CAA 96: Computer Applications and Quantitative Methods in Archaeology 1996 9781841710495, 9781407351803

Citation preview

BAR  S845  2000  

CAA 96

LOCKYEAR, SLY & MIHĂILESCU-BÎRLIBA (Eds)  

Computer Applications and Quantitative Methods in Archaeology Edited by

C

CAA 96

Kris Lockyear Timothy J. T. Sly Virgil Mihăilescu-Bîrliba

BAR International Series 845 9 781841 710495

B A R

2000

CAA96 Computer Applications and Quantitative Methods in Archaeology Edited by

Kris Lockyear Timothy J. T. Sly Virgil Mihailescu-Birliba

BAR International Series 845 2000

Published in 2016 by BAR Publishing, Oxford BAR International Series 845 CAA 96 © The editors and contributors severally and the Publisher 2000 The authors' moral rights under the 1988 UK Copyright, Designs and Patents Act are hereby expressly asserted. All rights reserved. No part of this work may be copied, reproduced, stored, sold, distributed, scanned, saved in any form of digital format or transmitted in any form digitally, without the written permission of the Publisher.

ISBN 9781841710495 paperback ISBN 9781407351803 e-format DOI https://doi.org/10.30861/9781841710495 A catalogue record for this book is available from the British Library BAR Publishing is the trading name of British Archaeological Reports (Oxford) Ltd. British Archaeological Reports was first incorporated in 1974 to publish the BAR Series, International and British. In 1992 Hadrian Books Ltd became part of the BAR group. This volume was originally published by Archaeopress in conjunction with British Archaeological Reports (Oxford) Ltd / Hadrian Books Ltd, the Series principal publisher, in 2000. This present volume is published by BAR Publishing, 2016.

BAR

PUBLISHING BAR titles are available from:

E MAIL P HONE F AX

BAR Publishing 122 Banbury Rd, Oxford, OX2 7BP, UK [email protected] +44 (0)1865 310431 +44 (0)1865 316916 www.barpublishing.com

Contents

1

2

3

4

5

6

7

A Bayesian approach to a problem Clive Orton 1.1 Background 1.2 Discussion 1.3 Aim .... 1.4 Method .. 1.5 A worked example 1.6 Conclusion Experiments with Detrended Kris Lockyear 2.1 Introduction ....... . 2.2 Examining a test data set 2.3 Conclusions ....... .

of archaeological

1

1 1 2 2

4 6

Correspondence

Analysis

9

9 10

14

Vessel volume as a factor in ceramic quantification: John Hawthorne 3.1 Introduction. . . . . . . . . . . . . . . 3.2 Ceramic quantification ........ . 3.3 African Red Slip Ware and economics 3.4 Data manipulation 3.5 Vessel capacity 3.6 Conclusion

The COMPASS method for the estimation Eugen S. Teodor 4.1 Introduction . . . . . . 4.2 The COMPASS system .

site evaluation

the case of African

of the capacity

Red Slip Ware

19

19 19 19 20 20 23 of pottery

vessels

25

25 25

Dating Stonehenge C. Bronk Ramsey and A. Bayliss 5.1 Introduction . 5.2 Mesolithic activity 5.3 Phase 1 5.4 Phase 2 5.5 The stone monument . 5.6 The Avenue 5.7 Later use 5.8 Conclusions

29

Stonehenge - Mapping the Stones Paul G. Bryan and Michael Clowes 6.1 The background 6.2 The project ....... . 6.3 The survey ....... . 6.4 Conclusion of the project 6.5 Future uses of the data ..

41

Multidimensional analysis of the archaeological site at Mitoc-Malu Galben Vasile Chirica and Andrei Cojocaru 7.1 The site (Vasile Chirica) .... 7.2 The program (Andrei Cojocaru)

29 30 30 32 35 35 37 37

41 41 43

47 47 discoveries

from the multiphase

Palaeolithic 49 49 51

8

The statistical

Jon 8.1 8.2 8.3 8.4 8.5 8.6 9

analysis

of ground

probing

radar data from "radar-weak"

55

sites

Bradley and Mike Fletcher Introduction ......... . Statistical feature extraction . The site . . . . . . . . Radar results . . . . . . . . . Verification of results ..... Conclusions and further work

An Application

of Neural

Networks

55 55 56 57 57 60 to Use-Wear

Analysis.

Some Preliminary

Results

J. A. Barcelo, A. Vila and J. Gibaja 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 Unsupervised v. supervised learning: the role of archaeological experimentation 9.3 Neural networks and experimental archaeology . . . . . . . . . . . . . . . . . . 9.4 Leaming the relationship between qualitative description of tools and prehistoric activities: the training set . . . . . . . . . . . . . . . . . . 9.5 Redundancy and conflicting data . . . . . . 9.6 Statistical analysis v. Artificial Intelligence 9. 7 Conclusions . . . . . . . . . . . . . . . . . .

63

63 63 65 66 66 67 68

10 Arch WEB: a web site for Dutch archaeologists M. Wansleeben and M. H. van den Dries 10.1 Introduction ........ . 10.2 Aims of the project .... . 10.3 Evaluation of the first year 10.4 The future of Arch WEB 10.5 Conclusion

71

11 GIS and visualising

81

71 71 72 76 78

the palaeoenvironment N. R. Burton and C. A. Shell 11.1 Introduction .............. . 11.2 Visualisation and visual prediction . . 11.3 The site, the setting and the situation 11.4 Data preparation . . . . . . . . . 11.5 The modelling process . . . . . . 11.6 Model display and enhancement . 11. 7 Prediction and prospection . . . 11.8 The next stage: comprehensive prospection data sets 11.9 Discussion ....................... .

12 GIS and Early Aland:

Spatial

analysis

in an archipelago

81 81 82 82 83 84 85 87 87 of south-western

Finland

Patrick Daly, Michael Frachetti and Jari Okkonen 12.1 Introduction . . . 12.2 Profile of A.land . . . 12.3 Project Overview . . 12.4 Concluding Remarks 13 An exploratory

GIS approach

to Andalusian

91

91 91 92 97 Archaeological

Heritage

F. Amores, L. Garcia, V. Hurtado, H. Marquez and C. Rodriguez-Bobada 13.1 Introduction . . . . . . . . . . . . . . . . . . . 13.2 The Sierra de Huelva - A Rural Case Study 13.3 The Historic Centre of Sevilla 13.4 Assessing but not Concluding 14 Looking at intra-site GIS Jeremy Huggett 14.1 Introduction . . . . . . . . . . . 14.2 A case study . . . . . . . . . . 14.3 The problem with intra-site GIS 14.4 Where does intra-site GIS go from here?

Records

101

101 102 108 111 117 117 117 118 119

ii

14.5 Conclusions

.........................

.

. 121

15 Spatial technology and archaeological theory revisited David Wheatley 15.1 Introduction ......... . 15.2 The need for theory .... . 15.3 An emerging theory of place . 15.4 Spatial technology and an archaeology of place 15.5 Conclusions ................ . 16 A GIS investigation of site location and landscape Tuscany Philip Perkins 16.1 Introduction. . . . . . . . 16.2 Landscape and settlement 16.3 Methodology . . . . . . . 16.4 Discussion and development 16.5 Interpretation of the models of settlement locations . 16.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . 17 Building theory into Alicia L. Wise 17 .1 Current theories 17.2 Ecology . . . . . 17.3 GIS and historical 17.4 Phenomenology 17.5 Discussion . . . .

GIS-based landscape

123

123 123 124 125 129 relationships

in the Albegna

Valley, 133

133 133 134 135 136 138

analysis

141

. . . . . . . . . . ecology

141 141 142 143 144

. . . . .

18 Computer Networks in Higher Education. Paul D. Bossons and D. E. Ord 18.1 Introduction. . . . . . . . . . . . 18.2 Information Technology services . 18.3 Network infrastructure . . . . 18.4 Strengths and weaknesses . . 18.5 How did the network evolve? 18.6 The future . . . . . . . . . . .

A Case Study:

Staffordshire

University

149

149 149 151 153 154 155

19 Retrospect on 1970: Looking back on the developments of computing archaeology nia since the Mamaia Conference John Wilcock and Silviu Sanie 19.1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.2 The 1970 conference Mathematics in the Archaeological and Historical Sciences 19.3 The 1970s and 1980s . . . . . . . . 19.4 Attendance at RECOMDOC 1992 . . . . . 19.5 The present day in Romania. . . . . . . . . 19.6 Attendance at the 1970 Mamaia Conference 19.7 Attendance at RECOMDOC 1992 . . . . . 20 Abstracts

in Roma157

157 157 162 163 164 164 165 169

lll

List of Figures

1.1

Prior probability that 0 > 0.5, as a function of b (after Nicholson & Barry 1995, Fig. 1).

2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11

Species map from CA of 241 Roman Republican coin hoards Sample map from CA of 241 Roman Republican coin hoards Detrending by segments ........... . Species map from ordinary CA of test data set ....... . Sample map from ordinary CA of test data set ....... . Species map from CA detrended by third order polynomials . Sample map from CA detrended by third order polynomials. Cumulative percentage curves for BPT, SEI, SPN, CRl, BOR, & CR2 Cumulative percentage curves for eight hoards from cluster group b Species map from detrended CA of 241 hoards Sample map from detrended CA of 241 hoards

10 10 10 12 12 14 14 15 15 16 16

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8

Sherd count .................. . Building inscriptions in Roman Africa, by year . Vessels per year ... Mean vessel volume . Second century bowls Third century dishes Mean breakage rate Overall volume

21 21 21 21 22 22 22 22

4.1

The COMPASSsystem

26

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11

Location of mesolithic features under the car park . Radiocarbon dates from the mesolithic features Model of the dating of phase 1 ........... . Phase plans of the development of Stonehenge .. . Location of articulated vertebrae in the secondary silting of the ditch The dating of phase 1 ..... The dating of phases 1 and 2 . The dating of phase 3 ... . The dating of the Avenue .. . Sections through Y Hole 30 Plan of Stonehenge and the Avenue, within the modern landscape

30 32 32 33 34 34 34 36 36 36 37

6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12

Aerial view of the stone circle at Stonehenge 1:3000 scale stereo pair of the stone circle .. 1:50 scale Topographic survey of stone circle 1:50 scale photogrammetric drawing of Muchelney Abbey, Somerset Stereopair of stone 30 from 1989 ....... . Wild P31 metric camera in use at Stonehenge Total Station Theodolite in use at Stonehenge Stereopair of stone 3 ............. . Leica SD2000 Analytical Photogrammetric Plotting Machine . Leica Helava DPW750 Digital Photogrammetric Workstation Zeiss Phodis ST30 Digital Photogrammetric Workstation .. 3D DEM of stone 3 formatted for viewing within AuTOCAD

42 43 43 44 45 45 45 46 46 48 48 48

7.1 7.2 7.3

Pendant from Mitoc-Malu Galben Aurignacian lithics .... Screenshot of the program ....

50 51 52 V

4

8.1 8.2 8.3 8.4 8.5 8.6 8.7

A typical radar image from a radar-strong site A typical radar image from a radar-weak site . The activity analysis process . . . . . . . . . . Topographical and crop-mark evidence at Hindwell Enclosure 1 Radar activity maps for a series of signal return time windows The resistivity results for Hindwell 1 Resistivity residuals at Hindwell 1.

56 56 58 58 59 60 60

10.1 10.2 10.3 10.4

Home page of the Arch WEB information service . The number of documents read during a visit The time visitors spent reading a document . . . A WWW document with an attractive magazine-like lay-out

72 74 74 77

11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8

Borehole data format . . . . . . . . . . . . Cross-section profiles of generated surfaces Altitude shading . . . . . . . . . . . . Hillshaded surface . . . . . . . . . . . . . . Perspective view with a fishnet drape . . . Perspective view with altitude shading and contours . Perspective view with hill-shading and base-map overlay Fenland Survey map showing the position of Bronze Age barrows

83 84 85 85 86 86 87 88

12.1 12.2 12.3 12.4 12.5

92 95 96 97

12.6

The location of Aland . . . . . . A three dimensional 'timescape' Past shorelines . . . . . . . . . . Tower style image ....... . The 'nodata' default standard for topology construction. These boundaries are obviously skew from those of the Geodata set, yet a more 'land form' styled image is produced. Temporal landscapes

13.1 13.2 13.3 13.4 13.5 13.6 13.7 13.8 13.9 13.10 13.11 13.12 13.13 13.14 13.15 13.16 13.17

The Sierra de Huelva UTM zones Sites mentioned in the text Intensity of survey: Administrative units Intensity of survey: Adequate areas for spatial analysis Municipal land-planning ................ . Municipalities covered by redefinition of archaeological sites as polygons Proportion of sites described as polygons . . . . . . . . . Proportion of new sites recorded after systematic survey . Land use v. distribution of archaeological sites ...... . The impact of re-afforestation on the archaeological record Potential risk evaluation: communications network Potential risk evaluation: mining ...... . The historic centre of Sevilla . . . . . . . . . . . . . Risk assessment: basic dating of constructions ... Risk assessment: ground plots for archaeological research Available surface for archaeological excavation

103 105 105 108 108 109 109 110 110 111 111 112 112 113 114 114 114

14.1 14.2 14.3 14.4

A 'typical' CAD-derived distribution plot ... An incidence matrix of daub fragments . . . . An E-W profile through the southern-most structure A provisional interpretation of the location of structures on the motte .

120 120 120 120

15.1

The colonisation of South America

128

16.1 16.2 16.3 16.4 16.5

Location of 5th century 2nd century Distribution Distribution

. . . . . . .

study region and sampled transects settlements (Etruscan) .. settlements (Roman) ... of altitude above sea level of slopes in the valley ... vi

97 98

134 135 135 135 135

16.6 16.7 16.8 16.9 16.10 16.11

Distribution of Model of areas Model of areas Model of areas Model of areas Model of areas

aspects in the valley .................. with associated with 7th century settlement . . . . . with a positive association with Etruscan settlement with a negative association with Etruscan settlement with a positive association with Roman settlement . with a negative association with Roman settlement

18.1 18.2 18.3 18.4

The The The The

19.1

Was this the true situation at the Mamaia Conference?

Stafford Campus network Stoke-on-Trent Campus network infrastructure .. external link for the network

137 137 137 137 138 138 150 151 153 154

Vll

163

List of Tables

1.1

Minimum size of sample N needed to achieve specified values of p and of Bv-

2.1 2.2

Hoards in test data set .... Ten hypothetical coin hoards .

11

3.1 3.2 3.3

Partial correlations for early period (AD 100 to AD 350) Partial correlations for late period (AD 355 to AD 600) Regression coefficients for multiple and simple analyses

23 23 23

5.1 5.2 5.3

Summary of reliable radiocarbon dates from Stonehenge Details of two samples measured in October 1995 .... Estimated dates for the stone settings in Phase 3 at Stonehenge

31 33 36

10.1

Monthly registered number of visits

73

12.1

Time blocks utilised ....

94

13.1 13.2 13.3

First character of the zone Second character of the zone Conversion from the Struve to Hayford ellipsoid

17.1 17.2

Sample suite of variables suitable for using Sample suite of variables suitable for using

4

13

GIS GIS

lX

105 105 106 in an ecological framework. in a phenomenological framework.

142 142

Cuvint Inainte Kris Lockyear, 1 Timothy J. T. Sly 2 and Virgil Mihailescu-Birliba 1 Institute

of Archaeology, University College London; of Southampton; 3 Institute of Archaeology, Ia§i

Computer Applications in Archaeology 1996 was an unusual occasion for a conference that had been until that time confined largely to the United Kingdom with only two forays abroad, to Denmark and The Netherlands. It was not, however, the first time that an international conference on the topic had been held in Romania. In 1970 the Mamaia conference was a ground-breaking event (see Wilcock this volume; Hodson et al. 1971; Mihailescu-Birliba & Chirica 1996). The inherent difficulties and expense of attending the conference in Romania resulted in a smaller attendance than Aarhus or Leiden had enjoyed, but this did not detract from the success of the event which was attended by delegates from sixteen countries including the United States and Argentina. The conference was jointly organised by: The Institute of Archaeology of the Romanian Academy, la§i, The 'Al. I. Cuza' University, la§i, CIMEC - the Information Centre for Culture and Heritage of the Ministry of Culture, Bucure§ti and the Department of Archaeology, University of Southampton. That this event could take place at all was largely due to the conference sponsors, to whom we wish to extend our thanks; they included: The Society of Antiquaries of London, The World Archaeological Congress, Taylor and Francis Publishers, English Heritage, The Mitropolia of la§i, The Mayor of la§i, the Ministry of Culture, Romania, The Soros Foundation for an Open Society, the Catholic Bishopric of la§i and the Institute of Computing of the Romanian Academy, la§i. It is a measure of our success that more delegates managed to attend the conference on some form of bursary than paid the full registration fee. We would also like to thank our colleagues who came out to la§i a few days before the conference to run workshops on a variety of themes including databases, multimedia, survey techniques and, the most oversubscribed of them all, Clive Orton's workshop on analysing pottery - from sherds to Correspondence Analysis in a day. Lastly, I would also like to thank our anonymous referees for approaching a thankless task so willingly. This volume is somewhat unusual in a number of ways. Firstly, we were asked by Dan Teodor and Mircea Petrescu- Dimbovita if we would be willing to publish their welcoming addresses, which we were glad to do. Secondly, remarkably few of the papers given at the conference were submitted for publication, and in particular only two papers from Romania were offered. As a result, this volume does not reflect the

2 Department

3

of Archaeology, University

proceedings very well, and so we have taken the editorial liberty of including the conference abstracts at the back of the volume. Thirdly, we were asked if we would accept papers which were given at the CAAUK meeting in Glasgow: those by Wise and Huggett have been included. The address heading each paper are those at the time of the conference, the contact details at the end the most up-to-date ones available to the editors at publication. I (KL) would particularly like to thank Timothy Sly for his patience and help in the production of this volume, and would like to acknowledge the help of my colleagues in London, particularly James Conolly. In a previous publication Lockyear (1996) I noted that an editor's worst problem was the authors, and vice versa. I would here like to thank those authors, in particular Jeremy Huggett, Clive Orton and Alex Bayliss, for providing me with carefully written texts and good quality graphics (it would be invidious of me to point out those papers where I have redrawn or rescanned every figure, or not had a single complete reference!). For those interested in such things, these proceedings have been typeset using lb-TE)C 2s using a modified version of the standard report class, and a variety of packages. The bibliography was generated with BrnTE)C using a custom-made style file. The final version was compiled using MikTE)C on a Windows NT workstation, distilled into PDF files, and written to CD for delivery to the publisher.

References HODSON, F. R., D. G. KENDALL& P. TAUTU (eds.) 1971. Mathematics in the Archaeological and Historical Sciences. Edinburgh University Press, Edinburgh. KAMERMANS,H. & K. FENNEMA (eds.) 1996. Computer Applications and Quantitative Methods in Archaeology CAA95. Institute of Prehistory, University of Leiden, Leiden. Analecta Praehistorica Leidensia 28. LOCKYEAR, K. 1996. 'Computer-aided publication in practice.' In Kamermans & Fennema (1996), pp. 531-543. Analecta Praehistorica Leidensia 28. MrnArLEscu-BiRLIBA, V. & V. CHIRICA 1996. 'A survey of the development of computer applications in Romanian archaeology.' In Kamermans & Fennema (1996), pp. 525-30. Analecta Praehistorica Leidensia 28.

Xl

Computer Applications Archaeology 1996

and Quantitative

Methods

.

Ill

Prof. dr. Dan Gh. Teodor Directeur de l'Institut d'Archeologie de la§i

Mesdames et Messieurs, J'ai l'honneur et le plaisir de vous saluer et de vous souhaiter la bienvenue au nom de l'Institut d' Archeologie de la§i. La promtitude avec laquelle vous avez repondu aux organisateurs de ce Congres, prouve clairement que, en cette epoque encore inquiete, les archeologues et les informaticiens des nombreux pays de l'Europe, en depit des differences de vue professionnelles, ressentent toujours la necessite de se reunir et de collaborer, animes par leur devoir envers le passe et je dirais envers l'avenir. Le Congres qui commence aujourd'hui a la§i a pour nous une signification a part etant la premiere manifestation internationale de ce genre clans notre ville concernant l'utilisation des ordinateurs et des methodes quantitatives clans archeologie, domaine clans lequel nous ferons maintenant a peine les premiers pas. Dans ce contexte, certainement, vos communications et interventions apporteront des informations et des points de vue nouveaux qui constitueront un bon benefice scientifique pour tous, que nous avons l'intention de le materialise par la publication d'un tome. Le volume extremement grand d'informations qui doivent etre analysees, ordonees et interpretees pendant un temps relativement limite, impose partout clans la recherche l'emploi permanent de l'ordinateur.

C'est une necessite imperieuse et on peut dire que sans ordinateur la recherche devient de plus en plus malaisee a faire, parfois meme difficile a finaliser. Done, l'ordinateur represente un instrument d'une grande utilite, une aide serieuse et precise qui peut ouvrir de nouvelles voies d'investigation, plus completes et plus sures que celles existantes jusqu'a maintenant; il affronte le temps en devenant un ami de confiance qui nous permet des realisations d'une grande valeur. On a obtenu de notables resultats grace a l'emploi de l'ordinateur clans de nombreux domaines de recherche et je suis convaincu qu'ils ne se laisseront attendre clans les investigations archeologiques de notre pays. Nous esperons que, grace aux communications presentees et aux contacts qui seront etablis entre les specialistes vous pourriez connaitre les realisations des savants de divers pays, ce qui donnera une impulsion considerable en ce qui concerne la modernisation des recherches archeologiques de notre pays et implicitement pour obtenir un important essor des informations scientifiques. En vous remerciant de votre presence a la§i, je vous souhaite chaleureusement une cooperation fructueuse, pleine de succes a nos travaux et un tres agreable sejour clans notre ville.

Xlll

Computer Applications Archaeology 1996

and Quantitative

Prof. dr. Mircea Petrescu-Dfmbovita,

Methods

.

Ill

Membre de l'Academie Roumaine

Institutul de Arheologie, la§i

Mesdames, Mesdemoiselles, Messieurs, Permettez moi, au nom de la Filiale de la§i de l' Academie Roumaine et de la Section des Sciences Historiques et d' Archeologie de l' Academie Roumaine, d'adresser un salut cordial a la vingt-quatrieme reunion sur le theme 'L'application de l'ordinateur et les methodes quantitatives en archeologie' dont les travaux se deroulent ces jours a la§i. A la suite des efforts des organisateurs de l'Institut d'archeologie de la Filiale de l'Academie Roumaine de la§i et du Departement de l'archeologie de l'Universite de Southampton de la Grande Bretagne, a cette importante manifestation scientifique internationale participe bon nombre des specialistes de differents pays de l'Europe et de l' Amerique, auxquels on souhaite beaucoup succes clans les debats des themes et clans l'activite clans les ateliers. Par cette breve allocution se voudrais mettre en discussion, comme archeologue, certains aspects generaux concernant l'application des mathematiques clans l'archeologie et l'etat de recherches clans ce domaine en Roumanie. Ainsi, il est bien connu qu'en ce qui concerne l'utilisation des mathematiques clans l'archeologie et l'histoire il y a des points de vue differents. Tandis que les adeptes de l'ecole historique neokantienne de Baden refusent d'admettre la contribution des mathematiques clans le domaine des sciences historiques, parce que, a leur avis, l'analyse quantitative n'a pas d'importance, l'histoire etant une synthese des phenomenes individuels et non pas une repetition, au contraire les neo-pozitivistes admettent l'importance des methodes mathematiques clans ce domaine, en les considerant parmi les elements fondementaux pour la connaisance des phenomenes sociaux. A propos de ce probleme, on considere que l'archeologie, en depit des methodes de travail d'autres domaines, y compris les sciences de la nature, techniques et mathematiques, reste incluse clans le domaine des sciences humanistes, en contribuant, par sa fonction de connaitre, a la formation et au renforcement de la conscience historique. Bien sur qu'au point de vue de la methodologie de la recherche clans l'archeologie il s'impose l'elargissement des investigations interdisciplinaires, a une base plus large, clans le cadre de la collaboration internationale pour le developpement efficace de l'archeologie. Pour le renforcement des

contacts interdisciplinaires s'imposent des projets de recherches communes, bien planifiees et conduites, qui doivent presenter interet, par une thematique majeure, aussi pour l'archeologie que pour d'autres disciplines, avec lesquelles l'archeologie, a l'heure actuelle, peut et doit etre en contact en envisageant l'esprit du renouvellement clans cette discipline. En ce qui concerne notre pays, on espere que bient6t seront couramment utilises les moyens offerts par le traitement automatique des donnees archeologiques a l'aide de l'ordinateur, operation qui se deroule avec succes au Centre National des specialistes du Ministere de la Culture, qui font paraitre aussi un Bulletin et qui jusqu'a present ont effectue un grand nombre d'enregistrements d'objets, ainsi que de monuments, sites et fouilles archeologiques. De meme, en 1995 on a fait publie a Bucarest par Monsieur Costin Scorpan le premier tome de terminologie archeologique selective (tresor de termes, descripteurs, termes alternatifs, ascripteurs). A ceux-ci s'ajoute une activite laborieuse au Musee National d'Histoire Bucure§ti et au Musee d'Histoire de la Transylvanie de Cluj-Napoca, concernant directement les plus importantes fouilles archeologiques pratiquees par les collaborateurs de ces deux musees. Outre ces activites s'ajoutent d'autres a ClujNapoca, ou fonctionne pres du Musee d'Histoire de la Transylvanie, un centre actif clans le domaine de l'utilisation de l'ordinateur en archeologie. Ce musee, en collaboration avec la Faculte des mathematiques, avec l'Institut de physique nucleaire et avec le Lycee d'informatique, organise chaque anne des sessions aux debats publies ou predominent l'analyse quantitative des donnees archeologiques et les determinations chronologiques. Moi-meme j'ai transmis a ClujNapoca, en vue d'une analyse a l'aide de l'ordinateur les donnees concemant les contextes closes (habitations, annexes et trous) de l'habitat de Tru§e§ti, du nord de la Moldavie, le plus grand fouille jusqu'a present clans l'aire de la civilisation eneolithique de Cucuteni, a l'espoir d'obtenir la periodisation inteme de cet habitat. En ce qui conceme cette activite a la§i, on mentionne que depuis plus d'une dizaine d'annees, un groupe restreint de l'ancien Institut d'Histoire et d'Archeologie 'A.D.Xenopol', en collaboration avec les

xv

specialistes du Centre d'Informatique de Bucure§ti, a realise une codification indexee pour certaines publications d'archeologie de la§i et de Bucure§ti, dont la methode utilisee a ete publie en 1987 dans la revue Arheologia Moldovei. J'ai cite seulement quelques activites deroulees dans ce domaine a Bucure§ti, a Cluj-Napoca et a la§i,

quoique, bien sur, je suppose qu'il y aussi d'autres dans les localites ayant une tradition des recherches archeologiques. A la fin, j'espere que les archeologues roumains realiseront aussi les resumes de leurs ouvrages du domaine de l'archeologie en utilisant la terminologie archeologique selon les regles du Tresor de termes.

XVl

1 A Bayesian approach to a problem of archaeological site evaluation Clive Orton Institute of Archaeology, University College London

1. 1

Background

The operation of field archaeology in England today is largely directed by two documents, neither of which has legal force, but which together form a strong framework. They are Planning Policy Guidance Note 16: Archaeology and Planning (Department of the Environment 1990), commonly known as PPG 16, and Management of Archaeological Projects (Andrews 1991), commonly known as MAP2. PPG 16 gives policy guidance on the preservation of archaeological remains in the context of rural or urban development. The general philosophy is that much of the conflict of recent years about the preservation of archaeological remains could be avoided if steps were taken, before the granting of planning permission, to ascertain the nature and extent of any surviving remains on the site to be developed. It states that 'Where nationally important archaeological remains ... are affected ... there should be a presumption in favour of their physical preservation' (Department of the Environment 1990, para. 8), but that 'The case for the preservation of archaeological remains must however be assessed on the individual merits of each case, ... , including the intrinsic importance of the remains' (Department of the Environment 1990, para. 27). Further, 'Where it is not feasible to preserve remains, an acceptable alternative may be to arrange prior excavation, during which the archaeological evidence is recorded.' (Department of the Environment 1990, para. 24). Such decisions clearly require the prior knowledge of the survival (or not) of archaeological remains on a development site. This knowledge may be provided by assessment, defined as 'desk-based evaluation of existing information: it can make effective use of records of previous discoveries, ... (Department of the Environment 1990, para. 24), or by field evaluation, defined as 'a rapid and inexpensive operation, involving ground survey and small-scale trial trenching', which is 'quite distinct from full archaeological excavation'. The rationale is that 'Evaluations of this kind help to define the character and extent of the archaeological remains that exist in the area of a proposed development, and thus indicate the weight which ought to be attached

to their preservation. They also provide information useful for identifying potential options for minimising or avoiding damage. On this basis, an informed and reasonable planning decision can be taken' (Department of the Environment 1990, para. 21). A framework for the practical implementation of this policy is provided by MAP2, which breaks down an archaeological field project into five phases (project planning, fieldwork, assessment of potential for analysis, analysis and report preparation, dissemination). It defines the management principles and procedures for each phases, but does not itself offer practical guidance for their implementation. It would not be reasonable to expect it to set out more than general principles, given the wide diversity of archaeological fieldwork.

1.2

Discussion

These two documents, while setting out a quasi-legal framework (and it is important to remember that neither carries the force of law), left something of a vacuum in the practicalities of implementing the policies they put forward. For example, it was not clear how such sites should be selected for evaluation, nor what would constitute a satisfactory level of investigation for a site evaluation. Since one of the aims of PPG 16 is to prevent the sorts of problems that have in the past arisen from the discovery of significant archaeological remains while development is in progress, the criteria for satisfactory evaluation must be that the existence, extent and nature of archaeological remains are established with a degree of certainty that will enable rational decisions to be made about the design of the proposed development, but at a reasonable cost. Clearly, total excavation would provide the required information, but at prohibitive cost (possibly exceeding that of the proposed development in some cases). The problem is essentially one of sample design - choosing the 'best' combination of techniques (geophysical survey, boreholes, excavation, etc.) with their respective samples, where 'best' means providing the required information at minimum cost. 1

1.4

The initial, nai:ve, expectation of archaeologists was that there would be a simple rule, probably expressed in percentage terms, for the definition of a satisfactory sample of a site for purposes of evaluation. A considerable educational effort was made to wean them off this particular dummy; this seems to have been achieved. Archaeologists now seem aware of the difficulties, but as yet no coherent theory or methodology has emerged.

1.3

1.4.1

Method Classical statistics

The problem can be stated thus: given a particular site, we want to design a sampling scheme, consisting of trial trenches and/ or boreholes, that will enable us to say with a chosen level of confidence, that there are no significant archaeological remains on the site, if our sample reveals no such remains. The definition of 'significant' is crucial for sample design. Clearly, we do not mean the presence of isolated artefacts, or even a small individual feature (e.g., an isolated post-hole), but the existence of either a substantial assemblages of artefacts (e.g., a flint scatter), or a functional feature or group of features (e.g., a group of post-holes comprising a hut). It seems reasonable therefore to initially define 'significant' in terms of size, and, to a lesser degree, shape (e.g., 'artefacts or features occupying an area more than 2m in maximum diameter'), while acknowledging that much more work needs to be done on this topic. The impetus for this paper came from an outstanding example of statistical serendipity. I attended a talk given to the General Applications Section of the Royal Statistical Society by Mike Nicholson of the Directorate of Fisheries Research of the Ministry of Agriculture, Fisheries and Food, Lowestoft. His work was concerned with the 'problem of making inferences about the distribution of a species [of shellfish] when it is not observed in a survey area. Absence of the species implies no more than that the sampling points did not coincide with the occurrence of the species, not necessarily that the species is absent from the area' (Nicholson & Barry 1995, p. 74; see also Barry & Nicholson 1993). This seems to me to be mathematically analogous to the archaeological situation: for 'area' read 'site', for 'sampling point' read 'trial trench' or 'borehole' (his 'points' are not literal points, but may be small quadrats or circles), for 'species' read 'archaeological remains' (e.g., flints). He distinguishes two reasons for such surveys:

Aim

The aim of this paper is to look at one small part of this question, the existence (or not) of archaeological remains on a site. The broader questions of assessing their extent, historical significance, likely survival under various schemes of mitigation, etc., will be left to others (e.g., Biddle 1994; Carver 1993). Ours is not such a trivial question as it might at first sight seem to be. While in rural areas valuable evidence may be provided by survey techniques - aerial and geophysical survey, field-walking, etc.- these may not be available or useful in urban areas, for instance because of the presence of existing buildings and services, and restrictions on over-flying. For this reason, evaluation in urban areas can be expected to be more speculative than in rural areas, with a higher proportion of 'negative' outcomes (i.e., evaluated sites on which no significant archaeological remains are found). This does appear to be the case: for example, in London in 1992, 134 out of 188 evaluations were negative, and in 1993, 160 out of 226 (71% in each year; McCracken & Phillpotts 1995, pp. 64-5). It might be thought that this represents a waste of effort, and that zoning should be applied, with evaluations carried out only in areas of high archaeological potential. This would lead to a situation in which fieldwork merely confirmed what archaeologists thought they already knew, and since our knowledge is based on previous fieldwork (itself reflecting the density of archaeological activity in an area, see Hodder & Orton 1976, pp. 20-24), there would be a cycle of self-reinforcing bias. In practice, important discoveries have resulted from speculative evaluation in areas not known for their archaeological potential, e.g., in London at Clapham (Densem & Seeley 1982) and Tulse Hill (Greenwood & Maloney 1995, p. 343). This argument provides the rationale for this paper. A methodology is needed for the evaluation of sites which will probably contain no archaeological remains, but which may contain unexpected remains of unknown nature and importance. The approach must be cost-effective (to avoid suspicion of archaeologists keeping themselves in work by investigating 'dead' sites), but must carry an element of quality assurance, i.e., a negative evaluation must really be negative, otherwise (i) valuable new information may be lost, or (ii) embarrassing discoveries may be made in the course of subsequent development.

1. when the species is beneficial, and we want to be

reasonably sure that we have not missed a significant (e.g., commercially exploitable) patch of it, as in the example of cockles on Holbeach Sands, The Wash (Barry & Nicholson 1993, pp. 35961). 2. when the species is undesirable, and we want to be reasonably sure that it has not invaded an area, as in the case of Manila clams in Poole Harbour (Nicholson & Barry 1995, pp. 76-7). Paradoxically, some archaeological situations may be of the latter type, since the aim may be to be reasonably sure of being right if we say that there are no significant remains on the site.

2

N

In their first paper, Barry & Nicholson (1993) consider the probability of detecting a single circular patch, of radius r, in an area where there is one sampling point per area d2 . That is, if there are N sampling points in an area of size A, d2 is defined by the equation d2 =AN./

log(l - 0v) as a formula (I) for the minimum sample size needed to meet specified values of p and of 0v. Some values of N, calculated for chosen values of 0 and of 0v, are shown in Table 1.1. Note that this table shows the sample size (N) needed for us to be p% certain that the proportion of a site occupied by archaeological deposits is less than a specified amount (0v), in the case when no deposits are actually found in the sample. For example, if we want there to be a 95% probability that the proportion is less than 2%, we need a sample size of about 114. So far, this is all straightforward Classical statistics. It reflects work done in the l"SA in the 1980s (e.g., Nance 1981; Read 1986; Shott 1987) and 1990s (Sundstrom 1993), and in England in the 1990s (Champion et al. 1995). Nicholson and Barry's formula (I) is essentially the same as that quoted by Read (1986, p. 484) and Shott (1987, p. 365) on regional and intrasite scales respectively. Shott considered the detection of a rare class of feature on a multi-feature site, rather than the detection of a 'patch' of archaeological deposits on a 'sparse' site, but the statistical argument is, so far, the same.

They discover that the key statistic is what they call the 'standardized patch radius' R, defined by R = r / d, i.e., the ratio of the patch radius to a notional distance between sampling points. They compare the average probabilities of patch detection for four sampling designs - square and triangular lattices, random design and transect design. They plot the values of this probability for values of R from 0 to 1.0, showing that the lattice designs are superior to the others (with the triangular slightly superior to the square, and the random slightly superior to the transect; Barry & Nicholson 1993, Fig. 4). This conclusion appears to conflict with archaeological expectations, which are perhaps unduly influenced by the experience of Flannery (1976), that lattice designs are inferior to random designs because their sampling points may fall between regularly-spaced archaeological features. The difference in perception arises because Barry and Nicholson's work concerns the probability of detecting a single patch in the area. Archaeologically, one could use this work to calculate a size of sample needed to detect (with specified probability) a single 'patch' of archaeological remains on a site. The higher this probability is, the higher the certainty that no patch is present if one is not detected. In their second paper, Nicholson & Barry (1995) show how the incorporation of prior information (e.g., about the likely density of a species in the area) can, by means of Bayesian statistical analysis, improve the efficiency of a sample design. They start in classical fashion with the case of 'an effectively infinite area in which 0(0 < 0 < l) is the probability that the species is present at a randomly spaced sampling point' and that a sample of N randomly spaced sampling points fails to detect any examples of the species. The estimate of 0 is then zero, but they show that the upper 100p% confidence limit for 0 can be constructed by finding the largest value 0v such that 1- p


1, values of 0 > 0.5 become progressively less likely, while with b < l, values of 0 above 0.5 are more likely (see Fig. 1.1). Prior belief that archaeological remains are unlikely to be found therefore corresponds to high values of b. They go on to show that the value of b acts as the 'sample size of a hypothetical survey from which the species was absent' and give the example that 'a prior belief expressed as being 95% (lO0p) sure that 0 is less than 0.1 (0prior) gives b = 28.4, equivalent to a hypothetical prior survey of 29 sampling points

i.e., 1- p

= _lo_g_(l_-_p_)

> _lo_g_(l_-_P_) log(l - 0v)

They quote McArdle (1990), who derived from this inequality the equation: 3

p

0.90 0.95 0.99

BP 0.01 229 298 458

0.02 114 148 228

0.05 45 58 90

0.10 22 28 44

0.20 10 13 21

0.50 3.3 4.3 6.6

Table 1.1: Minimum size of sample N needed to achieve specified values of p and of 0p,

probability (theta>0.5)

1

0.8 -

0.6 -

0.4 -

0.2 -

Figure 1.1: Prior probability that 0 > 0.5, as a function of b (after Nicholson & Barry 1995, Fig. 1).

0

3

4

5

b

1.5.1

in which the species was not observed.' (Nicholson & Barry 1995, p. 76). This may seem to the archaeologist like mathematical trickery, in that several sampling points have been created out of thin air (or our own prior beliefs). As a counter to such suspicions, Nicholson and Barry point out that the Classical formula (I) is identical to the corresponding Bayesian formula for N only when b = 0, which corresponds to a belief that 0 = l. Since archaeologists would not carry out an evaluation if 0 = l (which means that archaeological remains exist across the entire site), it is reasonable to accept a prior belief in a lower value of 0, and thus in a value of b which reduces the required sample size. Whatever prior belief we choose to employ, we can calculate a sample size that corresponds to chosen levels of p (our confidence level) and of Bp (the upper acceptable limit for the proportion of the site on which archaeological remains survive). The parameter BPmust be chosen to make the area of any surviving archaeological patches (i.e., ABp) 'insignificant'.

1.5

2

Setting the parameters

The archaeologist must first decide on (i) the appropriate definition of 'significant archaeological remains' and (ii) the acceptable posterior probability that there are no such remains on the site, if none are found in the evaluation. Such parameters may well be the subject of future guidelines, but for the time being (s )he is on his/her own. The choice for (i) depends on the type of remain that might be found. Well-defined types would suggest precisely-defined 'patches', for example if grubenhiiuser were anticipated, rectangular patches of 3m x 2m might be specified, while if Iron Age roundhouses were anticipated, circular patches of 10m diameter might be suspected. If there were no strong hints towards a particular type of remain, the preference might be to specify an area or proportion of the site. This might be particularly relevant to urban sites where the question might be not so much whether re~ mains of a particular type once existed on the site but how much (if any) has survived more recent dev~lopment. For purposes of illustration, this approach is adopted, and 'significant archaeological remains' are defined as remains with a total extent of 100m2 . or more on the site. Since 100/10, 000 = 0.01 = 1% this is equivalent to setting the critical value of 0 to 0.01. The choice for (ii) depends on the likely cost of being 'wrong': i.e., what will be the cost if the archaeologist says that there are no significant archaeological remains, but they are encountered in the development

A worked example

Suppose a development site of 1 ha. (100m x 100m = 10, 000m2 ) is to be evaluated for possible archaeological remains. None have been found on the site, but other work in the locality suggests that there may be ~ome. How does the archaeologists design his/her proJect? For purposes of illustration, we suppose that site conditions make the use of geophysical prospection either impossible or inconclusive. 4

(the situation that PPG 16 is supposed to prevent occurring). The greater the cost, the higher the posterior probability should be. This raises all sorts of questions - is the cost that of the delay to the developer caused by the investigation of unexpected remains, or of redesigning the development around them (both of which are likely to be high) or the cost to archaeology of the loss of unrecorded remains (which is so far unquantifiable)? This is a very wide and difficult subject; for the time being the archaeologist avoids it and makes an arbitrary decision, say 90%. To sum up, the archaeologist has decided that (s )he wants to be 90% certain that, if (s)he says there are no significant archaeological remains on the site, the total extent of any remains is less than 100m2 .

1.5.2

Devising

0.144 by the use of 2m-square test pits, and the corresponding value of n is reduced to 158, which appears to be much more expensive than 229 shovel-tests. But two points inflate the comparison: 1. the assumption that the archaeological remains

form a single square patch, and 2. that the remains are more likely to be seen in a large intervention than in a small one. We look at each in turn. 1. the effective area of the target patch is increased

by a proportion that depends crucially on its shape and nature (one patch? several? square? linear?). For example, a linear patch, say 50 x 2m, has an effective area of 52 x 4m, i.e., 208m 2 , so that 0e is increased to 0.0208, and n is reduced to 110. Considering several small patches, for example ten of 4 x 2.5m each, gives a larger increase in the effective area, in this case to 10 x 6 x 4.5 = 260m 2 , corresponding to 0e = 0.026 and leading to n = 87. The process could be carried to ridiculous extremes, for example 100 patches of 1 x Im gives an effective area of 100 x 3 x 3m 2 , 0e = 0.09, and n = 24. Clearly archaeological opinion about the likely size and shape of 'significant archaeological remains' is very important at this stage. It should be noted that the area excavated in even as few as 24 such test pits (24 x 2 x 2 = 96m 2 ) is more than the total area of 229 shovel tests (220 x 0.3 x 0.3 = 20.6m 2 ), suggesting that the most efficient strategy is a very large number of very small interventions. This would be the case, but for the next point.

the strategy

The archaeologist must now design a sampling procedure that will meet these aims at the lowest possible cost. The cost includes not only person-hours worked, but also the total length of the time spent on site (and hence the potential delay to the developer). The aspects that must be decided are: 1. the size and shape of the interventions (e.g., lm 2 test pits; 2m-wide trenches) 2. their number and location 3. the means of excavating them (by hand, mechanical excavator, etc.). These questions are inter-related, and the aim is to find the 'best' combined answer to all three. For purposes of illustration, we start from the mathematically simplest situation, in which the interventions are so small that they may be regarded as mathematical points on the site (e.g., boreholes, or perhaps 30cmsquare 'shovel tests', e.g., Krakker et al. 1983). Application of the Classical formula (I), with p = 0.90 and 0P = 0.01, gives n = 229 as the number of interventions that would be needed. But, as we have seen in 1.4.2, this corresponds in Bayesian terms to the unlikely situation of a prior belief that 0 = I. If we adopt a position of vagueness, e.g., that all values of 0 are equally likely, then b = I and the difference to n is trivial. But if we are prepared to make stronger statements about 0 we can reduce the value of n appreciably. The prospect of undertaking a large number of very small interventions does not appeal (Champion et al. 1995, p. 39), so we consider larger ones, say (for example) 2m-square test pits. Mathematically, increasing the size of the intervention is equivalent to increasing the area of the 'target patch'. For example, a 10 x 10m patch would be hit by points that actually fall within it, but it would be hit by 2m squares whose centres lie up to Im from its edge, i.e., within a total area of 12 x 12m = 144m2 . Thus, in this simple case, the effective value of 0, 0e, is increased from 0.1 to

2. a second important point is that, if significant archaeological remains are present, they are more likely to be detected in a large intervention than in a small one. This is called the 'visibility' of the remains, or the 'site detection probability' and has been discussed by Champion et al. (1995). Its effect is to make small interventions relatively less efficient, since they have a higher probability of not detecting remains even when they are present in them. It may be that at this stage the cost of the outcome of these considerations is seen as too high, perhaps in relation to the development value of the site. If so, it can be reduced only by reducing the value of p (i.e., of increasing the risk of not detecting significant archaeological remains), or of increasing the value of 0P (i.e., of increasing the size of the remains that the archaeologist is prepared to 'write off'). Thus the choices listed at the start of this section depend on a complex interplay between the size and shape of the sorts of patches of significant archaeological remains that might be expected, with their

5

visibility on interventions of various shapes and sizes and dug by various means, and what is perceived as a reasonable cost of evaluation.

1.5.3

(ii) provision of the infrastructure (software and training) that will enable archaeologists to achieve (i). This paper is a first step towards (i); funding for (ii) is currently being sought.

Designing the sample

Acknowledgements

After the archaeological and mathematical complexities of 1.5.1 and 1.5.2, this is a relative formality. The choice lies between a purely random and a more systematic layout of the chosen number of trenches and/or test-pits; there seems to be agreement between archaeologists and biologists that the systematic is probably better, and that within the systematic designs a triangular (also called hexagonal) lattice is probably the best (Champion et al. 1995, p. 39; Barry & Nicholson 1993).

1.6

My prime thanks must go to Mike Nicholson, whose excellent talk was the starting point of this work. I was encouraged to pursue it by Peter Hinton and Kris Lockyear, and this paper has been greatly improved through the perceptive comments of a referee.

References ANDREWS, G. 1991. Management of Archaeological Projects. English Heritage, London.

Conclusion

BARRY, J. & M. NICHOLSON1993. 'Measuring the probability of patch detection for four spatial sampling designs.' Journal of Applied Statistics 20(3): 353-361.

The design of archaeological field evaluations has already passed beyond the stage at which it was believed that there was a minimum sample size ( e.g., 2%) that was needed for a successful evaluation (Champion et al. 1995, p. 36). It is now accepted that design must be based on a complex interplay of factors - type and extent of expected remains, definition of 'archaeological significance', size, shape and method of archaeological interventions, and the probability of recognising different types of remains in different types of intervention. A valuable next step would be to require designs of evaluations to carry formal statements of quality assurance, the most fundamental of which would be a lower limit on the probability of a site containing no significant archaeological remains, given that none are found in the evaluation. This would raise archaeological (almost political) questions about the definition of significant archaeological remains, and about an acceptable probability level of faling to detect them. These problems have always existed, hidden beneath the cloak of professional judgement; it is better hat they be made explicit and discussed openly. At this point, the use of a Bayesian approach has much to offer. First, archaeologists seem to find arguments or requirements based on subjective probabilities easier to grasp intuitively than ones based on Classical hypothesis testing. Second, by incorporating prior knowledge, a Bayesian approach can reduce the sample size needed to meet a specification, and thus reduce fieldwork costs. There are technical questions to be answered, such as the suitability of the Beta distribution (other than its sheer convenience), and the choice of its parameters, but these are not the province of the archaeologist. However, archaeologists will need to gain experience in articulating their prior beliefs. A two-fold approach is needed to advance the subject: (i) advocacy of the practice of designing evaluations so that such probability statements can be made,

BIDDLE, M. 1994. What future for British archaeology? Oxbow Lecture 1. Oxbow Books, Oxford. CARVER, M. 0. H. 1993. Arguments in Stone: Archaeological Research and the European Town in the First Millenium. Oxbow Books, Oxford. CHAMPION, T., S. J. SHENKAN & P. CUMING 1995. Planning for the Past Volume 3: Decisionmaking and field methods in archaeological evaluations. University of Southampton and English Heritage, Southampton and London. Cox, D. R. & D. V. HrKKLEY 1974. Theoretical Statistics. Chapman and Hall, London. DENSEM, R. & D. SEELEY1982. 'Excavations at Rectory Grove, Clapham, 1980-81.' London Archaeologist 4(7): 177-184. DEPARTMEKTOF THE ENVIRONMENT1990. Planning Policy Guidance: Archaeology and Planning. Department of the Environment, London. FLANNERY,K. V. (ed.) 1976. The Early Mesoamerican Village. Academic Press, London. GREENWOOD, P. & C. MALONEY 1995. 'London fieldwork and publication round-up 1994.' London Archaeologist 7(13). HODDER, I. & C. R. ORTON 1976. Spatial Analysis in Archaeology. Cambridge University Press, Cambridge. KRAKKER, J. J., M. J. SHOTT & P. D. WELCH 1983. 'Design and evaluation of shovel-test sampling in regional archaeological survey.' Journal of Field Archaeology 10: 469-480. 6

SHOTT, :.VLJ. 1987. 'Feature discovery and the sampling requirements of archaeological evaluations.' Journal of Field Archaeology 14: 359-71.

MCARDLE, B. H. 1990. 'When are species not there!' Oikos 57: 276-277. McCRACKEN, S. & C. PHILLPOTTS 1995. Archaeology and planning in London. Assessing the effectiveness of PPG 16. Standing Conference on London

SUNDSTROM,L. 1993. 'A simple mathematical procedure for estimating the adequacy of site survey strategies.' Journal of Field Archaeology 20: 91-96.

Archaeology, London. NANCE, J. D. 1981. 'Statistical fact and archaeological faith: Two models in small site sampling.' Journal of Field Archaeology 8: 151-165.

Clive Orton Institute of Archaeology University College London 31-34 Gordon Square London WClH OPY

NICHOLSON,M. & J. BARRY 1995. 'Inferences from spatial surveys about the presence of unobserved species.' Oikos 72: 74-78.

[email protected]

READ, D. W. 1986. 'Sampling procedures for regional surveys: a problem of representativeness and effectiveness.' Journal of Field Archaeology 13: 477-491.

7

2 Experiments

with Detrended Correspondence

Analysis

Kris Lockyear Institute

2.1

of Archaeology, University College London

Introduction segments' (Hill & Gauch 1980). Fig. 2.3 shows, in a simplified fashion, how the method works. The detrending process takes place as part of the calculation of the ordination axes, not as a post-analysis transformation of the scores. The second method uses polynomial curves in the detrending process, rather than segments. The technique has, however, been met with some caution by writers on the subject. Greenacre states that:

Correspondence Analysis (CA) is now a widespread and accepted technique within archaeology (see, for example, the many papers in CAA87 - Ruggles & Rahtz 1988). A common, and oft-looked for, result in CA is a horseshoe curve (e.g., Lockyear 1996a, figs. 12). If the aim of the analysis is the seriation of the objects/variables included, whether the principal gradient be time, or perhaps social status, geographical location or some other variable, then the results can be judged successful. If, however, the principal gradient creating the sequence is already known, then the horseshoe effect can mask other variation which may be of interest. Simply examining lower order axes does not solve this problem as the quadratic curve becomes a cubic curve and so on (Hill & Gauch 1980, p. 48). This problem was encountered in my analysis of coin hoards of the Roman Republic (Lockyear 1996b). A CA which included 241 hoards was dominated by the time gradient, which was already known (Figs. 2.12.2). The dataset was sufficiently large that it was possible to divide it into 22 subsets on the basis of the closing date 1 of the hoard, and to analyse each set individually. The principal problem with this strategy is that it prevented sources of variation which crosscut the closing dates of the hoards to be identified. In order to look at the dataset as a whole, three further statistical techniques were tried. Two of these analyses using Dmax-based Cluster Analysis and Principal Coordinates Analysis have already been presented to CAA and published (Lockyear 1996a; see also Lockyear 1995). The remaining analysis using Detrended Correspondence Analysis (ocA), forms the subject of this paper. 2 This paper will not consider the theoretical background to the technique (for which see Hill & Gauch 1980) but will present an 'experiment' into the use of the method in the analysis of hoard data which will hopefully illustrate the uses and difficulties of using this method for other types of archaeological data. The aim of the method is to, in effect, straighten out the curve and to present the deviations from that curve. In the process, it is expected that the 'bunching' effect often seen at the ends of a horseshoe curve will also be eliminated. Two methods have been used to achieve this. The original method detrended 'by

In the process [of detrending], however, control over the geometry is lost and it is possible that. . . the detrending might introduce further artifacts into the results. (Greenacre 1984, p. 232).

Baxter is also unenthusiastic about the method stating: Since the horseshoe effect is natural to CA in the presence of seriation structure, and since in many archaeological uses an unambiguous ordering is all that is wanted, there may be little need to worry about it. (Baxter 1994, p. 120).

The only archaeological application of DCA of which I am aware is that by Shennan in which he examines assemblages of amber artefacts from Bronze Age Britain (Beck & Shennan 1991, pp. 85-98). Shennan states that: It was decided to carry out a detrended correspondence analysis to avoid the problems of the so-called horseshoe effect which introduces a correlation between the first and second principal axes (Beck & Shennan 1991, p. 85).

Unfortunately for our purposes the normal CA of the data was not presented. Shennan also used a further method, Detrended Canonical Correspondence Analysis which compares the ordination axes with a further set of axes derived from 'environmental' data, in this case associated finds, but found that his data were too sparse and the results were not significant (Beck & 9

0

0

o:i

o:i

+

+

...

..

•. ......

. .....: . . ~

-2.5

118+ +i14 113++ +fl:102 103 +

+2.5

+61 +58 00

101+ 2+ /

-2.5

1i~ 7+64

97

+

+73

70

+2.5

.... ..

72+tt-tt-7f+~ 86 T

., ..

...

Figure 2.1: Species map from CA of 241 Roman Republican coin hoards. Data points are years of issue BC.

Figure 2.2: Sample map from CA of 241 Roman Republican coin hoards. Data points are coin hoards.

0

0

t

0 0

0

0

0

0

0

0

0

0

Figure 2.3: Detrending by segments. The gradient along axis 1 is divided into a number of segments, and then within each segment the values on axis 2 are adjusted (after Hill & Gauch 1980, Fig. 3).

0

0

0

0

0

0 0

•

•

•

•

Axis 1

•

•

• •

•

0

•

• •

•

•

---+

Cluster f. This cluster consisted almost entirely of Italian hoards the majority of which closed in the 40s BC. Only two Romanian hoards were included in this cluster, one closing in 42 BC and one in 29 BC. Four Italian and the two Romanian hoards were selected.

Shennan 1991, p. 91). The final method he employed derived a correlation matrix between the ordination axes and the associated finds. 3

2.2

Examining a test data set

Cluster g. This cluster of hoards included 11 Romanian and 7 Italian hoards. The Italian hoards mainly date to the late 50s-early 40s BC whereas the Romanian hoards date to the mid- to late 40s BC. Two Italian and four Romanian hoards were included in the test data set.

It was decided to test the method on a relatively small data set. Twenty-four hoards were chosen consisting of 11,161 denarii. These hoards had been included in the cluster analysis previously (Lockyear 1996a) and came from clusters b, f and g.4 All the hoards either came from Italy or Romania (see Table 2.1). To recap, the principal features of these clusters were:

2.2.1

Cluster b. This was the largest cluster from the analysis and consisted principally of Italian hoards closing from 82-71 BC, or Romanian hoards which could close anywhere from 77 to 32 BC. Six Italian and six Romanian hoards where chosen from this group.

Analysis one -

'ordinary'

CA

CA was performed on the test data set using the package CANOCO. Asymmetric maps were produced (Figs. 2.4-2.5) 5 . The first two axes accounted for 48.2% of the variation in the data set. The results of this analysis illustrate a number of classic features encountered in the analysis of coin

10

cluster

code

hoard

country

b b b b b b b b b b b b

CAR cos CST ass PL2 VPT CUC FA2 FND GUR sos SF! BOR CRl CR2 SPN BPT SE! CAS GRA ILi PRS TI2 VIS

Carovilli Casa Castelnovo Ossero Palestrina Villa Potenza Cuceu Farca§ele II Frauendorf Gura Padinii Sala§ul de Sus Sfinte§ti Borzano Carbonara Carbonara Spoiano Bran Poarta $eica Mica Casaleone Grazzanise Ilieni Poroschia Tirnava Vi§ina

Italy Italy Italy Italy Italy Italy Romania Romania Romania Romania Romania Romania Italy Italy Italy Italy Romania Romania Italy Italy Romania Romania Romania Romania

f f f f f f g g g g g g

closing date

82 74 71 72 74 71 48 42 56 32 54 71 42 48 36 46 42 29 51 54 46 39 46 41

hoards (which in themselves may illuminate the results of other less well-dated archaeological assemblages). The sample map (Fig. 2.5) has cluster b hoards, those which have an 'Italian 70s BC' profile, plotted close together in the bottom left-hand quadrant of the map in a tight group. Although cluster g hoards form a separate group on the map, the four Romanian hoards are plotted close to the cluster b hoards while the two Italian hoards are plotted at the top of the second axis. Cluster f hoards are plotted towards the right of the map, relatively close together on the first axis, but spread out along the second. The two Romanian hoards (BPT and SEI) are 'pulled away' towards cluster b. The variable map (Fig. 2.4) shows a horseshoe curve with the variables, in this case years of issue BC, plotted in an approximate sequence. We can interpret the axes in the following manner: the first axis represents time with the earliest issues to the left, and the latest issues to the right; the second axis represents relative abundance of middle period coins (roughly those from the 60s and 50s BC) at the top, with relative (or absolute) lack of those issues to the bottom. This is a classic seriated sequence. If we take the six Italian hoards on the right side of the map, they are in order with Carbonara (CR2) closing in 36 BC, followed by Borzano (BOR, 42 BC) through to Casaleone (CAS, 51 BC), with Grazzanise (GRA, 54 BC) lying on the return of the curve. The Romanian hoards in clusters f and g, despite having overall profiles like their Italian counterparts as defined by the cluster analysis, are at a detailed

total

40 1999 391 465 357 411 484 113 563 232 103 91 582 383 2371 264 59 346 712 256 108 541 148 139

Table 2.1: Details of the hoards used in the correspondence analyses. The total number of corns cited are those coins which have been identified to a reasonable degree of accuracy.

level still 'drawn towards' cluster b with its 70s BC profile, i.e., still have more of this early coinage than the Italian hoards. This reinforces the results of the cluster analysis previously reported (Lockyear 1995, 1996a). The spacing of the hoards on the map is also of interest. The within-cluster variation of these groups (using the Dmax values as an approximate indication) is more-or-less the same, but the dispersion of the hoards on the maps is anything but. This can be explained relatively simply. Table 2.2 represents ten hypothetical hoards and 12 coin types. All hoards have types A-C, only 7 have E-F and 3 hoards have J-1. Although the hoards a- 1 vary amongst themselves just as much as hoards 0-K,, the fact that all the hoards have types A-C means that the variation appears less significant on the CA maps than the variation in types J-1. In other words, hoards a- 1 and types A-C will be plotted close together on the maps, whereas hoards 0-K, and types J-L will be spread out, despite intra-group variation being more-or-less equal. Another good example of this can be drawn from the analysis of two chronologically overlapping data sets, the first containing hoards closing 147-118 BC, the second containing hoards closing 118-108 BC (Lockyear 1996b, sections 8.3.2-8.3.3, pp. 166-173). In the analysis of the first data set the three hoards closing in 118 BC were very widely spaced on the resulting map, but the same three hoards were plotted very close together on the map from the second analysis (Lockyear 1996b, Fig. 8.15 cf. Fig. 8.18b).

11

C>

+

+69 +50

56++55 67+ +66 63+ + +58 +57 62 60+ +49

+127 211+ 179+ 141+ 118+ + +76 129+

10/

*t+

+61 +64

+s8

-1.0

+209 + 29. 31. 37

7~

48 54 +51

,;lj +

+1.0

:::42+43

65 39

.t

+36, 38

Figure 2.4: Species map derived from ordinary CA of 24 hoards as listed in Table 2.1. Data points are years of issue. First (horizontal) and second axes of inertia .

C>

• Group b Italy o Group b Romania ■ Group f Italy □ Group JRomania • Group g Italy b. Group g Romania

C>

+

•

cas

•spn

ti2 6

. '"

VlS

6 6

ill

prs

-1.0

n

+1.0

fa2o~~ cos

car

pl2~vpt oss sff osds

•

est □

bpt ■

cr2

Figure 2.5: Sample map from ordinary CA of 24 hoards as listed in Table 2.1. Data points are coin hoards. First (horizontal) and second axes of inertia.

C>

12

A

a f3

'

5 E

( Table 2.2: Table showing ten hypothetical hoards (a-i,;) with twelve hypothetical coin types (A-L). o represents a low occurrence of that coin type in the hoard; • represents a high occurrence. See text for details.

2.2.2

Analysis two -

detrended

T/

0 K

B

C

D

E

F

• • • • • • • • • • • • • • • • • • • • • •

G

H

I

J

K

L

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

• • • • • • • • •

0 0 0

• • • • • 0

0

0

0

$eica Mica on the sample map, but not Bran Poarta. Dmaxobs for these two hoards is 11.1% and the application of the two sample Kolmogorov-Smirnov test shows no significant difference at the 0.01 % level. Plotting these two hoards as cumulative frequency curves reveals their inherent similarity over the whole of the curve which, when allowing for variation due to Bran Poarta's small size, results in the two hoards appearing on the plot in much the same location - see Fig. 2.8, cf. Fig. 2.7. This clearly demonstrates the danger of interpreting symmetric maps without reference to the original data or diagnostic statistics. As Greenacre (1984, p. 65) states:

CA

CA"IOCOpresents the user with a wide variety of options. For this experiment it was decided to use detrending by third order polynomials, 6 and to produce asymmetric maps as before, which are shown in Figures 2.6-2.7. The first two axes accounted for 43.7% of the variation in the data set. As can be seen, the technique has removed the horseshoe pattern from the distribution of hoards on the map (Fig. 2.7). Obviously, the hoards are in the same order on the first axis as previous analysis but the second axis is somewhat different, and affects the overall map. Within group f the Italian hoards (SPN, CRl, BOR & CR2) now cluster tightly whereas the two Romanian hoards (sEI & BPT) form an isolated pair at the top of the plot. The twelve hoards from group b on the left of map are now split into Romanian hoards with negative scores on the second axis, and Italian ones with positive scores. Cuceu (cue) is, however, nearer to the Italian group. The final group, g, falls between the other two groups on the first axis but is spread along the second with, notably, the two Italian hoards in this group separated from the Romanian hoards and the latter group clustering near to the Italian hoards of group b. The detrended analysis has achieved its two aims of removing the horseshoe curve and counteracting the bunching effect discussed above. The analysis does raise some questions, which in light of Greenacre's comments regarding 'artifacts' in the results, must be examined. These are:

The display of each cloud of points indicates the nature of the similarities and dispersion within the cloud, while the joint display indicates the correspondence between the clouds. Notice, however, that we should avoid the danger of interpreting distances between points of different clouds, since no such differences have been explicitly defined.

This analysis is a classic example of the problem, and is one reason why I prefer to present CA maps as two separate figures rather than one joint map. The second question is how do these two hoards vary from the others in group f? Fig. 2.8 shows that the Italian hoards have relatively more new coin and relatively less old coin. The maximum cumulative difference being reached in 74 BC. The extremely jagged nature of the maximum difference line is mainly due to the small sample size of Bran Poarta. The last question is, are the two sub-groups of b really different? To examine this problem Fig. 2.9 plots eight of the hoards from that group. As can be seen, the Romanian hoards have relatively more old coin when compared to the Italian hoards which have relatively more new coin. This is despite the fact that three of the Italian hoards have early closing dates. Carovilli (CAR) has no coin until 136 BC but closes first in 82 BC. This hoard is, however, the smallest in the analysis and it is not surprising that it has little of the older coinage. The maximum difference between hoards is reached in c. 117 BC and remains relatively level until hoards start closing in the 70s.

1. In what way are $eica Mica and Bran Poarta

(sEI & BPT) similar to each other? 2. In what way are they different from the other hoards including those in their own group? 3. Is the division within group b real, and if so what is it? Five years (species) have extreme positive values on the second axis: 29, 31, 37, 207 and 65. $eica Mica has one coin from each of these years. Of the other hoards, only Cosa (cos) has a coin of 207 BC and Carbonara (CR2) has a coin of 65 BC; no other hoards have coins of 29, 31 or 37. This explains the position of

13

• Group b Italy o Group b Romania ■ Group J Italy □ GroupJRomania "Group g Italy t. Group g Romania

3Lp7 29

□

sei □

bpt

+2

~

•est pl2.vpt cos• • •carti1,;.ili oss

'!/1

48+

1i9 142+ +

42~43

-0.5

+ + + 32

0

-0.5

+0.5

.,.spn er! bor ■

cue prst.,

cr2•

+0.5

VlS

ii;++ !°if++ 51 57Ji..JH56 +68

sfiO ofa2 0

+59

gur

0

sds

0

fnd

Figure 2. 7: Sample map from CA detrended by third order polynomials. Data points are hoards. First (horizontal) and second axes of inertia.

Figure 2.6: Species map from CA detrended by third order polynomials. Data points are years. First (horizontal) and second axes of inertia.

2.2.3

Analysis three set

"

cas

the full data

These maps are capable of historical interpretation. Hoards from c. 150 BC to c. 55 BC come principally from three regions, Italy, Spain and Romania. The Romanian hoards, on the whole, resemble Italian hoards of the 70s BC as discussed previously; Spanish hoards are not hugely different from the Italian ones. Towards the end of the Republic coinage started to be struck at a wider variety of locations by the various protagonists in the Civil Wars. This leads to wider differences between hoards. Under Augustus, the Roman coinage system is imposed on the western provinces, while at the same time the incidence of hoarding in Italy decreases quite dramatically. The more widely dispersed hoards and the lack of a strong Italian 'benchmark' leads to very complicated patterning within these later hoards that makes the interpretation of the results of CA difficult ( e.g., Lockyear 1996b, pp. 240-246).

The full data set of 241 hoards was re-analysed using DCA and the maps presented in Figures 2.10-2.11. The first two axes accounted for 26.9% of the variation in the data set. The initial impression is that this analysis is little more use than the original one. The data points of the right side of the maps are plotted on the first axis in a tight sequence. On the left side of the map the distribution on the second axis is extremely spread out. Examining the species map (Fig. 2.10) we can see that the early issues /hoards are on the right- hand side of the map ( e.g., Petacciato, PET, which closes in 141 BC). On the other hand, most of the years in the top left quadrant of the map post-date the battle of Actium, i.e., are issues of Augustus. The issues at the extreme of the distribution in the bottom left quadrant are those of the 40s BC, i.e., during the Civil Wars. The hoards labelled in the top-left quadrant are from across Europe: Penamacor (PEN) in Portugal, Bourguiel (sou) in France, Breaza (BRZ)in Romania and Zara (ZAR)in Italy. The hoards labelled in the bottom-left quadrant were identified as having exceptional quantities of coins from the 40s BC.

2.3

Conclusions

How much use is DCA? The analyses presented here are often, with hindsight, relatively predictable when one compares them with 'ordinary' CA. The method does, however, sometimes reveal aspects of the data

14

100% 90% 80%

r 70%

1/

;,,

60%

1/,,,,'' /:,'

50%

(',:, '

!;.

40%

,,

'

30%

maximum /11ifference

,

,---"/_,-_/'

20%

- - -

r -

10% 0%

1111

157

I

111111

147

I

111

-

-

;

1111111111111111111111111111111111111111111111111111

137

127

117

107

97

I

87

111111

I

77

111111

67

111111

57

1111

I

I

47

I

I

111111111

37

year BC

Figure CRl,

2.8: Cumulative percentage curves for BPT (upper solid line), SE! (dashed line), SPN, & CR2 ( dotted lines). Bottom solid line is the maximum difference between hoards.

BOR,

100% 90% 80% 70% 60% 50% 40% 30%

maximum difference

20%

/

10% 0% 157

147

137

127

117

107

97

87

77

67

57

47

37

year BC

Figure 2.9: Cumulative percentage curves for eight hoards from cluster group b used in detrended correspondence analysis. CAR, cos, CST and oss from Italy (dotted line) and FA2, GUR, FND and sos from Romania (solid line). Bottom solid line is the maximum difference between hoards.

15

~

+

18+ 19+ 25+ 28+

+15

+32 33+ 12+ 29 4-36 17+ +11 +,t-

-t+ +2 + +52 ++53

-I+:+

+

47

-1.0

48~

207

+1.0

49 46

Figure 2.10: Species map from CA detrended by second order polynomials. Data points are years of issue BC; cf. Fig. 2.1.

~ '";-

pen• bou•

~

+

mog• •err

brz •

ezar

elmp

... . •

-1.0

+1.0

sen ■

p03e

•

p07

Figure 2.11: Sample map from CA detrended by second order polynomials. Data points are 241 hoards; cf. Fig. 2.2.

16

BECK, C. W. & S. J. SHENKAN1991. Amber in British Prehistory. Oxbow Monographs 8. Oxbow, Oxford.

not visible on the original maps, such as the division of group b in Figure 2.7, and perhaps makes one look more closely at features such as the $eica Mica and Bran Poarta hoards ( cf. Figs. 2.5 with 2.7). the technique can show some other aspects of very large data sets, such as the 241 hoards presented here, but can still suffer from simply being too big. Although DCArequires more careful interpretation than CA, it does seem worthwhile to try the technique if a data set exhibits a strong horseshoe curve, and to then compare and contrast the results of the two analyses. I would not recommend the use of the method without prior analysis by ordinary CA, and careful reference back to the original data is also a necessity.

GREENACRE,M. J. 1984. Theory and Applications of Correspondence Analysis. Academic Press, London. GREENACRE, M. J. 1993. Correspondence Analysis in Practice. Academic Press, London. HILL, M. 0. & H. G. GAUCH 1980. 'Detrended correspondence analysis: an improved ordination technique.' Vegetatio 42: 47-58. LOCKYEAR,K. 1995. 'The supply of Roman Republican denarii to Romania.' Studii §i Cercetari de Numismatica 11: 85-102. Published 1997.

Notes

LOCKYEAR, K. 1996a. 'Dmax based cluster analysis and the supply of coinage to Iron Age Dacia.' In H. Kamermans & K. Fennema (eds.), Computer Applications and Quantitative Methods in Archaeology CAAgs, pp. 165-178. Institute of Prehistory, University of Leiden, Leiden. Analecta Praehistorica Leidensia 28.

1. i.e., the date of newest coin in the hoard.

2. This type of analysis was originally provided by the program DECORANA but has been re-implemented in CANOCO (ter Braak 1987-1992) and WIN-BASP. CANOCO was used in the analyses here and I would like to thank the Dept. of Archaeology, University of Southampton for allowing me access to this program while I was working on my Ph.D. 3. Shennan also used the package

LOCKYEAR,K. 1996b. Multivariate Money. A statistical analysis of Roman Republican coin hoards with special reference to material from Romania. Ph.D. thesis, Institute of Archaeology, University College London.

CANOCO.

4. In this paper I have adopted the term 'clusters' for the groups derived from the cluster analysis to differentiate them from 'groupings' seen on the CA maps.

RUGGLES, C. L. N. & S. P. Q. RAHTZ (eds.) 1988. Computer and Quantitative Methods in Archaeology 1987, Oxford. British Archaeological Reports International Series No. 393.

5. For the difference between symmetric and asymmetric maps see Greenacre (1993) or Shennan (1997). 6. Detrending by second order polynomials produces very similar results to those presented here. Fourth order polynomials pull in the more extreme negative values on the second axis. Other experiments not presented here were also undertaken (Lockyear 1996b, 307-311).

SHENNAN,S. J. 1997. Quantifying Archaeology. Edinburgh University Press, Edinburgh, second edition. TER BRAAK, C. J. F. 1987-1992. CANOCO- a FORTRAN program for Canonical Community Ordination. Microcomputer Power, Ithaca, New York.

References

Kris Lockyear Institute of Archaeology, University College London 31-34 Gordon Square London WClH 0PY [email protected]

BAXTER, M. J. 1994. Exploratory Multivariate Analysis in Archaeology. Edinburgh University Press, Edinburgh.

17

3 Vessel volume as a factor in ceramic quantification: case of African Red Slip Ware

the

John Hawthorne Department of Archaeology, University of Southampton

3.1

Introduction

This paper suggests that by consistently overlooking the role of vessel size as a factor in assemblage formation, archaeologists may sometimes have misinterpreted patterns in the ceramic record. Using the example of African Red Slip Ware, it is shown that a pattern, which up until now has been interpreted as an economic decline in the middle of the production, may in fact be merely the result of a move to communal eating. It is concluded that although research on techniques of quantification is certainly desirable, it is to the actual manner of use of ceramics where more attention should be directed.

3.2

the nature of the ceramic record and its relationship with quantification, we may in some cases be wildly off the mark. Specifically, it is argued that the role of vessel volume, or vessel capacity, has been ignored and that it may be much more important for understanding the results of ceramic quantification than is generally recognised.

3.3

African Red Slip Ware and economics

African Red Slip ware (ARS) is a particularly common later Roman fineware, current from the late first century AD to approximately the seventh. It was made in the area of modern Tunisia and was exported all over the Mediterranean. It is orangey-red in colour and the form series consists mainly of bowls, plates and dishes. The standard work of reference remains Hayes' Late Roman Pottery (Hayes 1972, 1980), although this is partially supplemented by the Atlante (Carandini 1981). The British excavations at Carthage have yielded important, if controversial dating evidence for the later wares (Fulford & Peacock 1984). The importance of ARS to Roman archaeology lies in its abundance and the relative ease with which it can be dated. This has been of invaluable use to field survey in the west Mediterranean in particular, as it has allowed field-walkers to reconstruct settlement patterns. In recent years, however, quantification of ARS from sites all around the western Mediterranean has suggested that similarities in the relative quantities found may be related more to economic patterns within Africa than to local economics (Cambi & Fentress 1989; Fentress & Perkins 1988). One of the original conclusions of the South Etruria Survey (Potter 1979), that there was a decline in settlement in central Italy in the third century, can not be taken too seriouslv now as it has been shown that the same pattern ocdurs elsewhere at the same time (Millett 1991). :.\foreover, as Fentress & Perkins (1988) have shown, these fluctuations in quantity are paralleled by changes in the rate of urban construction in Africa. Figures 3.1 and 3.2 show the quantities of ARS through

Ceramic quantification

The quantification of archaeological ceramics is a well established part of the discipline, certainly in Britain, and increasingly elsewhere in Europe. Most archaeologists are at least aware of the distinctions between sherd and weight counts, and many are also familiar with Estimated Vessel Equivalents (EVEs). The reasons for counting pots are equally familiar: as everybody knows, pots are counted to provide information on trade, demography, site status and so on. However, there may be a potential problem in the making here; at the same time that there is complacency with regard to the reasons for quantification, and general acceptance that it is probably a good thing on the whole, there is also a move towards ever more sophisticated methods for actually doing the counting. Orton et al. (1993) provide an overview in layman's terms of some of the more exotic methods. The consensus seems to be that the way to get the pots to yield more information, or more accurate information, is to apply more complex techniques to them. Yet by doing this we are in danger of taking it for granted that we already understand the causes of variation in ceramics. It is assumed that the more accurate information provided by the different counting methods can be readily interpreted in terms of trade, demography, or whatever the particular study is interested in. It is the aim of this paper to demonstrate that this complacency is misguided, and that far from already understanding

19

3.4.1

time on consumer sites in the west Mediterranean and building inscriptions from Africa. The correlation seems clear, as this pattern in the inscriptions is only an African phenomenon (Duncan-Jones 1990). Interestingly, a similar pattern is also seen in the funerary inscriptions of Africa (Fig. 3.1; Meyer 1990). Again, this is only an African phenomenon. The evidence therefore would seem to point to an economic crisis in Africa. This conclusion, that there was an economic crisis is the most obvious one, given the preoccupations of Roman ceramicists in particular, and ceramic quantification in general. It will be argued here, however, that in fact it is not economics which is the culprit but eating habits, represented by changing vessel capacity.

3.4

Explaining

the patterns

The results of counting sherds in this way are shown in Figure 3.1. This is the mean of seven survey sites. The point of particular interest is the period from the late second through to the early fourth century. At this time there is a very definite peak, drop and then recovery in the frequency of sherds. This occurs at almost all sites in the west Mediterranean which were receiving ARS. As noted, this pattern has been interpreted as resulting from an economic crisis. Here it is argued that it makes more sense if seen as the result of a change in the size of the vessels.

3.5

Vessel capacity

Vessel capacity is a particularly under-studied aspect of archaeological ceramics. It has been looked at from several predictable angles, such as a means of classifying vessels (Rice 1987, pp. 219-25) and as a factor in vessel standardisation (Rottlander 1966, 1967). Peacock & Williams (1986), looking at Roman amphorae have considered it as a factor related to vessel efficiency in terms of the ratio of vessel capacity to weight. Yet it seems that it has never been connected to changes in vessel frequency in studies such as that of Fentress & Perkins (1988). This is perhaps strange given that the connection seems obvious: if pots become very large, it is only to be expected that fewer will be needed. This point is returned to below when a critique of the current philosophy behind quantification will be presented, but for now it is sufficient to consider the more prosaic aspects of capacity. The vessel capacities used in this study were calculated using the AutoCAD Advanced Modelling Extension. This makes the task of capacity calculation considerably easier than would be the case if the paper and pencil methods suggested by Rice (1987) were used; indeed, the lack of appropriate software immediately to hand for most archaeologists must be a very significant factor prohibiting its more general study. The capacity was calculated for each form in the series, over 100 in all.

Data manipulation

The methodology used to create the graphs in this paper is the same as that used by Fentress & Perkins (1988, ftn. 12). This involves the analysis of field survey data: as survey data is less affected by stratigraphy than excavation data, its use avoids the problem with excavations whereby, for example, fifth century pottery could be over represented simply because more fifth century earth has been excavated. The dating of the recovered ceramics therefore comes from the stylistic features of the pots themselves. All data used here began as rim sherd counts, as it is rare to find much more in the average :VIediterranean survey report. However, an attempt has been made to convert these figures into crude vessel equivalents by calibrating the sherd counts by the average breakage rate for each form (Fig. 3.3). This breakage rate has been calculated from over 500 sherds from Carthage and Lepcis Magna (data from the latter site kindly lent by M. Attree). This vessel data has been used for the final analysis presented here; otherwise sherd data has been used. Once the sherds are identified and dated, the number of sherds of each form are divided by the duration of that form in years. Thus if a form lasts 50 years and has 100 sherds, there will be two sherds per year. This is done for all forms, and totals are summed. This was done using a spreadsheet divided into five year periods. This is very crude, but as Cambi & Fentress (1989, p. 76) have noted, the data is simply not sophisticated enough for more complicated analysis. The suggestion that it may be more realistic to use a weighted distribution to divide up the sherds per form per year (Fentress & Perkins 1988, ftn. 12) is impractical given the crudeness of the dating: many forms are still only dated to half or quarter centuries, which, combined with a centrally biased distribution, for example, would have the effect of producing regular lumps and troughs on the graphs, centred on quarter and half century points. It is sufficient to use the methodology as it stands, as here we are only interested in very broad, general patterns.

3.5.1

Vessel capacity and counts

ARS

sherd

The change in mean vessel capacity for the ARS form series is shown in Figure 3.4. A comparison with the sherd count, Figure 3.1, is interesting. It can clearly be seen that the drop in sherds in the third century is almost exactly matched by a rise in average vessel capacity. The reason for this dramatic change in capacity is that in the third century the dominant form becomes the very large, flat based dish (Fig. 3.6), whereas in the second century the most common form was the small bowl (Fig. 3.5). The change is very dramatic indeed: the second century forms averaged

20

ARS % perannum

sherds

2.5

30

25

20

1.5 15

10

0.5

yeru-AD

year AD

Figure

3.1: Sherd count.

Figure 3.2: Building inscriptions in Roman Africa, by year. After Fentress & Perkins (1988, Fig. 4).

vessels per year

volume(cubic cm)

1.2

10000 9000

8000 7000

0.8

6000

0.6

5000

4000

0.4 3000 2000

0.2

1000

year AD

Figure

year AD

3.3: Vessels per year.

Figure

3.4: Mean vessel volume.

If we consider the whole period first, the following partial correlations may be observed:

around 20cm in diameter with an average capacity of c. 3 litres, those of the third were frequently up to half a metre in diameter, with an average capacity of around 10 litres. These parameters are remarkably rigid: of around two dozen second century forms, only one approaches the third century sizes, and of the ten or so late third century forms only two have variants which are sometimes found in second century sizes. The pattern in the fourth to seventh centuries is less clear-cut, as there is a mixture of forms and sizes, but the peak of sherds in the earlier sixth century does correspond to a time of increased use of smaller bowls, even though this is not apparent from Figure 3.4. This is the subject of work in progress. The most obvious question to ask next is: what sort of statistical interplay exists between sherd counts, vessel breakage rates and vessel capacity? Data is still being collected on vessel breakage rates in order to answer this question with complete confidence, but at present the analysis of over 500 rim sherds indicates that breakage rates are, for the first three centuries AD at least, less significant in explaining variation in sherd frequencies than vessel capacity.

capacity sherds break

capacity 1.000 -0.648 0.331

sherds -0.648 1.000 0.087

break 0.331 0.087 1.000

Substantially more variation in the sherd frequency is explained by capacity than by the breakage rate. In fact, the breakage rate does not seem to play much part at all. This is borne out by Figure 3. 7, which shows that the average breakage rate fluctuates little over time. In contrast, capacity has a strong negative correlation, as we would expect from Figures 3.1 and 3.4. Further support for the notion that breakage rate is of little significance comes from the total vessel calculation (i.e., sherds calibrated by breakage rate) shown in Figure 3.3. Comparison with Figure 3.1 shows that there is little difference, with the exception that the later period has less pottery than the earlier. Of considerable interest, however, is the fact that the large gap of the third century remains unchanged by the transformation of sherds into vessels.

21

10cm

Figure

10cm

3.5: Second century bowls. After Hayes (1972,

Figure

Fig. 3).

Fig. 12).

20 mean no. of sherdsper form

1,400,000

3.6:

Third century dishes. After Hayes (1972,

arbitraryunits

1,200,000

15

1,000,000

800,000

10 600,000

400,000

200,000

year AD

Figure

year AD

3. 7: Mean breakage rate.

Figure

3.5.2

Treating the whole period in this manner masks some more detailed patterns, however. If the series is divided into two halves, with the dividing point being AD 350, then it can be seen that the earlier period is much more strongly influenced by capacity than the later period. In the later period the correlations of sherds with capacity and breakage rate are almost equal, although the capacity coefficient is still negative. As we would expect in both cases there is a reasonable positive correlation of capacity with breakage rate: as the pots get bigger they get more fragile. However, the most striking aspect is that which we would not necessarily have expected: that vessel capacity has more explanatory power than the degree of fragmentation of the vessels. This is also supported by regression analysis. If the whole period is considered in a multiple regression, R 2 is 42.7%. This compares with 3.13% for breakage rate on its own in a simple regression. Vessel capacity, on the other hand, accounts for 43.31 % of the variation in a simple regression. The figures for each period are shown in Table 3.3.

3.8: Overall volume.

An alternative explanation the economics of ARS

of

This demonstration that vessel capacity correlates strongly with sherd frequency, and that vessel breakage rate is not a major factor, opens the way for new interpretations of the 'economic' patterns outlined at the start of this paper. The most dramatic re-interpretation we may make is that the apparent decline of African exports in the third century is not actually the result of an economic crisis, as Fentress & Perkins (1988, p. 213) and Cambi & Fentress (1989, p. 76) have argued. Instead, we may see it as perhaps resulting from the increasing size of the vessels at this point in time, with the effect that fewer vessels were required to eat the same amount of food as in the second century. However, the analysis thus far does not let us state this with confidence: we know that the increasing capacity is strongly correlated with the declining sherd count, and that this is not a product of the breakage rate, but we cannot be sure that there is not still economic decline, with the increase in capacity insufficient to make up for the drop in the

22

capacity sherds break

capacity 1.000 -0.738 0.308

sherds -0.738 1.000 -0.15

break 0.308 -0.15 1

Table 3.1: Partial correlations for early period to AD 350).

multiple R 2 simple capacity r 2 simple breakage r 2

(AD

capacity sherds break

100

capacity 1.000 -0.265 0.783

sherds -0.265 1.000 0.238

Table 3.2: Partial correlations for late period AD 600).

whole period 42.70 43.31 3.13

early period 68.90 69.55 34.64

break 0.783 0.238 1.000 (AD

355 to

late period 3.30 1.68 0.26

Table 3.3: Regression coefficients for multiple and simple analyses (percentages).

number of vessels. Therefore an additional calculation has been undertaken to address this. The aim of this new calculation is to see whether the total capacity of all the vessels in the average assemblage changes markedly over time. To do this, the sherd counts for each form were calibrated by the breakage rate to give an approximation of the number of vessels within each form class. This figure was then multiplied by the vessel capacity for each form and run through the allocation-to-years procedure described above. The result is Figure 3.8. This plot is rather spikey, certainly, but it can be clearly seen that in contrast to the sherds or vessels graphs, there is no major gap in the third century. This would seem to support the idea that the increase in vessel capacity was directly responsible for the decrease in the number of vessels. It remains then to ask what the change in vessel size means, and how could it possibly have such an effect - surely larger plates just means that people ate off larger plates? The answer would seem to lie in the move to communal eating at this point (Carandini 1981, p. 15; Hawthorne 1996, pp. 3-6). Larger plates in this context represent the serving vessels for several people at once. Whereas the second century bowls were small, about the same size as individual bowls found in any British house, the third century

dishes were of such a size that they must have been used by several diners at the same time. This communal dining system, where several people eat from the same dish, is well documented for medieval Europe (Hammond 1993; Farb & Armelagos 1980, pp. 204-8; Mennell 1985; Braudel 1973, pp. 124-39). It seems not unreasonable to conclude that such a system would require fewer plates per diner, although of a greater size than individual plates. This would then explain the drop in the number of vessels in the third century.

3.6

Conclusion

It has been argued that one of the most significant factors in explaining change in the quantities of ARS found around the western Mediterranean is the average vessel capacity. This means that the traditional view that a third century economic crisis is represented in the data is open to question; it seems more likely that the data represents merely a change in eating habits, with very large pots indicating a move to communal eating. This demonstration may also have implications for the way in which ceramic quantification is undertaken more generally, as it clearly shows that day-to-day factors like eating habits may influence the results of quantification as much, if not more, than traditional factors such as trade.

23

References

HAYES, J. 1972. Late Roman Pottery. British School at Rome, London.

BRAUDEL, F. 1973. Capitalism and Material Life, 1400-1800. Weidenfeld & Nicolson, London.

HAYES, J. 1980. A Supplement to Late Roman Pottery. British School at Rome, London.

CAMBI, F. & E. FENTRESS 1989. 'Villas to castles. first millenium AD demography in the albegna valley.' In K. Randsborg (ed.), The Birth of Europe: Archaeology and social development in the first millenium AD, Analecta Romana Instituti Danici, Supplementum XVI, pp. 74-86. L'Erma di Bretschneider, Rome.

MENNELL, S. 1985. All manners of food: eating and taste in England and France from the Middle Ages to the present. Basil Blackwell, Oxford. MEYER, E. A. 1990. 'Explaining the epigraphic habit in the Roman Empire. The evidence of epitaphs.' Journal of Roman Studies 80: 74-96.

CARANDINI, A. (ed.) 1981. Atlante della Forme Ceramiche. I. Ceramica fine romana nel bacino mediterraneo (media e tardo impero). Instituto della Enciclipodia Italiana, Rome.

MILLETT, M. 1991. 'Pottery: Population or supply patterns? the Ager Tarraconensis approach.' In G. Barker & J. Lloyd (eds.), Roman landscapes : archaeological survey in the Mediterranean region, pp. 18-26. British School at Rome, London.

DUNCAN-JONES,R. 1990. Structure and scale in the Roman Economy. Cambridge, Cambridge University Press.

ORTON, C. R., P. TYERS & A. VINCE 1993. Pottery in Archaeology. Cambridge University Press, Cambridge.

FARB, P. & G. ARMELAGOS1980. Consuming Passions. The Anthropology of Eating. HoughtonMiffiin, Boston.

PEACOCK, D. P. S. & D. WILLIAMS1986. Amphorae in the Roman Economy: An Introductory Guide. Longman, London.

FENTRESS, E. & P. PERKINS 1988. 'Counting African Red Slip ware.' In A. Mastino (ed.), L'Africa romana. Atti del V convegno di studio Sassari, 11-13 dicembre 1987, pp. 205-214. Dipartimento di Storia dell'Universita di Sassari, Sassari.

POTTER, T. W. 1979. The Changing Landscape of South Etruria. Paul Elek, London. RICE, P. M. 1987. Pottery Analysis. A sourcebook. University of Chicago Press, Chicago.

FULFORD, M. & D. P. S. PEACOCK 1984. Excavations at Carthage: the British Mission, vol 1. 2. The Avenue Habib Bourguiba, Salambo. The pottery and other ceramic objects from the site. Department of Archaeology and Prehistory, Sheffield.

ROTTLANDER,R. C. A. 1966. 'Is Provincial-Roman pottery standardised?' Archaeometry 9: 76-91. RoTTLANDER, R. C. A. 1967. 'Standardisation of Roman Provincial pottery II: Function of the decorative collar on form drag. 38.' Archaeometry 10: 35-46.

HAMMOND,P. W. 1993. Food and Feast in Medieval England. Alan Sutton, Stroud, Gloucestershire. HAWTHORNE,J. 1996. 'Commensalism and common sense: a new approach to archaeological ceramics.' Assemblage 1: 3-6.

John Hawthorne Department of Archaeology Cniversity of Newcastle l\"ewcastle upon Tyne NEl 7RU C nited Kingdom [email protected]

24

4 The COMPASS method for the estimation of the capacity of pottery vessels Eugen S. Teodor National History Museum, Bucharest

4.1

Introduction

The 24th Annual Meeting of Computer Applications in Archaeology, held in 1996 in la§i (Romania), gave me the opportunity to understand the concern of English archaeologists for the volumetric capacity of vessels (see, for example, Hawthorne, this volume). The discussions that developed persuaded me that this parameter could not only bring a new dimension to demographic matters but could also be interesting as a theme of cultural comparison. It took only a step to note that a recently published study (Teodor 1996) offers a useful starting point for the estimation of volume. The COMPASS SYSTEM works with complete morphological data, so it can re-build the shape from published figures. It is also the theoretical support for a database of the post-Roman and Slavic migration period pottery for Eastern and Central Europe. The COMPASS SYSTEM deals with both complete pots or sherds, the latter being the main aim and the challenge. The principal idea is to enable us to compare the entire object with the part of it. This is why this system does not employ Bezier functions or other mathematical methods (as described in Orton et al. 1993, pp. 155-162) which can only deal with entire shapes. The fact is that recovered shapes are a very low percentage of recovered sherds. The number of complete posts does not allow us to compare the majority of Romanian sites from the morphological point of view. The very poor decoration for 5th to the 7th centuries AD must also be taken into account. Concluding, the COMPASS SYSTEM cannot use volume functions.

4.2

the data taken for morphological purposes could be used also for the volume estimation. The estimation takes as a starting point the comparison of a closed vessel with two truncated cones, which have for their base the plane of the maximum body width. The volume of each 'half' of the pot is calculated from the truncated cone formula. The results are corrected by taking account of the profile thickness (morphological measurements being taken on the exterior of the pot), the curve (arch) of the body (positive correction), and the curve (arch) of the neck (negative correction). The height cf (between neck diameter and the rim diameter, see Fig. 4.1) is not considered to make a significant contribution to the volume. The separate estimation of the upper and lower volumes offers the ability to compare complete vessels and half-preserved vessels (which are more numerous) when the morphological data are similar. Volumetric data could improve functional criteria, especially for cultures with a very low diversity on morphological and decorative characteristics such as those from the left bank of the Lower Danube at the beginning of Early Middle Ages. It is quite probable that two pots from the same morphological group, with dimensions in the same size class ('middle size pots'), have different functions if one has a capacity of 0.5 litres and another of 2.5 litres. One can see in Figure 4.1, the correction calculation for the upper and lower volume are not identical. This is because the compass system has been designed, from the very beginning, as an interrogative tool for Early Middle Age pottery. This kind of pot often has, on the lower part, a 'foot', not very well executed and not very marked. This is why I no longer take measurements for angles from G and H (Fig. 4.1), as I do for B and C. This argument is even more valid looking at the inside-shape, where the 'inflection point' (Shepard 1974, pp. 1, 26) is hard to see. For cultures with more complex shapes, one must consider the angles from points G and H for morphological criteria, and the differences between the angle at G point and t.i. (inferior tangent) for volumetric correction.

The COMPASS system

As published (Teodor 1996), this system does not work with absolute dimensions (except the mouth width), but with proportions, such as the neck diameter relative to the maximum body width, the upper height (see Fig. 4.1) relative to the total height, and so on. However, in actual research, the whole calculation has been transferred to computer, using for primary data the measured dimensions (in millimetres). This is how

25

F

f

D C

kt

at

bf

ct

U.C./

C

b

111=

~-

S (suspension)

average = gn

Truncated cone formula: 1'./

(R2

+ r 2 + rR)

Superior volume:

= (( (1.

0467x ( ( ( ( [af] - [bf]) (([EE']/

x [scale])

x

[AB] )+1) )+

( ( [be] x [scale]) x ( ( (1- ( ( ( [ts] -90)(90- [UB])) ( ( ( ( ( ( [a] /2)- [grs]) x [scale]) [scale]))+ ( ( ( ( [2C' o] /2)- [grs]) x [scale]) [scale])))+ ( ( ( ( [2C' o] /2)- [grs]) x [scale]) [scale])))) x 0.000001)

/100)) +1) /2)))) X x ( ( ( [a] /2) - [grs]) x x ( ( ( [2C' o] /2) - [grs]) x ( ( ( [a] /2) - [grs])

x

1r/3 I ab (height ab) correction of the body arch I Qf. (height be) correction of the neck arch R 2 (without the thickness of the body) r 2 (ditto)

r R (ditto)

x

mm 3 ➔ dm 3 (= litres)

Inferior volume:

= ((1.

0467x ( ( ( ( [kf] - [af] )- ( ( ( [gri] + [grs]) /2) +[SJ)) x [scale]) ( ( [J J'] / [AG] )+ 1))) x ( ( ( ( ( ( [a] /2)- [grs]) x [scale]) x ( ( ( [a] /2) - [grs]) x [scale]))+ (( (( [2H' o] /2)- (( [grs] + [gri]) /2)) x [scale]) x ( ( ( [2H' o] /2) - ( ( [grs] + [gri]) /2)) x [scale])))+ (( (( [2H' o] /2)- (( [grs] + [gri]) /2)) x [scale]) x ( ( ( [a] /2) - [grs]) x [scale]))) x 0.000001)

x

1r/3 I (without inferior thickness) correction of the body arch R 2 (without the thickness of the body) r 2 (ditto)

r R (ditto) mm 3 ➔ dm 3 (= litres)

Figure 4.1: Schematic of the COMPASS measurements. In black (top) can be seen those measurements that are involved in calculating the capacity. The meaning of af, bf etc. in the formul~ can be deduced from the diagram.

26

Admittedly, the CO:v!PASS method for calculating the volume of a pot needs to be calibrated on some examples calculated in AUTOCAD. In the short time at my disposal, this program was not available. Also, the cost of AUTOCAD is too high for a research program which involves thousands of pieces. If this method could work with an error less than 2%, it would be preferable, because it does not take extra effort, using data that most morphologists usually take from pots.

A. 0. 1974. Ceramics for Archaeologists. Carnegie Institution of Washington, Washington D.C., 8th edition.

SHEPARD,

E. S. 1996. Sistemul Campas. Studiu de morfologie analiticii numericii aplicat ceramicii uzuale din perioada de migrafie a slavilor. Muzeul National de lstorie a Romaniei, Bucure§ti. With extensive abstract in German.

TEODOR,

Eugen S. Teodor Muzeul National de Istorie a Romaniei Galea Victoriei 12 704122 Bucure§ti Romania

References C. R., P. TYERS & A. VINCE 1993. Pottery in Archaeology. Cambridge University Press, Cambridge.

ORTON,

27

5 Dating Stonehenge C. Bronk Ramsey 1 and A. Bayliss 2 1 Oxford

5.1

Radiocarbon Accelerator Unit; 2 English Heritage

Introduction

English Heritage recently funded a project to write up and publish the twentieth-century excavations at Stonehenge (Cleal et al. 1995). To determine the site's chronology the radiocarbon dates were assessed for reliability on scientific and archaeological grounds. Sixteen samples had been measured between 1950 and 1994, although the rejection of six of them left only ten which we considered reliable. Forty-eight new radiocarbon determinations were commissioned in 1994 and 1995, although four of them were also rejected as unreliable on archaeological grounds. Consequently the analyses presented here include only the remaining 54 measurements, 44 from the new programme and 10 from previous research. Details of the samples are provided in Table 5.1 on page 31. The intention of this paper is to provide an account of the processes of archaeological and mathematical reasoning which led to the published model of the site's chronology. Details of this model and the dating programme have been fully published elsewhere (Allen & Bayliss 1995, http://www. eng-h. gov. uk/stoneh). In particular we report some of the models which were produced as part of the analytical process and, although not regarded as the most realistic by the authors, still provided useful information. The inclusion in the analysis of two further measurements taken in the last few months, also leads to slight changes in the preferred model. In addition to providing an absolute chronology for the monument, the dating programme was designed to address specific questions-to elucidate the sequence of major events and sub-phases where there is no recorded stratigraphic information, to assign specific features or groups of features to a phase by radiocarbon dating, and to estimate the duration of the phases. To achieve these objectives, we constructed a series of mathematical models of the chronology of the site which incorporate both the radiocarbon measurements and other, purely archaeological, information such as stratigraphy. These models include our archaeological interpretations of the relationships between the radiocarbon samples, their contexts, and the activities of prehistoric people which we are attempting to date. Potentially there are considerable dangers in this, since these relationships are complex (Reece 1994). How-

ever we can use the models as analytical tools to test whether certain interpretations are consistent with the radiocarbon evidence. A number of different models are possible for the site. Although these will certainly change when new evidence comes to light, such modelling does give a more accurate view of the chronology of the monument than that provided simply by calibrating the radiocarbon measurements in isolation. The first stage in our analysis was to relate the results to the calendar timescale by calibration. This was done by the usual probability method (Dehling & van der Plicht 1993; Stuiver & Reimer 1993; van der Plicht 1993) using data from Pearson et al. (1993, 1986); Pearson & Stuiver (1986); Stuiver & Pearson (1986), and Kromer & Becker (1993). Implicit in this method of calibration is the assumption that we have no other information about the date of the sample. To incorporate the information from site stratigraphy and archaeological interpretation we used methods based on Gibbs sampling techniques (Buck et al. 1992; Gelfand & Smith 1990). The methods have been applied using the program Ox CAL (v2.17) (Bronk Ramsey 1995, http: //sable. ox. ac. uk/rlaha), which was written specifically for this sort of analysis and is based largely on the original mathematical work of Buck et al. (1994a, 1991, 1994b, 1992). In addition to Gibbs sampling, statistical tests are included which check if the model is consistent with the dating evidence. These tests, indices of agreement, were devised specifically for the program, the threshold for acceptance being similar to the 5% x2 test. A test fails if the Gibbs sampler is forced to choose dates from very low parts of a probability distribution. In the diagrams which follow (Figs. 5.2, 5.3, 5.6, 5. 7, and 5.8) the constrained probability distributions, calculated using the stratigraphic information and archaeological interpretations in addition to the radiocarbon results, are shown in solid black; the original unconstrained distributions are shown in outline. This enables the reader to judge the effect of the mathematical modelling. All ranges derived from these constrained distributions are cited in italics, to distinguish them from simple calibrated date ranges.

29

Figure 5.1: Location of mesolithic features under the car park.

5.2

Mesolithic activity

be of widely differing ages, we cannot be absolutely sure that it is all of exactly the same age. The antlers could certainly have grown in several different seasons (see Bayliss et al. 1997 forthcoming for further discussion of this point). Instead, if all the material from the base of the ditch beneath the primary silt is assumed to be randomly selected from a uniformly deposited phase, the end of this phase can be taken as an estimate of the date of the digging of the ditch. The advantage of the assumption of a uniformly deposited phase is that we can then estimate the end of that phase, not just the last dated event (which will be somewhat earlier). In fact because this estimate is so well constrained (see below) the estimates provided by the two methodologies are very similar. We know that the primary silt starts to form almost as soon as a ditch is dug (Bell et al. 1996), and so the last of this material must have been collected very soon after the ditch was completed. The four dated animal bone deposits from ditch terminals were significantly earlier than the antlers. Indeed, if the results are modelled to include the constraint that the structured deposits are later than the digging of the ditch, the model is statistically significantly inconsistent (A= 30.2%; Fig. 5.3), even though they must have been placed in the ditch after it was dug! The samples date to when the animal died however, not to when the bones were put on the base of the ditch, so this can be explained by the curation of the material for some time before it was deposited. Analysis of the information currently available suggests that the period of curation was for between 70 and 420 years (95% confidence). Since there was no material suitable for dating from the primary fill or the activity on top of this, the date of the digging of the ditch can only be constrained by material from the secondary fills. These secondary fills certainly accumulated after the silting beneath them, but the crucial question is whether the material dated from these layers is residual. To minimise the problem we chose to submit a relatively large number of samples throughout the profile. It was hoped at least some of these would not be

The earliest activity so far identified at Stonehenge is a series of pits beneath the present day car park (Allen & Bayliss 1995, pp. 43-7; Fig. 5.1). They have produced five radiocarbon dates from pine charcoal, all of mesolithic age (Table 5.1). The function and significance of these features is obscure, although it is possible that they held upright posts. If the five dates are assumed to be randomly selected from a uniformly deposited phase, the span of dated events is estimated to be between 300 and 1600 years, with the events occurring between 8500-7650 cal BC and 7500-6700 cal BC (Fig. 5.2). This assumption weights a short phase more strongly, and, given the small number of results, does not produce useful estimates for the start and end of the phase. However it does indicate the longevity of the mesolithic activity.

5.3

Phase 1

This consists of the construction and initial use of the first monument, a segmented ditch with a bank and counterscarp bank, and the Aubrey holes as a ring of posts (Cleal et al. 1995, pp. 63-114; Fig. 5.4a). The only material recovered for dating which could be identified in the archive came from the base of the ditch. The potential samples included over 100 antlers and a number of animal bones. All of these were placed on the base of the ditch before any primary silt had accumulated. The antlers have been modified into tools, both picks and rakes, with many displaying clear signs of wear (Sergeantson 1995, pp. 414-28). For these reasons it seems likely that they were used for the digging of the main ditch and then immediately deposited on its base. Since antler tools cannot be kept for a substantial time before use because they become brittle, the date of the antlers should be a good estimate of the date of the ditch digging. Even so we did not choose to take a weighted mean of the results before calibration (Ward & Wilson 1978). This is because, although the material cannot

30

Context

Material

Laboratory

Postpit Postpit Postpit Postpit Postpit

Mesolithic WA9580 WA9580 WA9580 A B

Pinus Pinus Pinus Pinus Pinus

OxA-4919 OxA-4920 GU-5109 HAR-455 HAR-456

8520±80 8400±100 8880±120 9130±180 8090±140

7700-7420 7580-7090 8090-7580 8820-7730 7480-6590

cal cal cal cal cal

BC

Pre-phase 1 Sarsen Circle

Animal bone

OxA-4902

5350±80

4360-3990 cal

BC

Phase 1 Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch

Antler Antler Antler Antler Antler Antler Antler Antler Antler Animal Animal Animal Animal

UB-3787 UB-3788 UB-3789 UB-3790 UB-3792 UB-3793 UB-3794 BM-1583 BM-1617 OxA-4833 OxA-4834 OxA-4835 OxA-4842

4375±19 4381±18 4330±18 4367±18 4365±18 4393±18 4432±22 4410±60 4390±60 4550±60 4460±45 4455±40 4520±100

3085-2920 3095-2920 3030-2910 3040-2915 3040-2915 3095-2920 3305-2925 3340-2910 3330-2910 3500-3040 3350-2920 3340-2920 3510-2920

cal cal cal cal cal cal cal cal cal cal cal cal cal

BC

Charcoal

C-602

3798±275

3020-1520 cal

BC

Animal bone Animal bone Animal bone Animal bone Animal bone Bone chisel Antler Antler Animal bone (articulated) Animal bone (articulated)

OxA-4841 OxA-4843 OxA-4880 OxA-4881 OxA-4882 OxA-4883 OxA-4904 UB-3791 OxA-5981

4295±60 4315±60 3875±55 4300±60 4270±65 4300±70 4365±55 4397±18 4220±35

3040-2700 3100-2700 2560-2140 3080-2700 3040-2660 3100-2700 3300-2900 3095-2920 2920-2660

cal cal cal cal cal cal cal cal cal

BC

OxA-5982

4405±30

3300-2920 cal

BC

Phase 3 Sarsen Circle Sarsen Trilithon Sarsen Trilithon Sarsen Trilithon Bluestone Circle Bluestone Circle Bluestone Horseshoe Stonehole E Stonehole E Z Hole 29 Y Hole 30 Y Hole 30 Y Hole 30 'Beaker' burial 'Beaker' burial 'Beaker' burial 'Beaker' burial 'Beaker' burial

Antler Antler Antler Antler Animal Antler Antler Antler Antler Antler Antler Antler Antler Human Human Human Human Human

4023±21 3860±40 3985±45 3670±150 3740±40 3865±50 3695±55 3995±60 3885±40 3540±45 3341±22 3300±19 3449±24

2655-2485 2470-2200 2850-2400 2480-1680 2290-2030 2480-2140 2280-1940 2860-2350 2490-2200 2030-17 40 1735-1530 1675-1520 1880-1690

cal cal cal cal cal cal cal cal cal cal cal cal cal

BC

bone bone bone bone bone

UB-3821 OxA-4839 OxA-4840 BM-46 OxA-4878 OxA-4900 OxA-4877 OxA-4837 OxA-4838 OxA-4836 UB-3822 UB-3823 UB-3824 BM-1582 OxA-4886 OxA-5044 OxA-5045 OxA-5046

3817±27t

2460-2140 cal

BC

Avenue Stonehenge terminal Stonehenge terminal N r Avon terminal N side of A344

Antler Antler Animal bone Antler

OxA-4884 BM-1164 OxA-4905 HAR-2013

3935±50 3678±68 3865±40 3720±70

2580-2300 2290-1890 2470-2200 2350-1930

BC

Human bone Bone point

UB-3820 OxA-4885

2468±27 2840±60

775-410 cal BC 1260-840 cal BC

charcoal charcoal charcoal charcoal charcoal

bone bone bone bone

Reference

Radiocarbon

Age

(BP)

Calibrated date range (95% confidence)

BC BC BC BC

BC BC BC BC BC BC BC BC BC BC BC BC

Phase 1/2 Aubrey Hole 32

Phase 2 Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch Ditch

bone

cal cal cal cal

BC BC BC BC BC BC BC BC

BC BC BC BC BC BC BC BC BC BC BC BC

BC BC BC

Post-monument Palisade Ditch Sarsen Circle

Table

5.1:

3825±60BP,

Summary of reliable radiocarbon 3775±55BP, and 3715±70BP.

dates

from Stonehenge.

31

tweighted

mean

of 3960±60BP,

3785±70BP,

SEQ {A= 34.6%(A'c= 60.0%)) SEQ

..A...___

BOUND @end structured deposits PHASE structured deposits 6.7%

-~.)4

cos2 rptgrp+ --N cos4 2 24 rptgrp(5- tg2rp + {)T/2 + 4ri4 )+

(~>.)5 6 72() 1\i cos rptgrp+ T

(61 - 58tg2rp+ tg4rp + 270ri2

-

330tg2rpri2 )

• Original coordinates: X = 690.600 Y = 4.227.700

Zone= 29 • Transformed coordinates: X = 164.987 Y = 4.232.378

Where

Zone= 30 • Formulae (Reduction Polynomials):

N=

Y

a

(1-

= Y1c+ nC

X = x

e2 sen 2 rp)½

- eD

+ 500.000 = X1c+ nD + eC + 500.000

n = (Y - Yc)l/10

rie' cos rp and rp and >. are the latitude and longitude of the point. As can be seen, the transformation of the coordinates depends on the parameters of the ellipsoid taken as the reference by the projection. These parameters are, first, the longest axis of the ellipsoid (a), second the first and second eccentricity of the meridian ellipsoid (e,e'), and third the latitude of the point (rp). The partial underestimation of the importance of these procedures for coordinate conversion could explain some recent confusion concerning the location of old sites. These sites were originally located on maps that used the Struve ellipsoid, so that in order to locate them on modern maps using the Hayford ellipsoid it is necessary to transform the geodesic coordinates from one ellipsoid to the other. However, tables are available to perform this conversion (Rossignoli 1976), and in fact it can be also automatically be performed by a GIS like ARC-INFO. Finally, the third cartographic problem that was detected before the archaeological coverages could be generated was again related to the coordinate system. Since the region of Andalucia is divided into two different UTM zones (29 and 30), and, as it is well known, every geographic zone is a different plane with its own reference system (Fig. 13.2b), the coordinates corresponding to one of the zones had to be transformed

5; e

= (x - Xc)l/10

5

where YcYXc are the coordinates of the multiples of the closest 100km to the given coordinates.

A= b + nc - ec'

= b' + nc' + ec C = a+nA-eB D = a' +nB +eA B

where Y1c,X1c,a,a',b,b'c,c' are the tabulated coefficients (Rossignoli 1976, p. 178). Once the preliminary cartographic problems were detected and fixed, the information contained in a series of DBF tables was exported to ARC-INFO version 7.03 as two ASCIIfiles, the first one storing each site's unique ID number and pair of UTM coordinates (imported from the Arc module) and the other containing the unique ID number plus the associated variables (imported from the Info module). Thus, two coverages with archaeological sites were generated, one with 54 7 sites recorded as points and another with 111 sites recorded as polygons. The point coverage was further sub-divided into coverages based on chronology (Neolithic, Copper Age, Bronze Age, Iron Age and Roman), so that their distribution patterns can be examined individually in the future.

106

13.2.3

Data visualization.

General). The non-inclusion or incomplete inclusion of the record of archaeological sites in the local landplanning documentation has also been noted as a potential deterioration factor (lack of control at the lowest administrative level) of the archaeological record, as local councils are legally compelled to participate in the control and protection of archaeological sites and monuments (Tejedor et al. 1994). Figure 13.7 shows those municipalities where a number of sites have been converted into polygons. One more map of the area has been produced showing the degree of inbalance between the number of sites recorded as polygons (i.e., recorded with a cartographic precision of lm) and the number of sites recorded as points (i.e., recorded with a margin of error of 100m). As can be seen in Figure 13.8 only in 10 out of 44 municipalities have more than 50% of sites been recorded as polygons on 1:10,000 maps. Another map (Fig. 13.9) displays the percentage of new sites identified after a systematic survey had been carried out within the limits of some of the local administrative units. This map can be used as an indicator of the expected frequency of sites in those municipalities where the survey background is low. As far as preservation of the archaeological record is concerned, current land use is undoubtedly an essential variable. As can be observed in Figures 13.1013.11, one the main types of land use currently observed within the southern part of the area under study is industrial re-afforestation. The systematic plantation during the 1960s of an alien species such as eucalyptus within this area was accompanied by massive terracing of hills which had a devastating effect on the archaeological record. In fact it is hard to find any significant concentration of archaeological sites in any of the polygons representing this type of land cover (Fig. 13.11), even in those areas that have been subject to systematic surface survey. Finally, more predictive maps have also been produced with ARC-INFO. Figure 13.12 shows a buffering area of 500m in radius around all the main roads within the study area. As is well known, a typical case for rescue excavation is massive earth removal due to large-scale road or railway works. In this case, and with the available archaeological information, a map has been produced displaying the list of sites threatened by any road widening within the study area. Similarly, the existence of important mineral resources (Fig. 13.13) in the Sierra de Huelva has become a major threat to the archaeological record, due to the great impact that these economic activities have. Since studies are frequently published reporting the impact of new mining initiatives on the archaeological heritage, the use of this kind of buffering map can be of great help, saving time and reducing costs in the preliminary stage of documentation. In general, GIS mapping of the archaeological information produces a quick and intuitive assessment of those gaps where the lack of information points out

The GIS mapping of the inventory of archaeological sites of the Sierra de Huelva begins with the observation of the degree of correlation existing between the intensity of surface exploration and the number of locations actually recorded. A basic classification in four categories has been made of every municipality ( municipio) according to the intensity of the archaeological surveys carried out within its administrative limits: 1. Unsurveyed: all the archaeological sites recorded

were randomly detected. 2. Unsystematically surveyed: some kind of survey was carried out in the past, but it either did not produce a systematic list of archaeological locations (they were for example oriented to specific site types in chronological or functional terms) or it was carried out under non-explicit methodological assumptions. 3. Systematically surveyed: (a) The methodology and scope of the survey was explicitly declared and the inventory of sites included all chronological and functional site types. (b) The same as 3a but in this case the survey was carried out within the context of a systematic research project involving direct intra-site fieldwork (excavations, sondages,

etc.) As can be observed in Figures 13.4-13.5 (and as expected) there is a high degree of correlation between the intensity of survey and the number of sites recorded. This map can be assessed in two different ways. On the one hand, the three areas where the scarcity of archaeological sites is clearly associated with low levels of surface exploration, suggest that more attention should be paid to them in the future: in this case, the lack of information becomes in itself a parameter of risk for the archaeological record (Fig. 13.4). On the other hand, those areas with the highest concentration of sites become the only adequate areas for spatial analysis: any attempt to include within the same spatial analysis areas with low and high survey backgrounds would inevitably embody a strong bias (Fig. 13.5). Another variable of interest, from the perspective of the administrative documentation of the schedule of archaeological sites, is the type of land categories developed by each local (municipal) council. The spatial distribution of this variable (Fig. 13.6) in the Sierra de Huelva suggests a wide predominance of municipalities without specific land-use ordinances (Sin Planeamiento), while no one has yet developed the highest level of administrative land planning (Plan 107

Figure 13.4: Intensity of survey: Administrative units.

Figure 13.5: Intensity of survey: Adequate areas for spatial analysis.

the need for future investigation. From the perspective of the management of inventories of archaeological sites, this approach is of great help in order to visually detect gaps in the information and in order to generate fast spatially-referenced queries of the databases. From the point of view of the management of large rural territories, the conclusion drawn from this initial experience is that traditional non-computing archaeological planning and data processing can become more efficient and affordable with the support of a GIS.

13.3 13.3.1

one of the largest historic centers of Western Europe (250ha) offering today a wide diversity of historical places and monuments. Aware of the need for a more centralised and proactive protection of the archaeological record of the city, the regional administration of culture decided in 1994 to establish a permanent research group whose main purpose would be the control and coordination of the frequent archaeological excavations carried out within the limits of the historic city. As was soon realised, one of the main methodological tasks of this research group would be the production of an efficient cartographic base, supported with a computer system, suitable for fast updating and easy retrieval of the large amount of archaeological information constantly produced in a rapidly changing physical environment. This archaeological computer mapping of the historic center of Sevilla has two main purposes:

The Historic Centre of Sevilla Data sources

The second case under examination in this paper has its own methodological peculiarities. First inhabited during the Iron Age (7th century BC according to the archaeological evidence) the modern city of Sevilla has

Cognitive: It must provide a cartographic basis for the interpretation of the city as a single archae-

108

□ s1 N PLANEAMI ENTO □ DELI Ml TACI ON SUE LO

!I NOR

MAS

■ PLAN

SUBSI

DI

ARI

URBANO

AS

GENERAL

Figure 13.6: land-planning.

Municipal

Figure 13. 7: Municipalities covered by redefinition of archaeological sites as polygons.

ological site, facilitating the design and testing of hypotheses relative to the urban development of the settlement during different historic periods. At the same time, it is expected to provide a database containing variables such as depth of archaeological sediments, freatic level, deterioration of the archaeological stratigraphy, record of archaeological excavations carried out, type of construction above and under the ground, etc.

Again, a GIS like ARC- INFO provides an adequate set of tools to achieve the above, as urban archaeological information can be, first visually stored, queried and retrieved, second, easily updated, and third quantitatively analysed, as the program provides a module (Grid) for the quantitative assessment of raster information.

13.3.2

Evaluation: Also, it is expected to provide a dynamic and interactive basis for the evaluation and assessment of these different parameters in order to generate a more adequate strategy for the archaeology. This basically involves the elaboration of a series of archaeological maps ( Carta de Riesgo) appropriate for constant updating: maps showing loss of the archaeological record, maps of conservation areas, maps of accessibility of archaeological stratigraphy, etc.

Geographic and Archaeological data input

In this case no pre-existing digital maps at an adequate scale were available to map the archaeological information; it was necessary to create these ourselves. This process involved the following steps: 1. A series of raster images of the 43 sheets of

the cadastre map covering the historic centre of Sevilla were produced by means of a scanner.

109

D

%

< =5

0

>50

=1 00

%