1196–1203. Mech, L. D., & Goyal, S. M. 1993. Canine parvovirus effect on wolf population change and pup survival. Journal of Wildlife Diseases 29: 330–333. Menges, E. 1990. Stochastic modelling of extinction in plant communities. In: Conservation Biology: The Theory and Practice of Nature Conservation, Preservation and Management. P. L. Fiedler & S. K. Jain, eds., pp. 253–276. New York: Chapmam & Hall. Milinski, M. 1977. Experiments on the selection by predators against spatial oddity of their prey. Zeitschrift f¨ur Tierpsychologie 43: 311–325. Mills, L. S., Hayes, S. G., Baldwin, C., Wisdom, M. J. Citta, J., Matson, D. J., & Murphy, K. 1996. Factors leading to different viability predictions for a grizzly bear data set. Conservation Biology 10: 863–873. Mills, M.G.L. 1984. The comparative behavioural ecology of the brown hyaena (Hyaena brunnea) and the spotted hyaena (Crocuta crocuta) in the southern Kalahari. Koedoe (supplement) 27: 237–247. . 1985. Hyaena survey of Kruger National Park: August–October 1984. IUCN SSC Hyaena Specialist Group Newsletter 2: 15–25. . 1988. Wild dogs and wild dog research in the Kruger National Park. Quagga 21/22: 11–13.
R E F E R E N C E S ▪ 327
. 1990. Kalahari Hyaenas: Comparative Behavioural Ecology of Two Species. London: Unwin Hyman. . 1992. A comparison of methods used to study food habits of large African carnivores. In: Wildlife 2001: Populations. D. R. McCullough & R. H. Barrret, eds., pp. 1112–1124. London: Elsevier. . 1993. Social systems and behaviour of the African wild dog Lycaon pictus and the spotted hyaena Crocuta crocuta with special reference to rabies. Onderstepoort Journal of Veterinary Research 60: 405–409. . 1996. Methodological advances in capture, census, and food-habits studies of large African carnivores. In: Carnivore Behavior, Ecology and Evolution, Vol. 2. J. L. Gittleman, ed., pp. 223–242. Ithaca, NY: Cornell University Press. Mills, M.G.L., & Biggs, H. C. 1993. Prey apportionment and related ecological relationships between large carnivores in Kruger National Park. Symposia of the Zoological Society of London 65: 253–268. Mills, M.G.L., & Gorman, M. L. 1997. Factors affecting the density and distribution of wild dogs in the Kruger National Park. Conservation Biology 11: 1397–1406. Mills, M.G.L., & Mills, M.E.J. 1982. Factors affecting the movement patterns of brown hyaenas, Hyaena brunnea, in the southern Kalahari. South African Journal of Nature Research 12: 111–117. Mills, M.G.L., & Shenk, T. M. 1992. Predator-prey relationships: The impact of lion predation on wildebeest and zebra populations. Journal of Animal Ecology 61: 693–702. Moehlman, P. D. 1983. Socioecology of silver-backed and golden jackals, Canis mesomelas and C. aureus. In: Recent Advances in the Study of Mammalian Behavior. J. F. Eisenberg and D. G. Kleiman, eds., pp. 423–453. Lawrence, KS: American Society of Mammalogists. . 1989. Intraspecific variation in canid social systems. In: Carnivore Behavior, Ecology and Evolution. J. L. Gittleman, ed., pp. 143–163. Ithaca, NY: Cornell University Press. Monaghan, E. P., & Glickman, S. E. 1992. Hormones and aggressive behavior. In: Behavioral Endocrinology. J. Becker, M. Breedlove, & D. Crews, eds., pp. 97–130. Cambridge, MA: Massachusetts Institute of Technology Press. Monfort, S. L., Arthur, N. P., & Wildt, D. E. 1990. Monitoring ovarian function and pregnancy by evaluating excretion of urinary oestrogen conjugates in semi-freeranging Przewalski’s horses (Equus przewalksii). Journal of Reproduction and Fertility 91: 155–164. Monfort, S. L., Brown, J. L., & Wildt, D. E. 1993. Episodic and seasonal rhythms of cortisol secretion in male Eld’s deer (Cervus eldi thamin). Journal of Endocrinology 138: 41–49. Monfort, S. L., Wasser, S. K., Mashburn, K. L., Burke, M. A., Brewer, B. A., & Creel, S. 1997. Steroid metabolism and validation of noninvasive endocrine monitoring in the African wild dog (Lycaon pictus). Zoo Biology 16: 533–548. Moore, J. 1993. Inbreeding and outbreeding in primates: What’s wrong with the dispersing sex? In: The Natural History of Inbreeding and Outbreeding. N. W. Thornhill, ed., pp. 392–426. Chicago: University of Chicago Press. Moore, M. C., Thompson, C. W., & Marler, C. A. 1991. Reciprocal changes in corticosterone and testosterone levels following acute and chronic handling stress in the tree lizard, Urosaurus ornatus. General and Comparative Endocrinology 81: 217–226.
328 ▪ R E F E R E N C E S
Munck, A. P., Guyre, P. M., & Holbrook, N. J. 1984. Physiological functions of glucocorticoids in stress and their relation to pharmacological actions. Endocrine Reviews 5: 25–44. Murie, A. 1944. The Wolves of Mount McKinley. Washington, DC: U.S. Government Printing Office. Reprinted 1971. Murphy, K. M. 1998. The ecology of the cougar (Puma concolor) in the Northern Yellowstone ecosystem: Interactions with prey, bears, and humans. Ph.D. dissertation, University of Idaho, Moscow. Neill, S. R., & Cullen, J. M. 1974. Experiments on whether schooling by their prey affects the behaviour of cephalopod and fish predators. Journal of Zoology, London 172: 549–569. Neitz, W. O., & Thomas, A. D. 1938. Rickettsiosis in the dog. Journal of the South African Veterinary Medical Association 9: 166–174. Nunney, L., & Elam, D. R. 1994. Estimating the effective population size of conserved populations. Conservation Biology 8: 175–184. Ogutu, J. O., & Dublin, H. T. 1998. The response of lions and spotted hyaenas to sound playbacks as a technique for estimating population size. African Journal of Ecology 36: 83–95. O’Riain, M. J., Bennett, N. C., & Brotherton, P.N.M. 2000. Reproductive suppression and inbreeding avoidance in wild populations of cooperatively breeding meerkats (Suricata suricatta). Behavioral Ecology and Sociobiology 48: 471–477. Osterhaus, A.D.M.E., De Swart, R. L., Vos, H. W., Ross, P. S., Kenter, M.J.J., & Barrett, T. 1995. Morbillivirus infections of aquatic mammals: Newly identified members of the genus. Veterinary Microbiology 44: 219–227. Packard, J. M., Seal, U. S., Mech, L. D., & Plotka, E. D. 1985. Causes of reproductive failure in two family groups of wolves. Zeitschrift f¨ur Tierpsychologie 68: 24–40. Packer, C. 1986. The ecology of sociality in felids. In: Ecological Aspects of Social Evolution. D. I. Rubenstein & R. W. Wrangham, eds., pp. 429–451 Princeton, NJ: Princeton University Press. . 1990. Serengeti lion survey: Report to TANAPA, SWRI, Mweka and the Game Department. Serengeti Wildlife Research Centre, Seronera, Tanzania. Packer, C., & Caro, T. M. 1997. Foraging costs in social carnivores. Animal Behaviour 54: 1317–1318. Packer, C. & Pusey, A. E. 1982. Cooperation and competition within coalitions of male lions: Kin selection or game theory? Nature 296: 740–742. . 1983a. Dispersal, kinship and inbreeding in African lions. In: The Natural History of Inbreeding and Outbreeding. N. W. Thornhill, ed., pp. 375–391. Chicago: University of Chicago Press. . 1983b. Adaptations of female lions to infanticide by incoming males. American Naturalist 121: 716–728. . 1983c. Male takeovers and female reproductive parameters: A simulation of oestrus synchrony in lions. Animal Behaviour 3: 334–340. . 1987. Intrasexual cooperation and the sex ratio in African lions. American Naturalist 130: 636–642. Packer, C., & Ruttan, L. 1988. The evolution of cooperative hunting. American Naturalist 132: 159–198. Packer, C., Herbst, L., Pusey, A. E., Bygott, J. D., Hanby, J. P., Cairns, S., & Borgerhoff-Mulder, M. 1988. Reproductive success of lions. In: Reproductive Success. T. H. Clutton-Brock, ed., pp. 363–383. Chicago: University of Chicago Press.
R E F E R E N C E S ▪ 329
Packer, C., Scheel, D., & Pusey, A. E. 1990. Why lions form groups: Food is not enough. American Naturalist 136: 1–19. Packer, C., Pusey, A. E., Rowley, H., Gilbert, D. A., Martenson, J., & O’Brien, S. J. 1991. Case study of a population bottleneck: Lions of the Ngorongoro Crater. Conservation Biology 5: 219–230. Packer, C., Collins, D. A., Sindimwo, A., & Goodall, J. 1995. Reproductive constraints on aggressive competition in female baboons. Nature 373: 60–63. Palomares, F., & Caro, T. M. 1999. Interspecific killing among mammalian carnivores. American Naturalist 153: 492–508. Palomares, F., Gaona, P., Ferreras, P., & Delibes, M. 1995. Positive effects on game species of top predators by controlling smaller predator populations: An example with lynx, mongooses and rabbits. Conservation Biology 9: 295–305. Palomares, F., Ferreras, P., Fedriani, J. M., & Delibes, M. 1996. Spatial relationships between Iberian lynx and other carnivores in an area of south-western Spain. Journal of Animal Ecology 33: 5–13. Paquet, P. C. 1991. Winter spatial relationships of wolves and coyotes in Riding Mountain National Park, Manitoba. Journal of Mammalogy 72: 397–401. . 1992. Prey use strategies of sympatric wolves and coyotes in Riding Mountain National Park, Manitoba. Journal of Mammalogy 73: 337–343. Pascual, M. A., Kareiva, P., & Hilborn, R. 1997. The influence of model structure on conclusions about the viability and harvesting of Serengeti wildebeest. Conservation Biology 11: 966–976. Pence, D. B. 1995. Disease and Coyotes in Texas. In: Coyotes in the Southwest: A Compendium of our Knowledge, pp. 17–22. Austin: Texas Parks and Wildlife Department. Pereira, M. E., & Izard, M. K. 1989. Lactation and care for unrelated infants in forest-living rintailed lemurs. American Journal of Primatology 18: 101–108. Peterson, R. O. 1995. Wolves as interspecific competitors. In: Ecology and Conservation of Wolves in a Changing World. L. U. Carbyn, S. H. Fritts, & D. R. Seip, eds., pp. 315–323. Edmonton: Canadian Circumpolar Institute. Peterson, R. O., & Krumenaker, R. J. 1989. Wolves approach extinction on Isle Royale: A biological and policy conundrum. George Wright Forum 6: 10–15. Phillips, M. K., & Smith, D. W. 1996. The Wolves of Yellowstone. Stillwater, MN: Voyageur Press. . 1997. Yellowstone wolf project biennial report 1995–1996. Yellowstone Center for Resources Document YCR-NR-97–4. Yellowstone, Wyoming, 24 pp. Pienaar, U. de V. 1969. Predator-prey relationships amongst the larger mammals of the Kruger National Park. Koedoe 12: 108–176. Pierce, M. A., Laurenson, M. K., & Gascoyne, S. C. 1995. Hepatozoonosis in cheetahs and wild dogs in the Serengeti ecosystem. African Journal of Ecology 33: 273–275. Pitcher, T. J., & Parrish, J. K. 1993. Functions of shoaling behavior in teleosts. In: Behaviour of Teleost Fishes. T. J. Pitcher, ed., pp. 294–337. London: Chapman and Hall. Planning and Assessment for Wildlife Management. 1995. Returns from tourist hunting in Tanzania. In: Tourist Hunting in Tanzania. N. Leader-Williams, J. A. Kayera, & G. L. Overton, eds., pp. 71–80. Dar es Salaam: Tanzania Department of Wildlife. Poini, A., & Fletcher, T. 1994. Plasma levels of androgens and gonadal development
330 ▪ R E F E R E N C E S
of breeders and helpers in the bell miner (Manorhina melanorhys). Behavioral Ecology and Sociobiology 34: 31–41. Polis, G. A., Myers, C., & Holt, R. D. 1989. The evolution and ecology of intra-guild predation: Competitors that eat each other. Annual Review of Ecology and Systematics 20: 297–330. Pulliam, H. R. 1974. On the theory of optimal diets. The American Naturalist 108: 59–75. Pulliam, H. R., & Caraco, T. 1984. Living in groups: Is there an optimal group size? In: Behavioral Ecology. J. R. Krebs & N. B. Davies, eds., pp. 122–147. Sunderland, MA: Sinauer Associates. Pusey, A. E., & Packer, C. 1987. The evolution of sex-biased dispersal in lions. Behaviour 101: 275–310. Pusey, A. E., & Wolf, M. 1996. Inbreeding avoidance in animals. Trends in Ecology and Evolution 11: 201–206. Pusey, A. E., Williams, J., & Goodall, J. 1997. The influence of dominance rank on the reproductive success of female chimpanzees. Science 277: 828–831. Queller, D. C., & Goodnight, K. F. 1989. Estimating relatedness using genetic markers. Evolution 43: 258–275. Rabenold, K. N. 1990. Campylorhynchus wrens: The ecology of delayed dispersal and cooperation in the Venezuelan savanna. In: Cooperative Breeding in Birds. P. B. Stacey & W. D. Koenig, eds., pp. 157–196. Cambridge, UK: Cambridge University Press. Radinsky, L. 1973. Evolution of the canid brain. Brain, Behavior and Evolution 7: 169–202. Ralls, K., Ballou, J. D., & Templeton, A. R. 1988. Estimates of lethal equivalents and the cost of inbreeding in mammals. Conservation Biology 2: 185–193. Ralls, K., & White, P. J. 1995. Predation on San Joaquin kit foxes by large canids. Journal of Mammalogy 76: 723–729. Rapaport, L. & Haight, J. 1987. Some observations on allomaternal caretaking among captive Asian elephants (Elaphus maximus). Journal of Mammalogy 68: 438–442. Rasmussen, G. 1996. Highly endangered painted hunting dogs used as an excuse for stock loss. Farmer (Zimbabwe) 66: 30–31. . 1998. Coloured collars for painted dogs. Zimbabwe Wildlife 10/11: 20. Reeve, H. K., & Ratnieks, F. L. 1993. Queen-queen conflicts in polygynous societies: Mutual tolerance and reproductive skew. In: Queen Number and Sociality in Insects, L. Keller, ed., pp. 45–85. Oxford, UK: Oxford University Press. Reeve, H. K., Emlen, S. T., & Keller, L. 1998. Reproductive sharing in animal societies: Reproductive incentives or incomplete control by dominant breeders? Behavioral Ecology 9: 267–278. Reich, A. 1978. A case of close inbreeding in the African wild dog Lycaon pictus in the Kruger National Park. Koedoe 21: 119–123. . 1981. The behavior and ecology of the African wild dog, Lycaon pictus, in the Kruger National park. Ph.D. Dissertation, Yale University, New Haven, CT. Reimers, T. J., Gowan, R. G., Davidson, H. P., & Colby, E. D. 1981. Validation of assays for triiodothyronine, thyroxine, and hydrocortisone (cortisol) in canine, feline and equine sera. American Journal of Veterinary Research 42: 2016–2021. Reyer, H-U, Dittami, J. P., & Hall, M. R. 1986. Avian helpers at the nest: are they psychologically castrated? Ethology 71: 216–228.
R E F E R E N C E S ▪ 331
Roberts, G. 1996. Why individual vigilance dcelines as group size increases. Animal Behaviour 51: 1077–1086. Rodgers, A. 1974. The lion (Panthera leo, Linn.) population of the Eastern Selous Game Reserve. East African Wildlife Journal 12: 313–317. Rodgers, W. A. 1979. The ecology of large herbivores in the miombo woodlands of south east Tanzania. Ph.D. Dissertation, University of Nairobi, Nairobi, Kenya. Roelke-Parker, M. E., Munson, L., Packer, C., Kock, R., Cleaveland, S., Carpenter, M., O’Brien, S. J., Popischil, A., Hofman-Lehman, R., Lutz, H., Mwamengele, G.L.M., Mgasa, M. N., Machange, G. A., Summers, B. A., & Appel, M.J.G. 1996. A canine distemper virus epidemic in Serengeti lions (Panthera leo). Nature 379: 441–445. Romero-Herrera, A. E., Darbre, P. A., & Lehman, H. 1976. The myoglobin of the Cape hunting dog (Lycaon pictus). Biochimica Physica Acta 420: 350–357. Rood, J. P. 1987. Dispersal and intergroup transfer in the dwarf mongoose. In: Mammalian Dispersal Patterns: The Effects of Social Structure on Population Genetics. B. D. Chepko-Sade & Z. T. Halpin, eds., pp. 85–103. Chicago: University of Chicago Press. . 1990. Group size, survival, reproduction, and routes to breeding in dwarf mongooses. Animal Behaviour 39: 566–572. Rudnai, J. A. 1970. Social behavior and feeding habits of lion in Nairobi National Park. M.Sc. Thesis, University of Nairobi, Nairobi, Kenya. . 1973. The Social Life of the Lion. Wallingford, PA: Washington Square East Publishers. Runyoro, V. A., Hofer, H., Chausi, E. B., & Moehlman, P. D. 1995. Long-term trends in the herbivore populations of the Ngorongoro crater, Tanzania. In Serengeti II: Dynamics, Management and Conservation of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 146–168. Chicago: University of Chicago Press. Rushworth, J. E. 1973. Wankie National Park Notes, Nos. 6 & 7, cited in Childes (1988). Sachs, R. 1967. Liveweights and body measurements of Serengeti game animals. East African Wildlife Journal 5: 24–36. Saito, M., & Kato, H. 1985. Circadian anticipatory response to food intake in behavioral and endocrine functions. In: Circadian Clocks and Zeitgebers. T. Hioshige & K. Houma, eds., pp. 146–156. Sapporo, Japan: Hokkaido University Press. Saltzman, W., Schultz-Darken, N. J., Scheffler, G., Wegner, F. H., & Abbott, D. H. 1994. Social and reproductive influences on plasma cortisol in female marmoset monkeys. Physiology & Behavior 56: 801–810. Saltzman, W., Schultz-Darken, N. J., Wegner, F., Wittwer, D. J., & Abbott, D. H. 1998. Suppression of cortisol levels in subordinate female marmosets: Reproductive and social contributions. Hormones & Behavior 33: 58–74. Sapolsky, R. M. 1985. Stress-induced suppression of testicular function in the wild baboon: Role of glucocorticoids. Endocrinology 116: 2273–2278. . 1990. Adrenocortical function, social rank, and personality among wild baboons. Biological Psychiatry 28: 862–878. . 1992a. Stress, the aging brain, and the mechanisms of neuron death. Cambridge, MA: MIT Press. . 1992b. Neuroendocrinology of the stress response. In: Behavioral Endocrinology. J. B. Becker, S. M. Breedlove, & D. Crews, eds., pp. 287–324. Cambridge, MA: MIT Press.
332 ▪ R E F E R E N C E S
Sapolsky, R. M. 1994. Why Zebras Don’t Get Ulcers. New York: W. H. Freeman. Sargeant, A. B., & Allen, S. H. 1989. Observed interactions between coyotes and red foxes. Journal of Mammalogy 70: 631–633. Sargeant, A. B., Allen, S. H., & Hastings, J. O. 1987. Spatial relationships between sympatric coyotes and red foxes in North Dakota. Journal of Wildlife Management 51: 285–293. Savage, R.J.G. 1978. Carnivora. In: Evolution of African Mammals. V. J. Maglio & H.B.S. Cooke, eds., pp. 249–267. Cambridge, MA: Harvard University Press. Savage, D. E., & Russell, D. E. 1983. Mammalian Paleofaunas of the World. Boston: Addison-Wesley. Schaller, G. B. 1972. The Serengeti Lion. Chicago: University of Chicago Press. . 1996. Carnivores and conservation biology. In: Carnivore Behavior, Ecology and Evolution, Vol. 2. J. L. Gittleman, ed., pp. 1–10. Ithaca, NY: Cornell University Press. Scheel, D. 1993. Profitability, encounter rates and prey choice of African lions. Behavioral Ecology 4:90–97. Scheel, D., & Packer, C. 1991. Group hunting behaviour of lions: A search for cooperation. Animal Behaviour 41: 697–709. . 1995. Variation in predation by lions: Tracking a movable feast. In: Serengeti II: Dynamics, Conservation and Management of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 299–314. Chicago: University of Chicago Press. Scheepers, J. L., & Venzke, K.A.E. 1995. Attempts to reintroduce African wild dogs Lycaon pictus into Etosha National Park, Namibia. South African Journal of Wildlife Research 25: 138–140. Schemske, D. W., Husband, B. C., Ruckelshuas, M. H., Goodwillie, C., Parker, I. M., & Bishop, J. G. 1994. Evaluating apporaches to the conservation of rare and endangered plants. Ecology 75: 584–606. Schmidt, L. G., Bradshaw, S. D., & Follett, B. K. 1991. Plasma levels of luteinizing hormone and androgens in relation to age and breeding status among cooperatively breeding Australian magpies (Gymnorhina tibicen Latham). General and Comparative Endocrinology 83: 48–55. Schmidt, P. A., & Mech, L. D. 1997. Wolf pack size and food acquisition. American Naturalist 150: 513–517. Schoech, S. J., Mumme, R. L., & Moore, M. C. 1991. Reproductive endocrinology and mechanisms of breeding inhibition in cooperatively breeding Florida scrub jays (Aphelocoma c. coerulescens). Condor 93: 354–364. Schoech, S. J., Mumme, R. L., & Wingfield, J. C. 1996. Delayed breeding in the cooperatively breeding Florida scrub-jay (Aphelocoma coerulescens): Inhibition or the absence of stimulation? Behavioral Ecology and Sociobiology 39: 77–90. Schoener, T. W. 1971. Theory of feeding strategies. Annual Review of Ecology and Systematics. 2: 369–404. . 1983. Field experiments on interspecific competition. American Naturalist 122: 240–285. Schwabl, H. 1995. Individual variation of the acute adrenocortical response to stress in the white-throated sparrow. Zoology 99: 113–120. Schwabl, H., Bairlein, F., & Gwinner, E. 1991. Basal and stress-induced corticosterone levels of garden warblers, Sylvia borin, during migration. Journal of Comparative Physiology B. 161: 576–580.
R E F E R E N C E S ▪ 333
Scott, J. 1991. Painted Wolves: Wild Dogs of the Serengeti-Mara. London: Hamish Hamilton. Seaman, D. E., & Powell, R. A. 1996. An evaluation of the accuracy of kernel density estimators for home range analysis. Ecology 77: 2075–2085. Selous, F. C. 1908. African Nature Notes and Reminiscences. London: Macmillan. Severre, E.L.M. 1995. Setting quotas for tourist hunting in Tanzania. In: Tourist Hunting in Tanzania. N. Leader-Williams, J. A. Kayera & G. L. Overton, eds., pp. 57–58. Dar es Salaam: Tanzania Department of Wildlife, Planning and Assessment for Wildlife Management. Sherman, P. W., Jarvis, J.U.M., & Alexander, R. D. 1991. The Biology of the Naked Mole Rat. Princeton, NJ: Princeton University Press, Princeton. Sherman, P. W., Lacey, E. A., Reeve, H. K., & Keller, L. 1995. The eusociality continuum. Behavioral Ecology 6: 102–108. Sillero-Zubiri, C., & Gottelli, D. 1992. Population ecology of spotted hyaena in an equatorial mountain forest. African Journal of Ecology 30: 292–300. Sillero-Zubiri, C., King, A. A., & Macdonald, D. W. 1996. Rabies and mortality in Ethiopian wolves (Canis simensis). Journal of Wildlife Diseases 32: 80–86. Sinclair, A.R.E. 1995. Serengeti past and present. In: Serengeti II: Dynamics, Conservation and Management of an Ecosystem. A.R.E. Sinclair & P. Arcese, eds., pp. 3– 30. Chicago: University of Chicago Press. Skinner, J. D., Fairall, N., & Bothma, J. 1977. South African Red Data Book—Large Mammals. Johannesburg: South African National Scientific Programmes. Smith, J. S., Yager, P. A., & Baer, G. M. 1996. A rapid fluorescent focus inhibition test (RFFIT) for determining rabies virus neutralizing antibody. In: Laboratory Techniques in Rabies, pp. 181–192. Geneva, Switzerland: World Health Organization. Smith, M. S., & McDonald, L. E. 1974. Serum levels of luteinizing hormone and progesterone during the estrous cycle, pseudopregnancy and pregnancy in the dog. Endocrinology 94: 404–412. Smithers, R.H.N. 1983. The Mammals of the Southern African Subregion. Pretoria, South Africa: University of Pretoria. Smuts, G. L. 1978. Effects of population reduction on the travels and reproduction of lions in Kruger National Park. Carnivore 1: 61–72. Smuts, G. L. 1982. Lion. Johannesburg: Macmillan South Africa. Smuts, G. L., Whyte, I. J., & Dearlove, T. W. 1977. A mass capture technique for lions. East African Wildlife Journal 15: 81–87. Sokal, R. R., & Rohlf, F. J. 1995. Biometry, 3rd edition. New York: W. H. Freeman. Solomon, N. G., & French, J. A. 1997. Cooperative Breeding in Mammals. Cambridge, UK: Cambridge University Press. Stacey, P. B., & Koenig, W. D. 1990. Cooperative Breeding in Birds. Cambridge, UK: Cambridge University Press. Stander, P. E. 1991. Aspects of the ecology and scientific management of large carnivores in sub-Saharan Africa. Ph.D. dissertation, Cambridge University, Cambridge, UK. . 1992. Foraging dynamics of lions in a semi-arid environment. Canadian Journal of Zoology 70: 8–21. Stander, P. E., & Albon, S. D. 1993. Hunting success of lions in a semi-arid environment. Symposia of the Zoological Society of London 65: 127–143.
334 ▪ R E F E R E N C E S
Stander, P. E., Haden, P. J., Kaqece, //, & Ghau, // 1997. The ecology of asociality in Namibian leopards. Journal of Zoology, London 242: 343–364. Starfield, A. M., Shiell, J. D., & Smuts, G. L. 1981. Simulation of lion control strategies in a large game reserve. Ecological Modelling 13: 17–28 Steel, R.G.D., & Torrie, J. 1980. Principles and Procedures of Statistics: A Biometrical Approach. New York: McGraw-Hill. Steklis, H. D., Brammer, G. L., Raleigh, M. J., & McGuire, M. T. 1985. Serum testosterone, male dominance and aggression in groups of male vervet monkeys (Cercopithecus aethiops sabaeus). Hormones and Behavior 19: 154–163. Stephens, D. W. 1981. The logic of risk-sensitive foraging preferences. Animal Behaviour 29: 628–629. Stephens, D. W., & Charnov, E. L. 1982. Optimal foraging: Some simple stochastic models. Behavioral Ecology and Sociobiology 10: 251–263. Stephens, D. W., & Krebs, J. R. 1986. Foraging Theory. Princeton, NJ: Princeton University Press. Stevenson-Hamilton, J. 1939. The health of wild animals. Journal of the South African Veterinary Medical Association 10: 56–64. . 1947. The African wild dog. In: Wildlife in South Africa. pp. 224–240. London: Cassel & Company. Swanepoel, R., Barnard, B.J.H., & Meredith, C. D. 1993. Rabies in Southern Africa. Onderstepoort Journal of Veterinary Research 60: 325–346. Tanzania Department of Wildife. 1995. Selous Game Reserve General Management Plan. Dar es Salaam: United Republic of Tanzania Department of Wildlife. Taylor, C. R., Schmidt-Neilsen, K., Dmi’el, R., & Fedak, M. 1971. Effect of hyperthermia on heat balance during running in the African wild dog. American Journal of Physiology 220: 823–827. Taylor, C. R., Heglund, N. C., & Maloiy, G.M.O. 1982. Energetics and mechanics of terrestrial locomotion: I. Metabolic energy consumption as a function of speed and body size in birds and mammals. Journal of Experimental Biology 97: 1–21. Temminck, M.C.J. 1820. Sur le genre Hy`ene, et description d’une esp`ece nouvelle d´ecouverte en Afrique. Annales G´en´erale Scientifique et Physique de Bruxelles 3: 46–57. ¨ Thenius, E. 1954. On the origins of the dhole (Genus Cuon Hudson). Osterreich Zoologie 5: 377–388. Thenius, E. 1972. Verbreitungsgeschichte der Saugetiere. Stuttgart, Germany: G. Fisher. Thomas, N. J., Foreyt, W. J., Evermann, J. F., Windberg, L. A., & Knowlton, F. F. 1984. Seroprevalence of canine parvovirus in wild coyotes from Texas, Utah, and Idaho (1972 to 1983). Journal of the American Veterinary Medical Association 185: 1283–1287. Thompson, I. D. 1994. Marten populations in uncut and logged boreal forests in Ontario. Journal of Wildlife Management 58: 272–280. Thornhill, R. 1981. Panorpa (Mecoptera: Panorpidae) scorpionflies: Systems for understanding resource-defense polygyny and alternative male reproductive efforts. Annual Review of Ecology and Systematics 12: 355–386. Thurber, J. A., Peterson, R. O., Woolington, J. D., & Vucetich, J. A. 1992. Coyote coexistence with wolves on the Kenai Peninsula, Alaska. Canadian Journal of Zoology 70: 2494–2498.
R E F E R E N C E S ▪ 335
Tilman, D. 1986. Resources, competition and the dynamics of plant communities. In: Plant Ecology. M. J. Crawley, ed., pp. 51–74. Oxford, UK: Blackwell Scientific Publications. Treves, A., & Chapman, C. 1996. Conspecific threat, predation avoidance and resource defense: Implications for grouping in langurs. Behavioral Ecology and Sociobiology 39: 43–53. Turnbull, P.C.B., Bell, R.H.V., Saigawa, K., Munyenyembe, F.E.C., Mulenga, C. K., & Makala, L.H.C. 1991. Anthrax in wildlife in the Luangwa valley, Zambia. The Veterinary Record 128: 399–403. Turner, C. D., & Bagnara, J. T. 1976. General endocrinology, 6th ed. Philadelphia: W. B. Saunders Co. Turner, G. F., & Pitcher, T. J. 1986. Attack abatement: A model for group protection by combined avoidance and dilution. The American Naturalist 128: 228–240. Turner, M. 1987. My Serengeti Years. London: Elm Tree Books. TWCM. 1990. Selous Census, 1989. Dar es Salaam: Tanzania Wildlife Conservation Monitoring. . 1991. Wildlife Census, Selous, June 1991. Arusha: Tanzania Wildlife Conservation Monitoring. . 1994. Aerial survey of the Selous Game Reserve, Mikumi National Park and surrounding areas, dry season 1994. Arusha: Tanzania Wildlife Conservation Monitoring. Tyrrell, J. B., & Baxter, J. D. 1981. Glucocorticoid therapy. In: Endocrinology and Metabolism. P. Felig, J. D. Baxter, A. E. Broadus, & L. A. Frohman, eds., pp. 599– 625. New York: McGraw Hill. Uetz, G. W., & Hieber, C. S. 1994. Group size and predation risk in colonial webbuilding spiders: Analysis of attack abatement mechanisms. Behavioral Ecology 5: 326–333. USDA. 1990. Composition of food. Beef products: Raw, Processed, Prepared. United States Department of Agriculture, Agriculture Handbook Series, Number 8–13. van Heerden, J. 1979a. The transmission of canine ehrlichiosis in the wild dog Lycaon pictus (Temminck, 1820) and black-backed jackal Canis mesomelas (Schreber). Journal of the South African Veterinary Medical Association 50: 245–248. . 1979b. The transmission of babesia canis to the wild dog Lycaon pictus (Temminck, 1820) and black-backed jackal Canis mesomelas (Schreber). Journal of the South African Veterinary Medical Association 51: 119–120. . 1986. Disease and mortality of captive wild dogs, Lycaon pictus. South African Journal of Wildlife Research 16: 7–11. van Heerden, J., & Kuhn. F. 1985. Reproduction in captive hunting dogs Lycaon pictus. South African Journal of Wildlife Research 15: 80–85. van Heerden, J., Bainbridge, N., Burroughs, R.E.J., & Kriek, N.P.J. 1989. Distemperlike disease and encephalitozoonosis in wild dogs (Lycaon pictus). Journal of Wildlife Diseases 25: 70–75. van Heerden, J., Mills, M.G.L., van Vuuren, M. J., Kelly, P. J., & Dreyer, M. J. 1995. An investigation into the health status and diseases of wild dogs (Lycaon pictus) in the Kruger National Park. Journal of the South African Veterinary Association 66: 18–27. Van Lawick H., 1973. Solo. London: Collins. van Orsdol, K. G. 1984. Foraging behaviour and hunting success of lions in Queen Elizabeth National Park, Uganda. African Journal of Ecology 22: 79–99.
336 ▪ R E F E R E N C E S
van Orsdol, K. G., Hanby, J. P., & Bygott, J. D. 1985. Ecological correlates of lion social organization (Panthera leo). Journal of Zoology, London 206: 97–112. Van Valkenburgh, B. 1989. Carnivor dental adaptations and diet: A study of trophic diversity within guilds. In: Carnivore Behavior, Ecology and Evolution. J. Gittleman, ed., pp. 410–436. Ithaca, NY: Cornell University Press. . 1996. Feeding behavior in free-ranging large African carnivores. Journal of Mammalogy 77: 240–254. . 2001. The dog-eat-dog world of carnivores: A review of past and present carnivore community dynamics. In: Meat Eating and Human Evolution. C. Stanford & H. Bunn, eds., pp. 101–121. Oxford, UK: Oxford University Press. Vehrencamp, S. L. 1983. A model for the evolution of “despotic” versus “egalitarian” species. Animal Behaviour 31: 667–682. Venkataraman, A. 1995. Do dholes (Cuon alpinus) live in packs in response to competition or predation by large cats? Current Science (Bangalore) 69: 934–936. Venter, J., & Hopkins, M. E. 1988. Use of a simulation model in the management of a lion population. South African Journal of Wildlife Research 18: 126–130. Verhage, H. G., Beamer, N. B., & Brenner, R. M. 1976. Plasma levels of estradiol and progesterone in the cat during polyestrus, pregnancy and pseudopregnancy. Biology of Reproduction 14: 579–585. Viljoen, P. C. 1993. The effects of changes in prey availability on lion predation in a large, natural ecosystem in northern Botswana. Symposia of the Zoological Society of London 65: 193–213. Vleck, C. M., Mays, N. A., Dawson, J. W., & Goldsmith, A. R. 1991. Hormonal correlates of parental and helping behavior in cooperatively breeding Harris’ hawks (Parabuteo unicinctus). Auk 108: 638–648. Vucetich, J. A., & Creel, S. 1999. Ecological interactions, social organization and extinction risk in African wild dogs. Conservation Biology 13: 1172–1182. Vucetich, J. A., Peterson, R. O., & Waite, T. 1997. Effects of social structure and prey dynamics on extinction risk in gray wolves. Conservation Biology 11: 957–965. Wandeler, A. I. 1993. Wildlife rabies in perspective. Onderstepoort Journal of Veterinary Research 60: 347–350. Waser, P. M. 1996. Patterns and consequences of dispersal in gregarious carnivores. In: Carnivore Behavior, Ecology and Evolution, Vol. 2. J. L. Gittleman, ed., pp. 267–295. Ithaca, NY: Cornell University Press. Waser, P. M., & Jones, W. T. 1983. Natal philopatry in solitary mammals. Quarterly Review of Biology 50: 355–390. Waser, P. M., Creel, S. R., & Lucas, J. R. 1994. Death and disappearance: Estimating mortality risks associated with philopatry and dispersal. Behavioral Ecology 5: 135–141. Wasser, S. K. 1995. Costs of conception in baboons. Nature 376: 219–220 Wasser, S. K., Monfort, S. L., Southers, J., & Wildt, D. E. 1994. Excretion rates and metabolites of oestradiol and progesterone in baboon (Papio cynocephalus cynocephalus) faeces. Journal of Reproduction and Fertility 101: 213–220. Wayne, R. Geffen, E., Girman, D. J., Koepfli, K., Lau, L. M., & Marshall, C. R. 1997. Molecular Systematics of the Canidae. Systematic Biology 46: 622–653. Welsh, T. H., & Johnson, B. H. 1981. Stress-induced alterations in secretion of corticosteroids, progesterone, luteinizing hormone, and testosterone in bulls. Endocrinology 109: 185–190.
R E F E R E N C E S ▪ 337
Whateley, A. 1981. Density and home range of spotted hyaena in Umfolozi Game Reserve, Natal. Lammergeyer 31: 15–20. Whateley, A., & Brooks, P. M. 1978. Numbers and movements of spotted hyaenas in Hluhluwe Game Reserve. Lammergeyer 26: 44–52. White, P. J., & Garrott, R. A. 1997. Factors regulating kit fox populations. Canadian Journal of Zoology 75: 1982–1988. White, P. J., Ralls, K., & Garrott, R. A. 1994. Coyote–kit fox interactions as revealed by telemetry. Canadian Journal of Zoology 72: 1831–1836. Wilkinson, I. S. 1996. The 1994–95 wild dog photographic survey. Licaone 5: 6–8. Williams, G. C. 1966. Adaptation and Natural Selection: A Critique of Some Current Evolutionary Thought. Princeton, NJ: Princeton University Press. Williams, E. S., Thorne, E. T., Appel, M.J.G., & Belitsky, D. W. 1988. Canine distemper in black-footed ferrets (Mustela nigripes) from Wyoming. Journal of Wildlife Diseases 24: 385–398. Wingfield, J. C. 1988. Changes in reproductive function of free-living birds in direct response to environmental perturbations. In: Processing of Environmental Information in Vertebrates. M. H. Stetson, ed., pp. 121–148. Berlin: Springer-Verlag. . 1994. Modulation of the adrenocortical response to stress in birds. In: Perspectives in Comparative Endocrinology, pp. 520–528. Ottawa: National Research Council of Canada. Wingfield, J. C., & Silverin, B. 1986. Effects of corticosterone on territorial behavior of free-living male song sparrows Melospiza melodia. Hormones and Behavior 20: 405–417. Wingfield, J. C., & Moore, M. C. 1987. Hormonal, social, and environmental factors in the reproductive biology of free-living male birds. In: Psychobiology of Reproductive Behavior: An Evolutionary Perspective. D. Crews, ed. Englewood Cliffs, NJ: Prentice Hall. Wingfield, J. C., Hegner, R. E., Dufty, A. M., & Ball. G. F. 1990. The “challenge hypothesis”: Theoretical implications for patterns of testosterone secretion, mating systems, and breeding strategies. The American Naturalist 136: 829–846. Wingfield, J. C., Hegner, R. E., & Lewis, D. M. 1991. Circulating levels of luteinizing hormone and steroid hormones in relation to social status in the cooperatively breeding white-browed sparrow weaver, Plocepasser mahali. Journal of Zoology, London 225: 43–58. Wisdom, M. J., & Mills, L. S. 1997. Sensitivity analysis to guide population recovery: Prairie chickens as an example. Journal of Wildlife Management 61: 302–312. Woodroffe, R., & Ginsberg, J. R. 1998. Edge effects and the extinction of populations inside protected areas. Science 280: 2126–2128. Woodroffe, R. B., Ginsberg, J. R., & Macdonald, D. W. 1997. The African Wild Dog: Status Survey and Conservation Action Plan. Gland, Switzerland: World Conservation Union. Wozencraft, W. C. 1989. Classification of the recent carnivora. In: Carnivore Behavior, Ecology and Evolution. J. L. Gittleman, ed., pp. 569–593. Ithaca, NY: Cornell University Press. Wright, S. E. 1931. Evolution in Mendelian populations. Genetics 16: 97–159. . 1938. Size of population and breeding structure in relation to speciation. Science 87: 430–431. . 1943. Isolation by distance. Genetics 28:114–138.
338 ▪ R E F E R E N C E S
Wright, S. E. 1969. Evolution and the Genetics of Populations. II. The Theory of Gene Frequencies. Chicago: University of Chicago Press. Zeigler, T. E., Scheffler, G., & Snowdon, C. T. 1995. The relationship of cortisol levels to social environment and reproductive functioning in female cotton-top tamarins, Sanguinus oedipus. Hormones & Behavior 29: 407–424. Zielinski, W. J., & Vandenbergh, J. G. 1993. Testosterone and competitive ability in male house mice, Mus musculus: Laboratory and field studies. Animal Behaviour 45: 873–891.
Index
agonistic behavior. See aggression; dominance; injury aggression, territorial, 46–50; and dominance, 205–7. See injury behavior, adaptations to competition, 266– 68; avoidance of competitors, 247–53; interactions with competitors, 259–63; cooperation, 76–83; during dispersal, 187–90; denning, 50–51; during interpack encounters, 45–50; and epidemiology, 284–86; hunting, 71–73, 76–78; at kills, 259–63; reproductive, 159–61, 203–5; with scavengers, 262–63; territorial aggression, 46–50 body size, 1, 29 coloration, 1–3 competition, adaptations to, 266–68; and population density, 253–56; and diet, 257– 59; and spatial distribution, 247–53; and food intake, 262–63; impact of, 265–66; away from kills, 263–65; interference at kills, 259–63; interspecific, 245–68, 298– 99, 305–6 conflict, with humans, 8–9, 288–89 conservation, 7–11, and interspecific competition, 245–68, 298–99, 305–6; and conflict with humans, 8–9; and disease, 269– 87, 300–301, 306–7; and geographic distribution, 5; and spatial distribution, 248– 53, 266–67; effective population size, 175–78; causes of extinction risk, 7–11; quantitative extinction risks, 288–309. See also extinction risk; population cooperation, in hunting, 76–83; and reproductive success, 163–67, 172–73 demography. See life table; reproduction; survival dens, 50–51 density dependence, 150–54, 173–75 dimorphism, sexual, 1, 29 disease, anthrax, 277–79; canine distemper virus, 271–74; canine parvovirus, 279–82; in Chobe National Park, 271; and domestic dogs, 274, 285–86; in Kruger National
Park, 270; macroparasites, 283–84; and population dynamics, 286–87; 300–301, 306–7; protozoans, 283; rabies virus, 274– 77; in Serengeti National Park, 269–70 dispersal, and age, 184; comparison among populations, 190–91; definition of, 181– 83; distance moved, 186–87; floating, 187–90; group size, 184–86; and inbreeding avoidance, 179–80, 195–200, 226–40; and inclusive fitness, 238–40; and mate competition, 194–95, 180–200; mortality risk of, 191–94; and relatedness, 196–99; 226–40; and reproductive suppression, 194–95; sex-bias, 179–81, 184, 191–94. See also reproductive suppression distribution, and interspecific competition, 248–53, 266–67; geographic, 5; of hyenas in Selous, 249–53; of lions in Selous, 250–53; of wild dogs in Selous, 248–49 domestic dogs, 274, 285–86 dominance, 5, 201–5; and aggression, 205– 7; and fecundity, 159, 161–63; and hormones of females, 211–14; and hormones of males, 210–11; and mating rate, 207– 10; and rank of mating partner, 210; and reproduction, 159, 161–63; and stress, 215–16, 218–22; and survival, 148, 151; and parentage, 216–17 effective population size, demographic, 176– 78; genetic, 175–76 endocrinology. See hormones; reproduction; stress extinction risk, 288–309; causes of, 7–11; and disease, 300, 306–8; effective population size, 175–78; and immigration, 300– 302; individual-based models of, 295–304; Leslie matrix projections, 290–95; and lion density, 298–99; and litter size, 302– 3; and pack size, 292–94, 303–4; population growth rate, 154–57 fecundity. See reproduction; reproductive skew; reproductive suppression foraging success, 88–99. See hunting fossils, 3–4
340 ▪ I N D E X gestation, 159, 214–15 growth rate, of population, 154–57, 292–94. See reproduction; survival habitat, 38–40, and kill sites, 84; and lions, 56–59; map of, 40; and prey distribution, 55–59; and prey encounters, 247; selection and preference, 52–59 helping. See cooperation; inclusive fitness herd size, 124–44 home ranges, cores, 39–41; and den locations, 50–51, 64; and habitat, 53–54; mapping of, 36–37; overlap, 41–50; and pack size, 51–52, 60–61; shape, 41–43; size, 39–41, 60–65 hormones, nonbreeder lactation, 214–15; pseudopregnancy, 214–15; sex steroids of females, 211–41; sex steroids of males, 210–11; stress, 215–16, 218–22 hunting, behavior, 76–83; chase distance, 84, 86; cooperation in, 76–83; costs and benefits of, 84–102; during denning, 64–65; energetics of, 89–93, 116; foraging success, 88–89; and group size, 67–69, 84– 102; kill sites, 84; methods of observation, 69–76; multiple kills, 86; and pack size, 84–102; and prey selection, 74–76; rate of, 87; success of, 73–74, 84–85 hyenas, census of, 33–35; competition at kills, 259–63; diet of, 257–59; distribution in Selous, 249–53; hunting, 67–68; interactions not at kills, 263–65; and interspecific competition, 245–48; and wild dog density, 253–56, 288–89 impala, in carnivore diets, 257–59; herd size and predation, 128–30, 132–33, 139–44; predation, 74–76, 91, 104, 108, 110 inbreeding. See dispersal, relatedness inclusive fitness, of nondispersers, 231–38; of dispersers, 238–40; and reproductive skew, 240–44 injury, during hunting, 119; during fights among packs, 49; during fights within packs, 205–7 kills, locations, 84; multiple, 86 kin selection. See inclusive fitness lactation, by nonbreeders, 214–15 life table, 155–57. See reproduction; survival
lions, census, 33–35; diet, 257–60; distribution in Selous, 250–53; and extinction risk of wild dogs, 298–99; hunting by, 67, 99–102; and interspecific competition, 245–48, 298–99; predation on wild dogs, 263–65; and wild dog density, 253–56, 288–89, 305–6 methods, 25–35, of anesthesia, 28; for other carnivores, 246–48; days of observation, 26–27; for dispersal data, 181–83, 191– 92; for endocrine assays, 31–37; general, 25; and handling effects, 27–30; for hunting observations, 69–73; for models of extinction risk, 290–91, 296; of radiocollaring, 27–30; of estimating survival, 145 models, of extinction risk, 290–308; of prey selection, 114–22; of reproductive skew, 240–44 nondispersal. See dispersal; inclusive fitness; reproductive skew pack size, 4–5, and extinction risk, 303–4; and home range size, 51–52; and hunting preferences, 109–11; and hunting success, 74, 84–102, 111–12; and population growth rate, 292–94; and prey encounters, 107–8, 110; and prey selection, 109–10 philopatry. See inclusive fitness phylogeny, 3–4 population, density, 10, 23, 150–54, 173–75, 248, 288; size, 23–24, 275–78, 288–89; growth rate, 154–57, 292–94; of wild dogs in other ecosystems, 10, 246–48, 254–56; of other carnivores, 253–56 predation, and herd size, 124–44; and hunting preferences, 109–11, 124–26, 130; and hunting success, 111–12, 124–26, 130–33; and prey encounter rates, 105–8, 110, 115, 124–30; and prey mass, 75, 85; and prey selection, 112–14, 103–23; 114–22; quantitative models of prey selection, 114–22; and reproduction, 167–69; by species, 74– 76, 93–94 rainfall, 17–18 relatedness, 6, 191–16; of subordinates to breeders, 226–31; and inclusive fitness, 231–40; and reproductive skew, 240–94
I N D E X ▪ 341 reproduction, age-specific, 155–57; behavior, 159–61; best regression model of, 169; comparison to other populations, 169–72; and dominance, 161–66; and pack size, 163–67, 172–73; and prey availability, 167–69; and rank, 161–63; seasonality of, 159–61; skew, 240–44 reproductive skew, 240–44. See dispersal; inclusive fitness; reproductive suppression reproductive suppression, and aggression, 205–7; and competition for breeding opportunities, 223–26; and dispersal, 231– 40; effectiveness of, 216–17; and sex hormones of females, 211–14; and sex hormones of males, 210–11; and mating rates, 207–10; mechanisms of, 201–22; ranks of mating pairs, 210; and relatedness, 226–31; and reproductive skew, 240–44; and stress, 215–16, 218–22 Selous Game Reserve, animal populations, 21–22; climate, 17–19; habitats, 17–21; history, human activities, 15–16, 22; size, 15–17 social dominance. See dominance
social organization, 4–7; dispersal patterns, 179–200, 222; dominance and mating, 201–4; and encounters between packs, 45– 50; and pack size, 84–102; and reproductive suppression, 201–22. See cooperation; dispersal; dominance; pack size; reproduction; reproductive suppression; survival stress, and rank, 215–16, 218–22 survival, age-specific rates of, 146–51, 161– 63; best regression model of, 148; comparison to other populations, 151–54; methods of estimation, 145; and population density, 150–54, 173–75; and rank, 151; and sex ratio, 157–59 taxonomy. See phylogeny temperature, 17–19 utilization distribution. See home range warthog, 81, 94; in carnivore diets, 257–59 wildebeest, in carnivore diets, 257–59; herd size and predation, 126–31, 133–39; predation, 74–76, 78–81, 91, 103–4, 108, 110, 112