Trust and Community on the Internet: Opportunities and Restrictions for Online Cooperation [Reprint 2016 ed.] 9783110508666, 9783828203013

Citation preview

Trust and Community on the Internet Opportunities and Restrictions for Online Cooperation

Guest-Editors: Bernd Lahno / Uwe Matzat

Lucius & Lucius Stuttgart 2004

Bibliografische Information der Deutschen Bibliothek Die Deutsche Bibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.ddb.de abrufbar

ISBN 3-8282-0301-9 (Lucius & Lucius)

© Lucius & Lucius Verlagsgesellschaft mbH Stuttgart 2004 Gerokstr. 51, D-70184 Stuttgart www.luciusverlag.com

Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigung, Übersetzungen, Mikroverfilmungen und die Einspeicherung, Verarbeitung und Übermittlung in elektronischen Systemen.

Druck und Einband: Rosch-Buch, Scheßlitz Printed in Germany

Contents Editorial

l

1. Concepts and Backgrounds Henk de Vos: Community and Human Social Nature in Contemporary Society

7

Bernd Lahno: Three Aspects of Interpersonal Trust

30

Hans- Werner Bierhoff/Bernd Vornefeld: The Social Psychology of Trust with Applications in the Internet

48

Uwe Matzat: Cooperation and Community on the Internet: Past Issues and Present Perspectives for Theoretical-Empirical Internet Research

63

2. The Internet as an Environment for Trust Victoria McGeer. Developing Trust on the Internet

91

Philip Pettit: Trust, Reliance and the Internet

108

Russell Hardin: Internet Capital

122

Geoffrey Brennan/Philip

139

Pettit: Esteem, Identifiability and the Internet

3. Reputation and Online Auctions Chris Snijders/Richard Zijdeman: Reputation and Internet Auctions: eBay and Beyond

158

Gary E. Bolton/Elena Katok/Axel Traders

185

Ockenfels: Trust among Internet

Werner Güth/Hartmut Kliemt: The Evolution of Trust(worthiness) in the Net

203

4. Groups and Networks Coye Cheshire/Karen S. Cook The Emergence of Trust Networks under Uncertainty. Implications for Internet Interactions

220

Anabel Quan-Haase/Barry Networked Organization

241

Wellman: Local Virtuality in a High-Tech

Andreas Flache: How May Virtual Communication Shape Cooperation in a Work Team?

258

Margit Osterloh/Sandra Rota: Trust and Community in Open Source Software Production

279

5. Law and the Internet Eric Hilgendorf. Crime, Law and the Internet

302

Geoffrey Brennan/Rodgers & Hammerstein: Internet-Hymn

313

Authors

314

Editorial

Early studies in the area of Internet Research emphasized the deficiencies that computer-mediated communication as opposed to face-to-face communication would have. The chances for the evolution of cooperative relationships on the Internet were assessed sceptically. Present research findings correct this point of view. In spite of a missing central authority, without formal controls and sanctions, with anonymity and easy-to-use exit options there is not only chaos and anomy on the Internet. Rather, there is a surprisingly large amount of cooperation and social order. There is a successful exchange of individual goods, public goods are produced, and there are stable social networks, reliable communities and effective social norms. If people interact online under certain conditions, there seem to be incentives that stimulate cooperation and social exchange. Additionally, under favourable conditions the Internet can be used for the establishing of cooperative relationships that would hardly be possible in the case of pure face-to-face communication. Such positive effects, however, are not a technical automatism. The failure of many online communities, as well as the discussion about the social consequences of the Internet for the social integration of its users is a warning against nave optimism. In this volume the potential, the pre-conditions and the limits of the Internet for the emergence of trust and community building are discussed. Trust and community are social phenomena which are of utmost importance for social relationships and social order. The analysis of the Internet as a trust-building medium focuses on the conditions under which the Internet promotes social exchange and cooperation between individuals and between groups. The analysis of the Internet as an opportunity for community building focuses on the conditions under which it can be used for comprehensive forms of collective action and group living. Both questions are interrelated. Trust is a pre-condition for community building. At the same time, social bonds within a community facilitate the placement of trust under risky conditions. Trust is a fundamental element of cooperative relations. It can be understood as the willingness to make one's own well-being to some degree dependent on other people or institutions. The advantages of mutual cooperation can often only be gained when individuals are willing to take risks and to expose themselves to the danger of being disadvantaged if others do not cooperate. This means: in such situations cooperation is possible only if trust is placed. In spite of the risks, individuals cooperate with each other on the Internet to a significant degree, as the examples on online markets and auctions show. These examples suggest that there is trust to a significant degree. Since trust-building mechanisms are an important foundation for different forms of cooperation, knowledge about the opportunities

Editorial

2

and limitations for building trust on the Internet is crucial for assessing the possibilities for online cooperation and collaboration in general. One important question to be analyzed is in which parts of the Internet does cooperation take place and through which mechanisms is trust established and secured. It should be examined whether and to what extent the Internet as a technical medium contributes to an extension of trusting relationships and interaction. Which technical, organizational, social, and legal conditions further the extension of trusting interaction and in which way do these conditions depend on and contribute to a satisfying short- or long-term interaction between the participants? Conditions that are conducive to the placement of trust can be • direct or indirect exchange relations between participants: reciprocity; • trustworthiness or reputation of the actors; • one-sided prior contributions that initiate cooperation: gift-relations; • common norms; • communality through common goals or values; • affective bonds between individuals. The general limits to the emergence of trust-building conditions will be the limits to the placement of trust. In as far as such conditions prevail for online interaction, the Internet is a trust-building medium. If such conditions can be realized on the Internet for individuals or groups that could not build up trusting relations outside of the Internet—for example because of transaction costs being too high—, then the Internet is a trust-extending medium. Communities are characterized by a special feeling of solidarity between their members. They consist of groups of people who traditionally interact in locally restricted environments and who develop durable social relationships. Communities provide opportunities for individual and collective action with far-reaching consequences for society. But usually membership in a community does not only have an instrumental value for the pursuit of the common goals of its members. Additionally, membership has a highly affective value. Communities satisfy fundamental human needs, such as group attachment, identification, security, solidarity, and mutual understanding. They give individuals the feeling of being esteemed, and they are an important source of trust and thus provide for the stabilization of cooperation since they transmit common values and promote affective bonds. Communities have a pivotal function in integrating society and also produce social capital. The increasing mobility of modern societies contributes to the diminishing of traditional communities that are based on personal exchanges and the geographical closeness of their members. At the same time, mobility contributes to the popularity of communication media such as telephone and electronic mail that allow for interpersonal communication over long distances.

Editorial

3

It is an important question whether and to what extent communities exist on the Internet and what their emergence and continuation depend on. In this context, it is of special interest to analyze whether the Internet provides opportunities for the development of new communities that are as valuable for individuals as the traditional communities whose existence is endangered. An additional issue to be discussed is about the quality of online relationships and online groups. Under which conditions can relationships that were developed on the Internet acquire a satisfying quality for individuals and groups? The task of Internet research is to examine in how far conditions can be realized in online groups that are known to be community-advancing in general. Such conditions are • common goals and values; • special relational interests of the group members; • a high degree of interdependence among the members; • multiplexity of the relationships; • durability of the relationships; • easy accessibility of the members to each other. Traditional communities ensure the existence of (some of) these conditions usually by their local embeddedness. Since such a local anchoring of communities is not the rule for online groups, the question emerges as to the other mechanisms which help the Internet to promote the existence of favourable conditions for community development. Social, economical, psychological, legal and technical conditions have to be taken into account to examine the potential as well as the limitations of the Internet for the development of trust and communities. The evolution of the Internet is influenced by a variety of factors which suggests that a discussion cutting across the usual disciplinary borders is necessary and useful. This conviction was the background for an international and interdisciplinary conference on Trust and Community on the Internet. Opportunities and Restrictions for Online Cooperation. The conference took place from 31 July to 2 August 2003 at the Centre for Interdisciplinary Research (ZiF) at the University of Bielefeld and was organized by Michael Baurmann, Bernd Lahno and Uwe Matzat. The meeting was financed by the ZiF and the Deutsche Forschungsgemeinschaft (DFG). The organizers greatly appreciate this support and express their thanks to both institutions. The articles in this volume are all based on papers which were presented at the conference. The first group of articles consider Concepts and Backgrounds. They discuss general theoretical models and empirical analyses of conditions for the development of trust and communities and examine whether such conditions can be found in certain parts of the Internet or to what extent they can be created. Henk de Vos explicates the fundamental sociological concept of community. He points out that

4

Editorial

communal living has deep connections with human social nature and is strongly associated with well-being and health. Structural conditions for the emergence and maintenance of communities are specified and it is discussed in which way the Internet affects these conditions. Bernd Lahno differentiates between a behavioral, a cognitive, and an affective dimension of trust. He argues that affective aspects of trust must be included in any adequate account of the role of trust in social dilemma situations involving multiple equilibria. The design of institutions that can foster trustful cooperation in the context of the Internet is analyzed under this perspective. The article by Hans- Werner Bierhoff and Bernd Vornefeld deals with the social psychology of trust. Three levels of trust are delineated: trust in a specific person, trust in people in general and trust in abstract systems. The authors assume that, in the case of the Internet, trust in the Internet as a system is of crucial importance. Several possibilities for strengthening the trustworthiness of the Internet as a system are presented. Uwe Matzat gives an overview of the present state of theoretical-empirical Internet research. He emphasizes the fact that recent insights suggest that the Internet should no longer be regarded as a constant that has uniform effects but that the social consequences of Internet use depend on a number of contextual conditions. Some hypotheses as to which conditions and features of online groups could facilitate bilateral or group-level cooperation are discussed. The second section of articles focuses on The Internet as an Environment for Trust. Victoria McGeer argues that the Internet provides conditions in which rational individuals can in principle initiate and maintain relationships of interpersonal trust. But she underlines the importance of developing mature capacities for trust in an Internet context since immature trusters are particularly vulnerable to the liablities of Internet trust. Philip Pettit expresses his fundamental scepticism of the idea that trust between people can be formed on the basis of Internet contact alone. According to his view, a necessary precondition for the development of trust in a rich sense of thé term is not given when people are related only via the Internet and their personal identity remains invisible to others. Russell Hardin analyzes the Internet as a form of social capital that is not reducible in its characteristics to other forms of social capital: it enables us to do many things with radically greater efficiency than is possible without it. But Hardin suggests that relationships on the Internet are too thin to back trust and cooperation among those who do not have relationships off-line with each other. In their joint contribution, Geoffrey Brennan and Philip Pettit deal with the role of esteem in Internet relations. They argue that the desire for esteem not only serves an important function in ordinary social life but that Internet relations are also susceptible to esteem-related incentives. For them there is every reason to believe that a good e-reputation is an object of desire for real agents and therefore an important driving force in disciplining interactions on the Internet and supporting the operation of social norms. The next group of articles under the heading Reputation and Online Auctions analyze empirically the mechanisms that further the building of trust on the Internet especially in the field of economic transactions. Chris Snijders and Richard

Editorial

5

Zijedeman investigate the phenomenon that transactions in online auctions such as eBay and Ricardo seem to work relatively smoothly. In an attempt to replicate the results of recent research on online auctions, they study the conditions under which eBay's reputation mechanisms can prevent opportunistic behaviour. Gary E. Bolton, Elena Katok and Axel Ockenfels report on experiments suggesting that it is the interaction of social preferences and cleverly designed reputation mechanisms that solve the trust problem on Internet market platforms to a large extent. The experiments show that the seller's intrinsic motivation to be trustworthy is not sufficient to sustain trade when not complemented by a feedback system. Werner Guth and Hartmut Kliemt compare traits of an organizational design to promote trustworthiness as suggested by economic reasoning with those that have actually emerged on the Internet. They ask whether institutions like eBay will increasingly have to 'economize on virtue' although they have so far been able to rely on its spontaneous provision. The subject of the next four articles is Groups and Networks. Coye Cheshire and Karen S. Cook comment on the application of experimental sociological research to different types of computer-mediated social interactions, with particular attention to the emergence of 'trust networks'. They develop a classification system that helps to integrate the existing research on trust in experimental social psychology with the field of computer-mediated exchange. Anabel Quan-Haase and Barry Wellman state that the role of groups in networked organizations has remained unclear and that little is known about how computer-mediated communication is used to bridge group and organizational boundaries. They examine how employees of a high-tech company communicate with members of the work group, other colleagues in the organization, and colleagues outside the organization to understand their boundary-spanning communications better. The paper by Andreas Flache addresses the question of how virtual communication may affect cooperation in work teams theoretically. A formal social exchange model is used and the degree of team virtualization, i.e. the extent to which interaction between team members occurs online, is related to parameters of the exchange. Simulation results suggest both positive and negative effects of team virtualization on work-cooperation. Margit Osterloh and Sandra Roth investigate under which conditions open source models—such as the open source software production— may be successful in general. They show that a complex interplay of situational, motivational, and institutional factors have to be taken into account. It is argued that the success of the open source model is greatly facilitated by well-balanced intrinsic and extrinsic motivation and low costs for contributions and governance mechanisms. The last article in this volume by Eric Hilgendorf deals with Law and the Internet. The main forms of datanet crime on the Internet are described and some of the most important Internet-cases of the last decade are discussed. As one of the main problems of datanet crime is its global scope, a transnational criminal law for the Internet seems to be desirable and possible. But Hilgendorf points out that the problems of computer-related crime on the Internet cannot be solved by criminal law alone.

6

Editorial

The volume is completed by the Internet-hymn. The lyrics of the hymn were written by Geoffrey Brennan to the tune of the famous song Edelweiss by Richard Rodgers and Oscar Hammerstein II from the musical The Sound of Music. The Internet-hymn was performed for the first (and, so far, last) time at the Centre of Interdisciplinary Research at the final session of the conference. Geoffrey Brennan as the solo-singer was strongly supported by the chorus of the conference participants. Finally we wish to express our personal thanks to the staff of the ZiF for their exceptionally friendly and professional support, especially to Marina Hoffmann, Daniela Mietz and Trixi Valentin who were in charge of the organization of the conference and the well being of the participants. Not least because of their invaluable assistance the conference was a pleasure for all. Michael Baurmann, Bernd Lahno, Uwe Matzat

Analyse & Kritik 26/2004 (© Lucius & Lucius, Stuttgart) p. 7-29

Henk de Vos

Community and Human Social Nature in Contemporary Society*

Abstract: Although community is a core sociological concept, its meaning is often left vague. In this article it is pointed out that it is a social form that has deep connections with human social nature. Human social life and human social history can be seen as unflagging struggles between two contradictory behavioral modes: reciprocity and status competition. Relative to hunter-gatherer societies, present society is a social environment that strongly seduces to engage in status competition. But at the same time evidence increases that communal living is strongly associated with well being and health. A large part of human behavior and of societal processes are individual and collective expressions of on the one hand succumbing to the seductions of status competition and one the other hand attempts to build and maintain community. In this article some contemporary examples of community maintaining, enrichment and building are discussed. The article concludes with a specification of structural conditions for community living and a short overview of ways in which the Internet affects these conditions.

0. Introduction Community is brought back in social scientific discourse. Social scientists and even economists increasingly begin to warn against the dwindling of community with all kinds of negative consequences (Etzioni, P u t n a m , Fukyama, Lane, Frank, Easterlin, Layard). Often it is unclear what their messages precisely are. Community is a widely used concept, but it is rarely precisely defined. Also, while some take for granted that community is a positive thing, others associate the concept with backwardness and conformity. Neither of these parties shows much interest in a precise specification of the advantages and disadvantages of community. Also not much attention is paid to how community relates to human social nature and to the structural conditions for viable communities. This article starts with a short overview of the development of the idea of community in twentieth century sociology (Section 1). In Section 2 community is defined as a group of people who are connected to each other by way of reciprocity relations. Given that community provides well being and health benefits, it should be expected that people are motivated to maintain and enhance community living, and that these efforts are more or less successful. Therefore in " T h a n k s to Uwe Matzat and Michael Baurmann for comments on earlier versions of this article. Work on this article was partly supported by a grant from K P N Research. Opinions expressed are solely those of Henk de Vos.

8

Henk de Vos

Section 3 some examples of contemporary forms of community are discussed. The article concludes with a specification of the structural conditions for community (Section 4), a discussion of how the Internet affects these conditions (Section 5) and with a general conclusion and discussion (Section 6).

1. The Concept Of Community In Twentieth Century Sociology Community was an important concept in the work of both Durkheim and Weber. Durkheim described the old forms of community as the small, homogeneous settlements of the agrarian society and the bands or clans of 'primitive' society. He saw these forms of group living as based on a social bond (mechanical solidarity) that resulted from: in-group similarity of traditions, beliefs, skills, and activities; between-group dissimilarities (segmentation), and positive sentiments towards in-group members (altruism) and negative sentiments towards others. With the rise of the market, mechanical solidarity could not be preserved, because segmentation strongly decreased. Durkheim first counted on the market to produce a network of interests (of mutual obligations, of rights and duties) that would function as a substitute for the community of the past. At first he was quite positive about this new so-called organic solidarity. But he also stressed that social bonds could not be based on egoism alone: they are too superficial and only create external ties. In fact, he believed that men have a need for being interdependent with others: "Men cannot live together without acknowledging, and, consequently, making mutual sacrifices, without tying themselves to one another with strong, durable bonds. ... Because the individual is not sufficient unto himself, it is from society that he receives everything necessary to him, as it is for society that he works. Thus is formed a very strong sentiment of the state of dependence in which he finds himself." (Durkheim 1964, 228) Therefore he expected that occupational organizations would evolve into the required new forms of community, because they would be based on new similarities and interdependencies (expertise, norms). Next to that, it should be the task of schools to inculcate similarities and a sense of interdependency on the societal level. In short, Durkheim (1) equated the old form of community (living) with similarity and segmentation, (2) hoped that new forms could be based on the similarities and interdependencies inside occupational groups, or on society-wide similarities and a sense of interdependency that are inculcated by way of moral education, and (3) believed that the bonds of community need altruistic sentiments (Durkheim 1964). The distinction between altruism and egoism returns in Weber's distinction between the communal type of social relationship and the associative type. In the first type individuals orient their actions to a subjective feeling that they belong together, whereas in the second case they are oriented to their individ-

Community

and Human Social Nature in Contemporary

Society

9

ual interests. Community can have several affective, emotional or traditional bases: "eine pneumatische Brüdergemeinschaft, eine erotische Beziehung, ein Pietätsverhältnis, eine 'nationale' Gemeinschaft, eine kameradschaftlich zusammenhaltende Truppe" (Weber 1922, 22). Weber considered the family as closest to the ideal type of community. He emphasized that by far the most of contemporary social relations are partly communal and partly associational. Next to the household community, he analyzed other forms, such as the ethnic community and the community of neighbors. The community of neighbors is described as relatively new: the feeling of belonging together is only based on the realization that neighbors are dependent on provision of help on special occasions and in cases of emergencies and danger. Exchange of help is mostly delayed, and if not, than the conditions of exchange are chosen on the basis of what is considered reasonable and fair. This implies that communal relations are durable and personal, contrary to the impersonal (and partly anonymous) relations on the market and in bureaucracies. Weber stressed that similarity of qualities, of situational contexts and of behavior, is not a sufficient basis for community. The latter is only present if "eine gefühlte Zusammengehörigkeit" (ibid.) exists on the basis of a similarity. After Dürkheim and Weber, sociology went through a period in which it was considered to be important to develop into an autonomous discipline. This was accomplished by stressing the social-cultural aspects of social life that are seen as determining perceptions, cognitions and behavior, putting aside altruistic and egoistic motivations (human social nature) as 'psychology' or 'economics'. In this way of thinking community was seen as a group of people who share perceptions and cognitions about common identity and mutual obligations, largely ignoring motivations (e.g. Gusfield 1975). At the end of the previous century some sociologists associated themselves with economics and its theoretical toolbox. This led to the emergence of so-called rational-choice sociology. Precisely within this paradigm much attention was given to the concept of community. For example, Taylor (1982) emphasizes the economic aspects of community: the possibilities to increase collective welfare by way of reciprocity and by way of producing social order. He explicitly designates 'psychological' aspects (friendship and sense of belonging) as things that are often connected with community, but are not an indispensable part of it. And Bowles and Gintis (1998) see communities as special governance structures that solve coordination problems, by way of low costs of information and punishment, and restrictions on interaction and migration. Community is again considered solely from an economic perspective, that is, under the assumption of a simple version of self-interest. Largely the same happens in many contributions that use the (economic) term of 'social capital' in stead of community (e.g. Coleman 1990). Finally, the social network approach defined community by way of formal network characteristics such as density and multiplexity. Density refers to the extent to which all possible links among members of a network are in fact present, that is, the degree to which everyone knows everyone else. Multiplexity refers to the extent to which individuals who are linked in one type of relationships—

10

Henk de Vos

say kinship—are also linked in other types—say co-residence. This approach is indeed extremely 'formal'. Motivations, egoistic or altruistic, do not enter the picture. Concluding, it seems that after Durkheim and Weber sociologists approached community either without referring to human social nature (egoistic or altruistic motivations) or exclusively from the perspective of egoistic motivations. This development can be understood against the background of two different intellectual fashions that dominated the social sciences during most of the twentieth century. Traditional sociology's emphasis on social-cultural aspects is an expression of social determinism: the idea that human behavior results from socialculturally determined perceptions, cognitions and attitudes. The social network approach was easy to reconcile with this model, because of its total neglect of individual motivations. This intellectual fashion, known as the Social Science Standard Model (Cosmides/Tooby 1992) faded away in the second half of the twentieth century. The more recent approach of rational-choice theory had as a positive point that (self-interested) motivations re-entered the scene. But it was also part of the more general social-scientific fashion that reigned for most of the twentieth century, namely: to ignore (altruistic) emotions as proper objects of scientific research. This wave also faded away, starting in psychology (with the rise of evolutionary psychology). In fact, what happened at the end of the twentieth century is that human nature was brought back into sociology. After Durkheim and Weber (and Marx) frankly spoke about human social nature, sociology entered into a period in which the assumption of the 'blank slate' reigned (Pinker 2002). It was taken for granted that humans are able to learn everything that is offered to learn ('equipotentiality'). The withering away of this assumption was an important development within the social sciences and allowed for a more realistic perspective on community.

2. Community and Reciprocity Two late twentieth century developments within the social sciences made the return of human social nature possible. First, evidence accumulated for the designation of a limited set of human modes of interpersonal behavioral patterns and relationships. Second, a profusion of findings about the beneficial effects of 'social support' showed that reciprocity relationships enhance human well being and health. In the following two paragraphs sets of evidence are shortly summarized and interpreted in terms of human social nature. In the third paragraph this information is used to show that community can and should be defined by way of grounding it in human social nature. 2.1 M o d e s of Interpersonal Behavior: Status C o m p e t i t i o n and Reciprocity Following an earlier made distinction between economic and social exchange, Clark and Mills in a series of experiments provided evidence for humans enter-

Community

and Human Social Nature in Contemporary

Society

11

ing into either an exchange or a communal mode of relationship (Mills/Clark 1994). In an exchange mode partners have mutual expectations of monitoring, calculativeness and concern with equity. They expect a service in return without delay or if this is not possible at a contractually agreed future moment. People's orientation resembles the Tit-for-Tat strategy in a sequential two-persons Prisoner's Dilemma Game. Real life expressions would be market and bargaining relations, in general: relations with strangers. Typical for the exchange mode is the interpersonal expectation of an instrumental orientation towards the relationship. In contrast with this, the communal (or reciprocity) relation is seen as a value in itself. Also, partners are vague about the moment that help should be returned and they don't mind very much about the equivalence of values of help given and help returned. They have positive feelings about this mutual noncalculating attitude. A would get angry in case B responded to his receiving help by insisting on balancing the books immediately. The underlying idea is that partners are oriented towards each other's needs. That's why transfers are seen as 'help' instead of 'services'. This orientation towards each other's needs also points to the phenomenon of feeling betrayed when help is not provided and of feeling guilty when failed to help (both without good reasons). The reciprocity mode exists for example in relations between family and friends. The distinction between the exchange and reciprocity modes became well known in the social-scientific and primatological literature (see Smaniotto 2004 and Silk 2003 for overviews). Others suggested that more than these two modes (or algorithms) should be distinguished (e.g. Fiske 1992). The search for these patterns soon becomes arbitrary if structuring principles are lacking. A way to structure the search is to start with well-known biological mechanisms for social behavior. This has the advantage that not only perhaps uniquely human patterns can enter the picture, but also those patterns that humans share with other animals. In fact, the neurochemical basis of much of human social behavior is similar to that of animal social behavior (Panksepp 1998). Modern biology has three biological mechanisms to offer for structuring the study of social behavior: status competition, kin altruism and reciprocal altruism. The first one was already well known and researched by social ethologists, e.g. Tinbergen (1965); the other two are products of neo-Darwinian evolutionary theory (Hamilton 1964; Trivers 1971). Status competition refers to the selection for those motivations and abilities that help in competing with conspecifics over resources (food, territory, mates). Status competition behavior is a complex set of patterns comprising assessment, (ritual or real) fight and eventually flight and avoidance (Parker 1974). It also exists in the form of stable dominance-submission relations or hierarchies, if parties 'agree' who wins and who loses (Chase 1982). The winner acts dominating and the loser submissively. Parties who are involved in a status competition are totally enmeshed in it: all attention is devoted to the ritual or real fight. Of course a fight goes along with a high degree of tension. But also in the stable dominance-submission relation parties still are predominantly oriented towards each other's behavior. The winner continuously verifies whether the loser acts sufficiently submissive, and the loser continuously verifies whether the winner is

12

Henk de Vos

sufficiently reassured by his (the loser's) submissive behavior. Both states are full of tension, leaving little room for play and exploration. The state parties are in is called the agonic mode (Chance 1988). T h e kin altruism mechanism refers to the selection for motivations and abilities that further altruistic behavior (care) towards offspring and other related individuals. Caring behavior comprises feeding, protecting, sharing and, especially in humans: coaching and instruction. It creates circumstances t h a t promote selection for attachment behavior: to stay close to caregivers and to act in ways t h a t further receiving care (Bowlby 1982; Bell/Richards 2000). Adult expressions of care and attachment probably form the human pair bond, in which partners care for and feel attached to each other (Zeifman/Hazan 1997). Reciprocal altruism was suggested as a possible explanation of altruistic behavior towards nonkin. Altruistic or helping behavior (that is, behavior with positive fitness consequences for Alter and negative consequences for Ego) could be a viable strategy under the provision t h a t at a later time Alter returns help to Ego, with net fitness advantages for both (see e.g. De Vos/Smaniotto/Elsas 2000). If humans are selected for motivations and abilities to perform this kind of behavior, this does not necessarily imply that the proximate motivation of the behavior involves the explicit expectation on Ego's side of future return nor a calculating attitude towards help given and received. More plausible is t h a t humans were selected for extending their care and attachment emotions to nonkin, thereby producing conditions that put the ultimate mechanism of reciprocal altruism into operation (De Vos/Smaniotto/Elsas 2000; Bell 2001; Smaniotto 2004). For a good understanding of this possibility it is essential to distinguish between the proximate motivations for altruistic behavior towards nonkin (care and attachment) and the resulting effect that relations of mutual help develop. The fact that such relations, and the individuals within these relations, exist can potentially be explained by the ultimate mechanism of reciprocal altruism. This shows t h a t it is wrong to associate reciprocal altruism automatically with (proximate) bookkeeping expectations and attitude, as often happens (e.g. Gintis/Bowles/Boyd/Fehr 2003). Kin altruism and reciprocal altruism make relationships possible with a completely different neurochemical basis and with different emotions (care and attachment) t h a n dominance-submission relationships. The parties involved are provided with a safe haven, the experience of which promotes playfulness and exploration (Panksepp 1998). Therefore Chance (1988) called the state the parties are in, the hedonic mode. This type of relationship can be called communal (cf. Mills/Clark 1994) or reciprocal (cf. De Vos/Wielers 2003). The three biological mechanisms underscore t h a t there exist two behavioral patterns and relationships for which humans are biologically prepared, in the sense t h a t these patterns are easily and spontaneously learned in a 'normal' social environment (Cummins/Cummins 1999): status competition (ritual or real fight, dominance and submission) on the one hand and reciprocity (care/attachment) behavior on the other hand. Until now we do not know of other mechanisms that could provide similar arguments for distinguishing more than these two patterns. Of course humans are able to act and interact in other ways. For example, they

Community and Human Social Nature in Contemporary

Society

13

can be calculative and bookkeeping oriented, such as in an exchange relationship as studied by Mills and Clark (1994). This is the kind of behavior that is often required in contemporary society with a well-developed market. But at the same time it seems that market behavior is not so easily learned and that market participants often fall back upon either status competition or reciprocity (De Vos/Wielers 2003). Along the same line it can be argued that (Weberianlike) bureaucratic and citizenship behavior is difficult to learn and to perform because people are continuously tempted to shift to either status competition or reciprocity. 2.2 Reciprocity, Status Competition, Well-Being and Health A second set of evidence about human social nature is provided by research findings pointing to positive effects on well being and health of social support and of communal life ambitions. An array of findings shows that social support directly enhances well being and health, specifically related to beneficial aspects of the cardiovascular, endocrine and immune system (Uchino/Cacioppo/Kiecolt-Glaser 1996). At the same time these findings strongly suggest that social support consists of a state of being embedded in a network of reciprocity relations larger than the own household (e.g. Weiss 1980), that not only provides instrumental but also emotional support (e.g. Hobfoll/Stokes 1988), and that not only provides opportunities to receive but also to give support (Antonucci/Fuhrer/Jackson 1990). This constitutes evidence for reciprocity behavior and the maintenance of reciprocity relations not only being easy to learn, but also being a kind of 'natural' way of interpersonal behavior that contributes to overall functioning of the human organism. A similar set of evidence does not exist pertaining to status competition. Should this be interpreted as contrary evidence for status competition being a natural behavioral pattern? Probably not. Findings about the well-being and health consequences of different life ambitions shed light on this issue. Kasser, Ryan and others empirically distinguished two clusters of life ambitions that people endorse: a communal and a status competitive cluster. The communal cluster consists of aspirations such as having good relations with others, helping the world become a better place and personal growth, and the status competitive cluster of aspirations such as financial success, fame and keeping up with fashions. These different ambitions are associated with different levels of well being and health. People who more endorse communal ambitions report more self-actualization and vitality and less depressive symptoms. Also they report less physical symptoms, such as headaches and backaches, and less use of tobacco, alcohol and drugs. And they are less narcissistic (see Kasser 2000 for a review of findings). This is in line with the evidence about the beneficial consequences of social support. But then, why do a certain number of people choose to endorse status competitive life ambitions? A possible answer is that the nature of people's ambitions depends on the nature of their social environment. Hypothetically, the

14

Henk de Vos

degree to which endorsing communal ambitions has beneficial consequences for someone is not only determined by the degree to which this person himself acts according to these ambitions, but also by how many others in his environment do so. If you have communal ambitions and act accordingly, but if at the same time all others in your environment act according to the status competitive mode, your outcomes in terms of well-being and health may differ considerably from a situation in which these others also have communal ambitions and act accordingly. It seems plausible that your worst outcomes are the ones in which you are the only one in your local environment who endorses a specific set of ambitions. So if you expect or observe others to choose for status competition, your will be better off doing the same. Hypothetically, just as in an N-persons Prisoner's Dilemma Game, all of you would have been better off if everyone had chosen for the communal alternative. Now what does it mean that Kasser (2000) found that 'communals' (those who have communal ambitions and act accordingly) as well as 'competitors' (those who have status competition ambitions and act accordingly) both exist? Probably that these groups differ in the nature of their local environment: the communals are more embedded in networks of reciprocity relationships (consisting of communals) than competitors. And competitors have more a local environment consisting of competitors. This arrangement will have emerged by way of each individual reacting to the proportion of communals and reciprocals in his environment and by altering this proportion by this reaction. The process may result in equilibrium, but probably this is not guaranteed. The main point is that status competition may be a natural mode of human interpersonal behavior. Actually, status competition is present in all mammals and reptiles, and therefore is evolutionary much older than the reciprocity mode. But it is triggered in special circumstances, namely by a social environment mainly consisting of competitors. In non-human mammals and reptiles this is the normal social environment. That humans were selected for the motivations and abilities of the reciprocity mode, without the status competition mode being erased, implies that the human social environment can be heterogeneous. A mainly competitive social environment triggers status competition. This will bring the competitors low levels of well being and health, but (hypothetically) higher than in the case that they unilaterally acted communally. Of course the same would apply to reciprocity behavior: it is a natural mode of interpersonal behavior, but it is triggered in an environment of many communals. To the degree that communals cluster, they are predicted to have higher degrees of well being and health than competitors, a prediction that is in agreement with the findings of Kasser c.s. To the degree that they do not cluster, the prediction would be that they shift to the status competitive mode. If they would fail to do so, they would be continuously harassed and exploited by competitors, with of course negative effects on their well being and health. The two modes probably are frequency-dependent strategies, that is, strategies that are enacted dependent on how many others (in the local environment) do the same. This should not suggest that they are consciously chosen. The triggering of a mode probably is a complex process, consisting of two stages. First, although the

Community and Human Social Nature in Contemporary

Society

15

disposition to learn a mode is innate, the actual learning of it needs a social environment in which the individual is engaged and socialized. This is triggering in the sense that a potential to learn is realized and a specific mode of behavior is added to the behavioral repertoire. Second, in the future the learned mode of behavior is triggered or not, dependent on the nature of the local environment. This reasoning implies that altruism, as part of the reciprocity mode, should not be considered and studied as an individual process but as a social process. This is precisely in accordance with a conclusion drawn from an overview of studies of prosocial behavior, namely that the level of community should be the unit of analysis instead of the level of the individual (Bierhoff 2002). 2.3 Community Defined We now arrive at a point at which a definition of community is easy to come up with: a community is a group of individuals with an internal structure of reciprocity relations. The foregoing implies that a community consists of communals, that is, of individuals who act in the communal mode. But at the same time it should be realized that a precondition for people enacting this mode is the existence of the community, that is, the existence of many other communals in one's local environment. The individual and the group level are deeply connected. Also we should realize that the status competition mode is not erased from individuals constituting a community. The mode may be asleep, but there may arise occasions in which it wakes up. This points to the well-known phenomena that individual tendencies to domineer are always strongly reacted to by community members, by way of ridiculing and belittlement, or stronger, ostracizing or even violence (Boehm 1999). For a moment returning to the issue of human social nature: perhaps humans have an innate disposition to punish (within a community context!), even if this is costly, those who cannot withstand the temptations to shift to the status competition mode (cf. Fehr/Gachter 2002). According to this definition the reciprocity relation is the building block of the social system of community. Just as the dominance-submission relation is the building block of hierarchies and the exchange relation is the building block of markets. The deep connectedness of individual and group level in a community implies that a community is more than a simple aggregation of reciprocity relations, that is, the whole is more than the sum of its parts. This points to the fact that a community makes generalized exchange possible: A helps B and at a later moment help is returned not by B but by C. Because all individuals are connected to each other, the identity of the person who returns help doesn't matter any more. This is an important element of the 'efficiency' of a community compared to a single reciprocity relation (cf. Takagi 1999). In a single relationship the probability that help can be returned at the moment that A needs help is much lower than in a community with generalized exchange. The process of generalized exchange makes the group a special entity. It generates a common interest in the continuing existence of the group as a whole.

16

Henk de Vos

3. Community in Contemporary Society 3.1 The Quest For Community Emerging No doubt community was much more prevalent in the societies of the past than in contemporary society. In the remote past, in the long period that humans did not produce food but gathered it on a day-to-day basis, this way of living most probably was essential for their survival and reproduction. Division of tasks and sharing of resources by way of maintaining reciprocity relations probably were the human solution to the challenges of a risky and poorly predictable environment. This solution also involved typical human phenomenon as egalitarianism, empathy, pair bonding, paternal care, extended period of juvenile dependency, long lifespan, and large group size (e.g. Deacon 1997). This communal way of living was deeply modified when humans started to produce their food, by way of domestication of plants and animals. Although the process took thousands of generations, it eroded community and let markets arise, first on a local basis, then regional, and nowadays global. This process is often studied as an exclusively economic one, but it had (and has) a tremendous impact on social life. Reduced to its essentials, it meant that the communal structure disappeared, at least in its pure form. By profiting from economic processes such as specialization, monetarization, and economies of scale, meeting material needs became possible to a formerly unknown degree. But at the same time, social environments of exclusively or mainly communals became less and less a normal thing. The domain of reciprocity relations fragmented into nuclear families and scattered remains of the communities of the past. This went along with an increase of the domain of anonymous and impersonal relationships between people who have at best an interest in a good reputation. Economic welfare, everything you can procure by spending money, increased. But social welfare, the well being and health that you receive by way of being a communal between communals, decreased. This large-scale historical process resulted in societies with low degrees of communal living and social support. And in which overall happiness does not or hardly increase in spite of the tremendous increase of economic welfare (e.g. Lane 2000). Nowadays people are born in a social environment that does not unequivocally guide to communal action. This is not saying that reciprocity behavior is not learned. In general, children's potential to learn reciprocity behavior is realized step by step from early childhood to adolescence, within the confines of family and friendship relations (Laursen/Hartup 2002; Clark/Jordan 2002). But at the same time the large domain of impersonal and anonymous relations becomes part of daily life while growing up. And this domain easy triggers the status competition mode. Assuming that people have at least a rudimentary notion of the beneficial effects of communal living, we should observe individual and collective efforts to maintain and extend their communal domain. For a large part, these efforts require consciousness and explicit attention. In the past, communal life could be taken for granted, but nowadays it is a sort of profession (Willmott/Young 1967, 112).

Community and Human Social Nature in Contemporary

Society

17

It should be noted that this quest for community (Nisbet 1953) does not contradict that people also value freedom and autonomy. A popular opinion is that the communities of the past were oppressive, clinging to outdated traditions and hostile to individual creativity and development. This picture applies to the very specific kind of 'communities' that existed in agrarian societies, with their often high degrees of inequality and oppression. Since it is not so long ago that people lived in these circumstances, the concept of community is easily associated with this specific social form. But agrarian societies only existed for a very short period relative to the existence of humankind. Communal living in the long period before the beginning of food production was very egalitarian. And creativity and autonomy must have been strongly encouraged because they were important communal assets in the struggle for survival. Nowadays people who endorse communal ambitions still are the ones who value personal growth, in contradistinction to those who are engaged in status competition (Kasser 2000). People in contemporary societies have two non-exclusive options for contributing to community living. First they can try to maintain and enrich the remains of community life that are transmitted from the past or that are based on existing long-term relationships. People grow up in a family and make friends in the neighborhood or the village of their youth. They can keep in contact with family and friends, even if distances increase because they and/or the others relocate. And they can try to build a community life if they find themselves surrounded by others with whom they share an expectation of keeping in each other's company in the future. This opportunity arises for example if a person arrives in a neighborhood or village with co-residents who intend to stay for a considerable while, or if he accepts a new job in a company with a low turn-over of staff members. In all these cases the individual finds himself in an environment of already existing long-term relations or he shares with others the expectation of possible future long-term relations. Second, there may be occasions for building community more or less from scratch. This is for example the case if people who are socially isolated happen to meet each other and become friends. Or in a neighborhood without much contact between residents, one or a few persons start to facilitate contacts by announcing a neighborhood party and mobilizing others to contribute to its organization. Or a few persons with a specific need for exchange of information and social support, that is not easy provided by family and friends, start a selfhelp group and mobilize partners in adversity to join. Or unemployed people with poor financial means but a lot of time find each other and start a barter system. Of course, in cases of successful community building, the resulting communities provide opportunities for others to join. For the joiners the option to join is similar to the first option mentioned above. The difference is that in this second case the long-term relationships or the expectation of long-term relationship are not residual, but are intentionally generated. In the following section I discuss some examples of forms of community that belong to either the first or the second option. The examples were chosen so as to have a high degree of diversity of life domains.

Henk de Vos

18

3.2 Maintaining and Enriching Community Life Some well known examples of opportunities for maintaining and enriching community are: family, neighborhood and village, organizational community, and civil society. If applicable, attention is paid to the contribution t h a t communication technologies, the Internet in particular, make or can make. Families In general, families are socially isolated to a high degree. They may be somewhat integrated in their neighborhood or village (see below), but often neighborly relations are non-existent or of limited intimacy and intensity. Because friends often live spatially separated, meetings are infrequent and need planning and coordination. Also members of the wider family often live at other locations. Households themselves became smaller, among other things by way of a strong decrease of coresident adults sharing a home and responsibilities for children (Goldscheider/Hogan/Bures 2000). Nevertheless, quite naturally people try to maintain family relationships even if geographically dispersed. Until far in the twentieth century the exchange of letters was an important medium (Thomas/Znaniecki 1958). Later the telephone had a tremendous social impact (Fischer 1992), just as the automobile. And nowadays, the use of electronic mail for maintaining family contact increases fast. On the one hand, geographical dispersion enhances the demand for these facilities. But on the other hand, the existence of these facilities makes it easier to disperse, and still keep in contact. Although families are less economically interdependent than in the past, there is still a lot of (mutual) support and care (e.g. Marks 1996). And people's well being increases if they care when levels of work-family conflict are held constant across caregiving and noncaregiving employed adults. Caregiving contributes to having more positive relations with others, to feelings of personal growth and of having a purpose in life (Marks 1998). Most caregiving requires physical contact. Often this is not feasible, but it seems t h a t decisions to move or stay are influenced to a considerable degree by the wish to live close to family members. W i t h larger distances, communication by electronic mail may fulfil an important function for keeping in touch and for giving emotional support. Compared with the telephone, and of course faceto-face meetings, the advantage of electronic mail is its asynchronicity. On the other hand, the telephone will remain an important facility for sociability and for reassurance (Dimmick/Kline/Stafford 2000). Neighborhoods

and villages

Neighborhoods and villages have in general a low average length of residence and a low degree of interdependency of its residents. Sheer spatial vicinity of living places has low social significance. Even in the more recent past there existed much community life in some stable neighborhoods, working-class quarters in particular (e.g. Young/Willmott 1977), and in some villages with a high proportion of agrarian occupations. There is no systematic evidence, but probably the number of stable neighborhoods strongly decreased. And apart from t h a t ,

Community

and Human Social Nature in Contemporary

Society

19

neighborhood community life is hampered by the increase of adults with paid work and by sprawl, that is the increase of transport time as proportion of time spent to work, education, household work and leisure (Putnam 2000; De Vos 2003). Communal life in villages has been strongly harmed by the decrease of people working in agrarian occupations. This transformed many villages that formerly had an abundance of economic activities, in sheer living places, for the old, or in dormitory places, for commuters. The spread of the automobile during the second half of the former century was of course a most important factor. First, it contributed to the concentration of economic activities, which, in turn, contributed to sprawl. This must have been an important cause of the decreasing density and multiplexity of social networks. And second, it contributed to the concentration and scaling up of retailing. Shops and hawkers disappeared from neighborhoods and villages. Meeting opportunities and places disappeared, and the sphere of personal relations between the local shopkeepers and their clients, with informal credit, became a phenomenon of the past. In short, local community life is not a thing that people find ready-to-use. Nevertheless, people use the few possibilities that the local environment still has for maintaining a communal life. First, people who live in neighborhoods with higher residential stability do more participate in local communal activities (Sampson/Morenoff/Earls 1999). Second, local interdependency may foster communal life and often is explicitly sought. In the United States residential community associations (RCA's) increased vastly (Kennedy 1995). Also local self-governing communities can be quite successful (Ostrom 1995; Ellickson 1994). This suggests that delegating some governmental authority to a more local level could bring about some degree of communal revival of localities. A further development in this direction would require quick and easy ways of communication, deliberation and decision making, such as by way of local online communities (e.g. Cohill/Kavanaugh 1997). Organizational

communities

Some considerations suggest that organizations can have a community character, but there are also reasons to expect severe limitations. Two considerations suggest that organizations function as communities to some degree. First, many firms have to deal with complex production processes and therefore highly interdependent task structures. This generates a need for employees with high contextual performance and affective commitment to the organization, and because of that a need for long-term relations with employees. Second, it is well known that your colleagues may be also your friends. This suggests that the work place can be a source of communal life. This is confirmed by evidence showing that becoming unemployed not only has negative financial, but also severe negative social-emotional consequences (cf. De Vos 1990). On the other hand, three observations point to the limitations of organizational community building. First, there are and will remain organizations with rather simple production processes with a low need for contextual performance

Henk de Vos

20

and for long-term relations. In these organizations there is not much to be expected of a communal life. Second, one important ingredient of communal life, the multiplexity of relations, can hardly be attained within the context of an organization. Social life on the job and at home mostly are strongly separated, because of work-home distance (sprawl). Third, organizations are hierarchies and therefore easily induce competition for promotion possibilities, which triggers status competition. The nation-state

and citizenship

The increase of the nation-state, and especially the welfare state, is of course furthered by the decline of community. Tasks and functions that originally belonged to communities were taken over by governments (social security, social order, education). This inspired the idea of citizenship and civil society: the nation-state was, or should be, a social form in which individuals' rights and obligations find expression in a way that is similar, although more abstract, to authentic community (Marshall 1964). More concretely it is hypothesized that citizens experience their relation with government as a reciprocity relation, albeit an imagined one. They would not be (very) calculative and consumeristic towards governmental arrangements, not engaging e.g. in tax evasion and fraud. Many countries indeed know a high degree of tax compliance. And more than eighty percent of Dutch citizens agreed with statements that clearly express a reciprocity perspective towards their government (Ester/Halman 1994). Nevertheless, in the Netherlands the expansion of the welfare state was accompanied by increasing concerns about a too consumerist attitude of citizens toward governmental provisions. Indications of an increasing abuse of these provisions piled up. Legislation and implementation were readjusted in order to keep abuse under control. This more strict policy had as a negative side effect that the signaling of the communal mode got even weaker. There are of course serious impediments to the functioning of a nation-state as a community. The nation-state is a highly abstract social construction and some mental effort is required to discern the effects of governmental policies on one's daily life and the consequences of one's own actions on governmental arrangements. It is really somewhat difficult to see oneself standing in a reciprocity relationships with the nation-state. That so many people nevertheless tend to have this view testifies to the strength of their need for community. But it should also be noted that the rise of mass media had important consequences. Television made it easier for voters to develop 'parasocial' relationships (cf. Rubin/Perse/Powell 1985) with politicians. And computer-mediated communication has high potentials for citizen-with-government and citizen-with-citizen communication. 3.3 Building Community A n e w For people who are born and who grow up in contemporary society, it is difficult to realize that their social environment is rather strange seen from the perspective of the history of humankind. Personal experiences with a more communal way of living have disappeared within a few generations, handing down nothing

Community and Human Social Nature in Contemporary

Society

21

else than stories about 'the good old days' and 'paradise lost', that are difficult to grasp. Nevertheless feelings of something being wrong can lead to attempts to intentionally build community anew. They exist of initiatives to bring people together with the aim of building a communal way of living on their own, with varying degrees of exclusion of outsiders, and with varying degrees of efforts to propagandize this way of living to the wider world. In general, these phenomena are less visible, but they do definitely exist in recent periods of modern history. Some forms in which they exist are communes and Utopias, self-help groups and local exchange trading systems (LETS). Communes

and Utopias

Societies in upheaval have always been a breeding-place for communes and Utopias, or, as they are often called: intentional communities. The greatest wave of community building in America occurred in the middle of the nineteenth century. Well-known examples are the Shakers, the Oneida Community, the Amana, and the Hutterites, of which the last two exist until the present day (Smith 1999). Their strivings for building a communal life were concisely described as: "They attempt to repersonalize a society that they regard as depersonalizing and impersonal, making person-to-person relations the core of their existence." (Kanter 1972, 213) A second wave of community building in the United States occurred in the 1960s and 1970s. Kanter (1972) extensively studied both waves. She found that the successful communities succeeded in instilling a high degree of commitment in their members. Her description of the meaning of commitment deserves to be quoted at some length. "A person is committed to a group or a relationship when he himself is fully invested in it, so that the maintenance of his own internal being requires behavior that supports the social order. A committed person is loyal and involved; he has a sense of belonging, a feeling that the group is an extension of himself and he is an extension of the group. ... Commitment thus refers to the willingness of people to do what will help maintain the group because it provides what they need. In sociological terms, commitment means the attachment of the self to the requirements of social relations that are seen as selfexpressive." (Kanter 1972, 66) This clearly is a picture of a social environment that overwhelmingly triggers the reciprocity mode. A special problem is that these intentional communities have to found and maintain a community within a predominantly competitive society. This requires a radical departure from the rest of society. Indeed, communes and Utopias represent the most radical and encompassing ways of trying to realize the ideals of community living. Other ways to build community anew are easier to reconcile with living in contemporary society. Self-help groups The development of self-help groups is generally ascribed to Alcoholic Anonymous, which was initiated in 1935. Since then such groups spread and evolved (Lieberman/Snowden 1993). Self-help groups are cost free, member governed, and peer led and made up of people who share the same problem and who pro-

22

Henk de Vos

vide mutual help and support. They are an alternative or an addition to professional services. Services provided vary widely. Examples are: phone help, buddy systems, home visiting programs, peer counseling, training of peer counselors, transportation facilities, newsletters, public education, speaker bureau, and prevention programs for the community at large (Lieberman/Snowden 1993). People join self-help groups mainly for two reasons (Katz/Bender 1976). First, in a society with a low level of community living many of those people who happen to run into a serious individual problem, are not embedded in a network of reciprocity relations. For them it is difficult to find support in their own social vicinity. A self-help group offers the opportunity to meet others with the same problems. This not only gives them access to others who can provide information, advice and support, but also provides opportunities to inform, advice and support others. So they do not only meet others who are useful to them, but also they can make themselves useful. And because they share the same problem, they will easily understand each other. Second, professional services are inevitably always more or less bureaucratic and impersonal. Although these services may meet the latest scientific and professional insights, their way of deliverance is always highly impersonal, and the services may be poor in terms of empathy and emotional support. Joining a self-help group therefore often is chosen as an important addition. In fact, many self-help groups are professionally initiated and supported. This suggests that self-help groups are an important source of the benefits of community. Running into an individual problem of some sort may be the immediate cause for joining a group, but the consequences of joining may be positive for general well being and health. Are self-help groups indeed much more than an oddity? Lieberman and Snowden (1993) estimate t h a t in 1992 7.5 million Americans participated in a self-help group. Wuthnow (1994) found t h a t in the 1990s forty percent of the adult population claimed to be involved in "a small group t h a t meets regularly and provides caring and support for those who participate in it". In general, members of self-help groups somehow must have found each other. Probably the real demand for self-help groups is larger than the supply, because of an information problem. Clearinghouses can offer some combination of the following services: compilation and distribution of self-help group listings, information and referral to groups, technical existence to existing groups, assistance to persons starting new groups, public awareness, professional education (Meissen/Warren 1993). The rise of computer mediated communication is highly relevant for this information problem. Probably Internet and electronic mail boosted up the development of use of self-help groups during the 1990s. Another problem for self-help groups is t h a t it that (potential) members do not live close to each other, for example because their particular problem is a rare phenomenon. For this reason many groups already used the telephone as a main communication medium. Computer mediated communication facilitates accessibility and alleviates problems of social stigmatization (Dunham/Hurshman/Litwin/Gusella/Ellsworth 1998; Davison/Pennebaker/Dickerson 2000).

Community

and Human Social Nature in Contemporary

Local Exchange Trading Systems

Society

23

(LETS)

LETS are self-organized local exchange systems based on generalized reciprocity, formalized by being recorded in a local currency. The systems are generalized, or multilateral, in the sense that a debt can be repaid to anyone else in the system, not only to the person from whom it was incurred. Starting with a local initiative in Canada, the phenomenon spread to the United Kingdom, Australia, New Zealand, the Netherlands, and to other European countries. In 2000 there existed about 90 LETS in the Netherlands, with a total number of participants of about 6500 persons (Hoeben 2003). In 1994 there were over 300 LETS in the United Kingdom, with a membership of more than 20,000 (Pacione 1998). A LETS is run by a steering group that provides for a bookkeeper who records the transactions (Pacione 1989). Members indicate which goods and services they can offer and they receive a directory that lists all the goods and services available in the system. They can then arrange to trade with each other, paying in the local currency, although no member is obliged to accept any particular invitation to trade. No warranty as to the value, condition, or quality of services or items exchanged is expressed or implied by virtue of the introduction of members to each other through the directory. An open statement of each member's credit or debit balance is provided to all members on a regular basis, and any member is entitled to know the balance and turnover of another member's account. An individual's credit/debit balance does not affect ability to trade. No interest is charged. Prices are a matter of agreement between the parties. Goods and services that are traded are for example: arts and crafts, painting, plumbing, decorating, childminding, dogwalking, baking, gardening, massage, haircutting, shopping, legal advice, car maintenance, tuition, and so on. Many members see LETS as collective effort to build a personalized economy, as an explicit alternative to the official, impersonal, and anonymous market. The phenomenon can indeed be considered as a social invention that reintegrates the economy into a system of reciprocity relations. It is however yet too early to ascertain the degree to which these efforts will be able to continue their initial success. It seems that LETS have low levels of trading activity as a persistent problem (Pacione 1998; Hoeben 2003). Probably LETS could profit considerably from computer mediated communication.

4. Conditions Favoring Communities The foregoing provides several suggestions regarding conditions that are favorable to the viability of communities in present society. In short: the more people are interdependent, the more they have and/or expect to have long-term relationships, the more they have multiplex relationships and the easier they are mutually accessible, the more they will develop mutual reciprocity relations (which means: the more they are a community). Probably community needs all four of these conditions for its (successful)

24

Henk de Vos

existence, but present knowledge is not sufficient to be able to specify the relative contributions (cf. Fischer 1984). The interdependency

condition

People more develop reciprocity relations, the more they expect to have 'things' (goods, help, advice) to offer to each other that lend themselves to delayed exchange. That is: in the long run everyone is expected: (1) to be sometimes in a position of having something to offer that has value for at least one other member, and (2) to be sometimes in a position of needing something that is offered by at least one other member. For example, villagers and neighbors in less developed market societies were more interdependent than they are nowadays, with all the facilities of the market, with money to spend, and with a government sponsored social security system. The awareness of interdependency triggers the communal mode. The condition of (expected) long-term

relationships

People develop more reciprocity relations, the more they expect to have a common future, that is, the more they expect, per relevant time period, small proportions of exit from the group and small proportions of entrance into the group. Actually this condition seems already to be implied by the interdependency condition. But (expected) long-term relationships may exist without much (or any) interdependencies. However, the sheer existence of (expected) durability induces people to deliberately create interdependencies, with the aim of widening one's circle of reciprocity relations. In those cases it is less the material interests that one aims at, than the well-being and health effects of having reciprocity relations. This happens for example by way of trying to find out what you can do to make yourself useful to others. Or you invite others, for example your new neighbors after you moved. In these cases the sheer existence of expectations of long-term relationships induces to signaling the communal mode. The multiplexity

condition

People develop more reciprocity relations, the more dimensions their relationships have. This is so because with more dimensions, there are more opportunities for helping each other. If people are only related to each other as colleagues, they have only the domain of work tasks as a source of opportunities. If they are also neighbors, or members of the same family, or friends, or members of the same association, assistance rendered on one dimension can be returned on another dimension. So the frequency of help exchanged will be much higher than on a one-dimensional basis. Multiplexity is related to interdependency: the more dimensions a relationship has, the higher the potential interdependency of the partners. The easy accessibility condition The more easily people who need something and people who offer something find each other, the more they develop reciprocity relations. Easy accessibility refers to low transaction costs relative to the value of help given. Accessibility

Community and Human Social Nature in Contemporary Society

25

has cognitive, spatial as well as technical aspects. The cognitive aspects refer to easy mutual understanding. People may be in each other's vicinity, but if they don't speak the same language, they may have difficulties in exchanging information about needs, resources and intentions. But even if there is easy mutual understanding, interaction may be hindered by geographic distance (the spatial aspect). On the other hand, the relevance of distance diminishes with cheap facilities for transport and long-distance communication, such as the telephone and electronic mail (the technical aspect).

5. Internet and Community I already mentioned some ways in which computer-mediated communication can help in maintaining, enriching or building community. A more systematic view on the potentials and limitations of the Internet for community living can be obtained by looking at the ways in which the existence of the Internet influences the degree to which the conditions for community are satisfied. Than it is immediately clear that the only effect the introduction of the Internet has is that mutual accessibility becomes easier. On itself the Internet does not change the degrees to which people are interdependent, to which they expect long-term relationships and to which their relations are multiplex. Nevertheless the decrease of costs of accessibility may have significant consequences for maintaining, enriching and building community. For maintaining and enriching community the introduction of the Internet provided an added communication device in the form of electronic mail. The new element in electronic mail is its fast asynchronicity, as compared to the slow asynchronicity of the exchange of postal letters and the synchronicity of the telephone and of course face-to-face communication. This asynchronicity makes communication easier because direct contact is increasingly difficult to bring about under circumstances of high mobility and sprawl. This on itself is favorable to community maintenance and enrichment. Additionally, it may be that the relaxation of time constraints makes communication via electronic mail richer in content and in self-disclosure and empathic connection than face-to-face communication (Davison/Pennebaker/Dickerson2000; Dimmick/Kline/Stafford 2000). It is also possible that simply the act of writing about personal experiences and problems, to which the sheer existence of email invites, has favorable consequences for the participants and their relationships (Pennebaker 1997). Nevertheless, the empirical evidence about the effects of electronic mail on community is scarce and sometimes negative (Kraut/Patterson/Lundmark et al. 1998). Possibly more important is the potential of the Internet for community building. Although the Internet does not affect the interdependency between people, it strongly enhances the probability that people who are interdependent, in the sense that they could help each other, learn about each other and eventually start a relationship. In this way an existing potential for community is realized by people getting to know each other. Of course the nature of the interdependencies is restricted to what can be communicated by way of writing: exchange of valued

26

Henk de Vos

information, sharing of personal experiences and emotional support. Probably the Internet strongly stimulated participation in illness support groups (Davison/Pennebaker/Dickerson 2000). Other examples are newsgroups and bulletin boards, although these often are much less personal and communal t h a n support groups. Finally, local online-communities may contribute to vitalization of neighborhood communities (Cohill/Kavanaugh 1997).

6. Discussion Although community is a core sociological concept, its precise meaning is often left vague. In this article it is pointed out that it is a social form that has deep connections with human social nature. Human social life and human social history can be seen as unflagging struggles between two contradictory modes of human social nature: reciprocity and status competition. Relative to huntergatherer societies, present society is a social environment t h a t strongly seduces to engage in status competition. But at the same time evidence increases t h a t communal living is strongly associated with well being and health. A large part of human behavior and of societal processes are individual and collective expressions of on the one hand succumbing to the seductions of status competition and one the other hand a t t e m p t s to build and maintain community. This is not only a big individual challenge, but also a collective one, and therefore an important policy problem. Government policy is predominantly influenced by economic considerations. This leads to policies t h a t strongly rely on the market mechanism as the main source of everything that people value. But the growth of the market makes people less personally interdependent. And it increases mobility and sprawl and therefore is detrimental to long-term relationships and multiplexity. This points to negative effects of policies to further economic development on the conditions for community. It also points to the fact t h a t a considerable part of economic growth is spent on attempts to maintain and reinstall community, such as transport and communication, by way of increasing accessibility. So economic growth is partly used, by way of earning and spending money, to 'procure' a certain degree of community that was for free earlier.

Bibliography Antonucci, T. C . / R . Fuhrer/J. S. Jackson (1990), Social Support and Reciprocity: A Cross-Ethnic and Cross-National Perspective, in: Journal of Social and Personal Relationships 7, 519-530 Bell, D. C. (2001), Evolution of Parental Caregiving, in: Personality and Social Psychology Review 5, 216-229 — / A . J. Richard (2000), Caregiving: The Forgotten Element in Attachment, in: Psychological Inquiry 11, 6 9 - 8 3 Bierhoff, H.-W. (2002), Prosocial Behaviour, New York Boehm, C. (1999), Hierarchy in the Forest. The Evolution of Egalitarian Behavior, Cambridge/MA Bowlby, J. (1982), Attachment and Loss. Vol. /, New York (first printing 1969)

Community

and Human Social Nature in Contemporary

Society

27

Bowles, S./H. Gintis (1998), The Moral Economy of Communities: Structured Populations and the Evolution of Pro-Social Norms, in: Evolution and Human Behavior 19, 3-25 Chance, M. R. A. (1988), Introduction, in: M.R.A. Chance (ed.), Social Fabrics of the Mind, London, 1-35 Chase, I. D. (1982), Dynamics of Hierarchy Formation: The Sequential Development of Dominance Relationships, in: Behaviour 80, 218-240 Clark, M. S. / S. D. Jordan (2002), Adherence to Communal Norms: What It Means, When It Occurs, and Some Thoughts on How It Develops, in W.G. Graziano/B. Laursen (eds.), Social Exchange in Development, New York, 3-25 Cohill, A. M./A. L. Kavanaugh (1997), Community Networks. Lessons From Blacksburg, Virginia, Boston Coleman, J. S. (1990), Foundations of Social Theory, Cambridge/MA Cosmides, L./J. Tooby (1992), Cognitive Adaptations for Social Exchange, in: J. H. Barkow/L. Cosmides/J. Tooby (eds.), The Adapted Mind. Evolutionary Psychology and the Generation of Culture, Oxford, 163-228 Cummins, D. D./R. Cummins (1999), Biological Preparedness and Evolutionary Explanation, in: Cognition 73, B37-B53 Davison, K. P./J. W. Pennebaker/S. S. Dickerson (2000), Who Talks? The Social Psychology of Illness Support Groups, in: American Psychologist 55, 205-217 Deacon, T. (1997), The Symbolic Species. The Co-Evolution of Language and the Human Brain, London De Vos, H. (1990), The Welfare State and Unemployment: A Theoretical Analysis, in M. Hechter/K.-D. Opp/R. Wippler (eds.), Social Institutions: Their Emergence, Maintenance, and Effects, New York, 205-222 — (2003), Geld en 'de Rest'. Over Uitzwerming, Teloorgang van Gemeenschap en de Noodzaak van Gemeenschapsbeleid (Money and 'the Rest'. About Sprawl, Collapse of Community and the Necessity of Community Policy), in: Sociologische Gids 50, 285-311 — / R . Smaniotto/D. A. Elsas (2000), Reciprocal Altruism under Conditions of Partner Selection, in: Rationality and Society 13, 139-183 — / R . Wielers (2003), Calculativeness, Trust and the Reciprocity Complex: Is the Market the Domain of Cynicism?, in: B. Nooteboom/F. E. Six (eds.), The Trust Process in Organizations, Cheltenham, 75-104 Dimmick, J. W./S. Kline/L. Stafford (2000), The Gratification Niches of Personal Email and the Telephone, in: Communications Research 27, 227-249 Dunham, Ph. J./A. Hurshman/E. Litwin/J. Gusella/C. Ellsworth (1998), ComputerMediated Social Support: Single Young Mothers as a Model System, in: American Journal of Community Psychology 26, 281-306 Dürkheim, E. (1964), The Division of Labor in Society, New York (first printing 1893) Ellickson, R. C. (1994), Order Without Law. How Neighbors Settle Disputes, Cambridge/MA Ester, P./L. Halman (1994), Consumentisme en Waardenorientaties. Is er Sprake van een Consumentistisch Ethos in Nederland? (Consumerism and Value-Orientations. Is There a Consumeristic Ethos in the Netherlands?) in: P. Ester/L. Halman (eds.), De Cultuur van de Verzorgingsstaat. Een Sociologisch Onderzoek naar Waardenorientaties in Nederland, Tilburg, 11-38 Fehr, E./S. Gächter (2002), Altruistic Punishment in Humans, in: Nature 413, 137140 Fischer, C. S. (1984), The Urban Experience, New York

28

Henk de Vos

— (1992), America Calling. A Social History of the Telephone to 1940, Berkeley Fiske, A. P. (1992), The Four Elementary Forms of Sociality: Framework for a Unified Theory of Social Relations, in: Psychological Review 99, 689-723 Gintis, H./S. Bowles/R. Boyd/E. Fehr (2003), Explaining Altruistic Behavior in Humans, in: Evolution and Human Behavior 24, 153-172 Goldscheider, F. K./D. Hogan/R. Bures (2001), A Century (Plus) of Parenthood. Changes in Living With Children, 1880-1990, in: History of the Family 6, 477494 Gusfield, J. R. (1975), Community. A Critical Response, Oxford Hamilton, W. D. (1964), The Genetical Evolution of Social behavior (I and II), in: Journal of Theoretical Biology 7, 1-52 Hobfoll, S. E./J. P. Stokes (1988), The Processes and Mechanics of Social Support, in S. W. Duck (ed.), Handbook of Personal Relationships, New York, 497-517 Hoeben, C. (2003), LETS' Be a Community. Community in Local Exchange Trading Systems, Amsterdam Kanter, R. M. (1972), Commitment and Community. Communes and Utopia in Sociological Perspective, Cambridge/MA Kasser, T. (2002), The High Price of Materialism, Cambridge/MA Katz, A. H./E. I. Bender (1976), The Strength in Us. Self-Help Groups in the Modern World, New York Kennedy, D. J. (1995), Residential Associations as State Actors: Regulating the Impact of Gated Communities on Nonmembers, in: Yale Law Journal 105, 761-793 Kraut, R./M. Patterson/V. Lundmark/S. Kiesler/T. Mukopadhyay/W. Scherlis (1998), Internet Paradox. A Social Technology that Reduces Social Involvement and Psychological Well-Being, in: American Psychologist 53, 1017-1031 Lane, R. E. (2000), The Loss of Happiness in Market Democracies, Yale Laursen, B./W.W. Hartup (2002), The Origins of Reciprocity and Social Exchange in Friendships, in: W. G. Graziano/B. Laursen (eds.), Social Exchange in Development, New York, 27-40 Lieberman, M./L. R. Snowden (1993), Problems in Assessing Prevalence and Membership Characteristics of Self-Help Group Participants, in: Journal of Applied Behavioral Science 29, 166-181 Marshall, T. H. (1964), Class, Citizenship and Social Development, Chicago Marks, N. E. (1996), Caregiving Across the Life Span: National Prevalence and Predictors, in: Family Relations 45, 27-36 Marks, N. E. (1998), Does It Hurt to Care? Caregiving, Work-Family Conflict, and Midlife Well-Being, in: Journal of Marriage and the Family 60, 951-967 Meissen, G. J./M. L. Warren (1993), The Self-Help Clearinghouse: A New Development in Action Research for Community Psychology, in: Journal of Applied Behavioral Science 29, 446-464 Mills, J./M. S. Clark (1994), Communal and Exchange Relationships: Controversies and Research, in R. Erber/R. Gilmour (eds.), Theoretical Frameworks for Personal Relationships, Hillsdale, 29-42 Nisbet, R. A. (1953), The Quest for Community, Oxford Ostrom, E. (1995), Constituting Social Capital and Collective Action, in R. O. Keohane/E. Ostrom (eds.), Local Commons and Global Interdependence, London, 125— 160 Pacione, M. (1998), Toward a Community Economy—An Examination of Local Exchange Trading Systems in West Glasgow, in: Urban Geography 19, 211-231

Community and Human Social Nature in Contemporary

Society

29

Panksepp J . (1998), Affective Neuroscience: The Foundations of Human and Animal Emotions, Oxford Parker, G. A. (1974), Assessment Strategy and the Evolution of Fighting Behavior, in: Journal of Theoretical Biology 223-243 Pennebaker, J. W. (1997), Opening Up: The Healing Power of Expressing Emotion, New York Pinker, S. (2002), The Blank Slate. The Modern Denial of Human Nature, London Putnam, R. D. (2000), Bowling Alone. The Collapse and Revival of American Community, New York Rubin, A. M./E. M. Perse/R. A. Powell (1985), Loneliness, Parasocial Interaction, and Local Television News Viewing, in: Human Communication Research 12, 155— 180 Sampson, R. J . / J . D. Morenoff/F. Earls (1999), Beyond Social Capital: Spatial Dynamics of Collective Efficacy for Children, in: American Sociological Review 64, 633-660 Silk, J . B. (2003), Cooperation Without Counting. The Puzzle of Friendship, in P. Hammerstein (ed.), Genetic and Cultural Evolution of Cooperation, CambridgeMA, 37-54 Smaniotto, R. (2004), Scorekeeping or Bonding? What Do Studies on Hunter-Gatherer Food Sharing Tell Us About Proximate Mechanisms of Reciprocity? (Under review) An Examination of Nontraditional Smith, W. L. (1999), Families and Communes. Lifestyles, London Takagi, E. (1996), The Generalized Exchange Perspective on the Evolution of Altruism, in W. B. G. Liebrand/D. M. Messick (eds.), Frontiers in Social Dilemma Research, Berlin, 311-336 Taylor, M. (1982), Community, Anarchy and Liberty, Cambridge Thomas, W. I./F. Znaniecki (1958), The Polish Peasant in Europe and America, New York (first edition 1918-1920) Tinbergen, N. (1965), Social Behaviour in Animals, London Trivers, R. (1971), The Evolution of Reciprocal Altruism, in: Quarterly Review of Biology 46, 35-57 Uchino, B. N./J. T. Cacioppo/J. K. Kiecolt-Glaser (1996), The Relationship Between Social Support and Physiological Processes: A Review With Emphasis On Underlying Mechanisms and Implications For Health, in: Psychological Bulletin 119, 488-531 Weber, M. (1922), Wirtschaft und Gesellschaft, Tübingen Weiss, R. S. (1980), Loneliness. The Experience of Emotional and Social Isolation, Cambridge/MA Wuthnow, R. (1994), Sharing the Journey. Support Groups and America's New Quest for Community, New York Willmott, P./M. Young (1967), Family and Class in a London Suburb, London Young, M./P. Willmott (1977), Family and Kinship in East London, Harmondsworth (first printing 1957) Zeifman, D./C. Hazan (1997), Attachment: The Bond in Pair-Bonds, in: J. A. Simpson/D. T. Kenrick (eds.), Evolutionary Social Psychology, Mahwah, 237-263

Analyse & Kritik 2 6 / 2 0 0 4 ( © Lucius & Lucius, Stuttgart) p. 3 0 - 4 7

Bernd

Lahno

Three Aspects of Interpersonal Trust

Abstract: Trust is generally held to have three different dimensions or aspects: a behavioral aspect, a cognitive aspect, and an affective aspect. While there is hardly any disagreement about trusting behavior, there is some disagreement as to which of the two other aspects is more fundamental. After presenting some of the main ideas concerning the concept of trust as used in the analysis of social cooperation. I will argue that affective aspects of trust must be included in any adequate account of the role of trust in social dilemma situations involving multiple equilibria. Cooperation in such situations requires coordination even though information on what another player might do is not available. A trusting person can handle such problems of cooperation by framing the situation in a way that goes beyond cognitive trust and solves what I shall call the problem, of normative consent. I will conclude with some remarks about the design of institutions that foster trustful cooperation, especially in the context of the Internet.

1. Introduction The aim of this paper is to present some of the main approaches to the concept of trust as used in the social sciences, in psychology and philosophy. This should provide the theoretical background for the analysis of the problem of trust in Internet communication. I will illustrate my considerations with some simple examples suited to the context of Internet communication, but my argument is concerned with the social problem of trust in general. I will discuss the problem of trust in the context of the general problem of social cooperation. The problem of cooperation is analyzed as composed of three sub-problems, which I will call: • the problem of cooperative incentives, • the problem of information, and • the problem of normative consent. Most theorists agree that trust is a complex phenomenon consisting of behavioral elements, as well as cognitive and affective elements. These elements or aspects of trust are closely related to the three sub-problems of cooperation mentioned above. Trusting behavior is characterized by a certain form of risk, namely the risk that another person may act in undesired ways (section 2). Most cooperative endeavors are characterized by such a risk. A rational individual can incur such

Three Aspects of Interpersonal

Trust

31

a risk only if she has reason to believe that others will respond cooperatively. So, the most basic problem of trusting behavior is a problem of cooperative incentives (section 3). For the most part, finding out what others are motivated to do tends to be difficult. Rational trust does not only require that the incentives of others are in fact such as to motivate cooperative behavior, but a trusting person must also be sufficiently informed of these incentives. This problem of information (section 4) points to the importance of cognitive elements of trust and, thus, to the concept of cognitive trust (section 5). Trust is affective in character in that it induces the trusting person to perceive his partner and the relevant conditions of their interaction in a specific way.1 Such affective trust (section 6) may become important, if a trusting decision cannot be based on information about the incentives of others. The problem of normative consent typifies a class of social situations where such is necessarily the case (section 7). As stated above, there is wide agreement among scholars that trust in general is characterized by behavioral as well as cognitive and affective elements. But when it comes to a definition of trust—surprisingly—one finds predominantly cognitive accounts of trust (I will give a few examples in section 5). To be sure, there are accounts of affective trust as well (I will give a few examples in section 6). Yet, these do not seem to play an essential role in the analysis of social cooperation. However, I will argue that this is due to the misguided concentration on the problem of cooperative incentives and the problem of information. As I will show, there are significant cooperative endeavors, which are characterized by a problem of normative consent. The problem of trusting behavior within these situations cannot be solved without some affective trust. So, if the problem of normative consent is taken into account, it becomes clear that affective elements of trust may very well play an important role in social cooperation. Although I am mainly concerned with conceptual issues here, it is not my intention to give one single, all-encompassing definition of trust. Of course there are problems in the context of social cooperation, the analysis of which afford a specific definition of trust. But this may depend on the specific properties of the problem at hand. So, a general definition of the concept of trust may not be very helpful in such cases. 2 One can, however, identify the main aspects of trust that may play major roles in the analysis of social cooperation. Thus, based on a discussion of some of the more prominent approaches to the phenomenon of trust, I will indicate what I take to be the decisive aspects of trust in this regard. In specifying the concepts of behavioral trust, cognitive trust and affective trust, I hope to provide a conceptual framework that may guide us in our study of trusting cooperation, within the internet as well as in other social contexts.

1

Cf. Lahno 2001 for a more detailed analysis of affective or emotional attitudes. One can still try to give a definition of trust that comes as close as possible to the concept of trust that we use in everyday intercourse, especially, when we speak of 'real' trust as opposed to pure reliance. Philosophers usually try to give such a definition; I do in Lahno 2002. But here I am concerned with trusting cooperation in the most general sense and not with 'real' trust as opposed to pure reliance. 2

Bernd Lahno

32

2. Trusting Behavior Risk is generally held to be one of the main characteristics of trust: the risk that another person may act in ways, which are disadvantageous to the trusting person. In fact, nobody can escape such risks completely. Given the world as it is, we have to face the fact that other people do interfere with our personal plans every now and then. But a trusting person is characterized by the fact that he willingly incurs such risks. He takes these risks, even though he is in a position to rule out such risks. He acts in ways that actually produce relevant opportunities for the person being trusted to do something harmful to him. This is what I call trusting behavior or behavioral trust: A person is showing trusting behavior or behavioral trust, if she (he) is making herself (himself) vulnerable to the actions of another person. Trusting behavior is needed in most cooperative projects. This is, of course, the main reason, why social scientists are interested in trust. Figure 1 shows a simple model of a problem of cooperation to illustrate this point.

A

Figure 1: Trust Game

One may think of the situation after bidder Adam won an auction at eBay in which seller Berta offered some product, say, a first edition of Hume's Treatise. Adam now has to send Berta the payment (plus the transaction costs) as promised in his bid to Berta before she in turn should send the book. If he does in fact send the money (which is CA in the model), she can keep it and still decide on how to react. For the sake of simplicity, let us assume that Berta has only 2 options: she can either keep the book (depicted here as Dg) or send it to Adam (CB)By mutually cooperating, both agents can improve their initial situation. It is assumed that there is some standard of value for each of them such that the

Three Aspects of Interpersonal Trust

33

initial situation, which will be retained if Adam does not send the payment (D^), may be rated by each of them as of value 0, while the cooperative solution bears a prospect of 1 for each. Both would gain from the realization of the exchange (QA; CB). But the cooperative surplus can be achieved only if Adam chooses option CA, that is, if he decides to send the money first. Yet, then Berta—being already in the possession of the money—can decide independently on her own move. If she doesn't send Adam the book, she can gain even more (depicted by b > 1), while Adam suffers from a loss of his money (depicted by a < 0). So, by sending the money, Adam is making himself vulnerable to the actions of Berta.

3. The Problem of Cooperative Incentives In the given game, which is known as the 'trust game' (Kreps 1990), it is assumed that Berta would gain from keeping the book, that is, from 'defecting,' If this is in fact the case, and if Adam and Berta are solely motivated by the given payoffs, and if they are perfectly informed of this, no cooperation is rationally possible. A rational Berta will defect, and a rational Adam anticipating this will not cooperate in the first place. (D^, DB) is the unique equilibrium of the trust game. Without a solution to the underlying incentive problem, there can be no rational trusting behavior in a situation like the one described here. Now, any situation of successive exchange seems to be basically of this sort and, in the same way, bilateral simultaneous exchange is usually a Prisoner's Dilemma situation. Yet, as we all know the exchange of goods and services is indeed possible in the real world. If we do not want to assume widespread irrationality, we are driven to infer that there must be certain aspects of social interaction which solve or at least avoid the problem of cooperative incentives. Three examples of mechanisms that can make cooperation possible are: 1. Internalized morals: Berta might be motivated to repay a cooperative move for moral or fairness reasons. 2. Institutional enforcement: Berta may be deterred from defection by formal sanctions (e.g., she might fear being taken to court). 3. Reputation: Berta may cooperate because that might give her additional opportunities to sell goods at eBay, for if she defects, no one will trust her in that forum again. It is quite plausible that in our eBay example mechanisms of all three sorts are actually at work. It should be noted, however, that these mechanisms work in very different ways. Internalized morals are effective by having a direct impact on the utility functions of actors. Enforcement mechanisms work by effectively changing the consequences of action. And a reputation mechanism operates by placing the situation into a larger social context. But, in the end, all three mechanisms have the same effect: defection becomes less attractive and possibly utterly unattractive for Berta.

34

Bernd Lahno

If mechanisms like these are effective, the game is changed in fundamental ways. One may think that, given such a mechanism, the actual game looks like the one depicted in figure 2. This game differs from the foregoing game in one single respect, namely that Berta assesses one-sided defection now as less valuable than mutual cooperation. This is represented by a different utility value b* » < » » [ Browse I

Sri!

Il S ^ c ^ f S w r c h j H g t p \ o w v t e t » [ a l t I nevw 1 t v e n t T l p e o p l e

Valuae

IssBJsàt

eôBFy ta a community that encourage* 00**' and tioneat communication among alt (ta m«mb«fa. O u community i» g u x M c y f w » lurxJamentat values « • • • •

We We We We We

betieve peiopte a r t basically good believe everyone ha* something t o coninOut» batleve that an honest, open environment c a n tying B. That is, the esteem that is forthcoming if you are observed is larger the smaller the likelihood ex ante that you would be observed. On this basis, E = A - (A - B) • P

This equation is consistent with our stipulation that when P = 0,E = A; and when P =1,E = B. Now, what value of Pi would maximize expected esteem? On the one hand, esteem is higher if the probability of being observed is lower; but then there is a chance that you won't be observed at all, and then you will get no esteem. So it can't be the case that it is best for you when P is zero. But equally, expected esteem is not necessarily maximized when the probability of being observed is one. Expected esteem is: P • E = P{A - {A -

B)P]

which is maximized: either when P = 1; or when P = A/2{A — B), which value of P we denote P*. In this latter case, the optimal value of expected esteem is A/2. In the former case, the value of esteem is B. So if B > A/2, then it will be desirable to have P as high as possible. But if B < A/2, then there is a range where expected esteem and probability of being observed are inversely related. A diagram may help here. Consider Fig 1. On the horizontal axis we show the value of P, ranging from zero to one. On the vertical axis we depict both the value, E of the esteem accruing, if observed, and the expected value of the esteem—the product of P and E. We can see by appeal to the diagram that if B < A/2, we have an interior solution, with P* less than 1. If B > A/2, we have a corner solution in which the highest expected esteem occurs when P is 1. In Fig 1, we have shown the former case. So far, we have taken it that the probability of being observed is an exogenous factor. But individuals can work to alter the probability of being observed: they can thrust themselves into the light—they can blow their own trumpet.

Esteem, Identifiability and the Internet

151

Figure 1: Optimal Probability of Being Observed

Or they can hide their light under a bushel, or modestly change the subject when their accomplishments become the topic of conversation. These strategies are themselves esteem-relevant: modesty tends to be positively esteemed; selfaggrandisement and bragging tend to be disesteemed. This fact introduces a further complication. Consider, for example, the case where P happens to fall precisely at P*. Then the esteem-maximising individual reason will act to reduce the probability of being observed, because his modesty will earn him further esteem. Indeed, those incentives will be operative even if P is originally somewhat below P*, provided that P lies not too far below. 6 At the same time, if P is low, it will pay the esteem-maximising individual to work to increase observability, despite the esteem cost of the self-aggrandising actions involved. These considerations explain why it may be in the (esteem) interests of an individual to court some measure of secrecy, even when the acts undertaken are esteem-worthy ones. There is no puzzle involved when a person who engages in scurrilous conduct seeks secrecy: such a person has an interest in minimizing the likelihood of discovery. But the thought that secrecy can be an esteemmaximising strategy in cases where the action reflects credit on the actor is

6

These considerations also bear in the case where B > A / 2 and hence P* = 1.

152

Geoffrey Brennan/Philip

Pettit

a mildly puzzling one. This possibility is, however, perfectly consistent with plausible assumptions about the nature of esteem and its pursuit. Consider in the light of this discussion, the situation of a person who writes under a pseudonym and acquires thereby a first-class reputation as an admirable writer. It is not self-evident that the best strategy in maximizing esteem is to declare one's 'true identity' immediately. Perhaps people will think that you are self-aggrandising. Perhaps they will be less inclined to buy your books once it is revealed that their author is just grey old you. Best, of course, if people come to discover that you are the 'famous author' more or less by accident—or later, when you are ready to retire and have built up a reputation not just for your writing but, by implication, for your modesty. After all, you always have the option of revealing your identity at any point. You can, if you choose, keep your esteem in reserve—stored at the bank rather than spent, as it were. It need not be the best strategy to go public immediately.

5. Anonymity as an Ad Hoc Strategy Apart from the general considerations related to insurance and modesty, there are a variety of more or less ad hoc reasons, some more relevant in one of our three sorts of cases than in others, why people might be prompted to seek anonymity; in particular, why they might be prompted to seek anonymity out of a concern for their esteem and reputation. We look at two. 5.1 I n - G r o u p E s t e e m Consider first the case of the secret society. Within any such society, members operate both as performers and as privileged observers. They earn esteem from each other and from no-one else. That is, outsiders cannot provide esteem for me by virtue of my being a member of the Bourbaki group because they don't know that I am one. Keeping membership secret is then a mechanism for declaring to all members that theirs is the only esteem that counts as far as each member is concerned. But this declaration can be a very significant source of esteem to members. It is a commonly observed property of esteem that people especially value the esteem of those that they esteem. Hume (1978, book 2, chapter 9) puts it thus in his discussion of fame: "though fame in general be agreeable, yet we receive a much greater satisfaction from the approbabtion of those whom we ourselves esteem and approve of, than of those whom we hate and despise." So I discover, when asked to join the secret society, that my esteem matters to these people who I know I have reason to esteem. And if my esteem matters to them then that is in itself a source of esteem for me. Further, all members are quite special in this regard. These others who I esteem apparently care not a whit about the esteem of outsiders: they seem to care only about the esteem they get from me and the other members. Return now to the pseudonym case. My identity as 'the famous author' is almost always known to some people—my editor, my agent, my inner circle of friends and family. And I always have the option of telling the secret to special

Esteem, Identifiability and the Internet

153

others, of course swearing them to secrecy in the process. Those in the know form a secret society of a kind. It is not just that they are party to a secret—that they know something that others do not know and perhaps would like to know. It is also that when I admit them to my inner circle I declare to them that their esteem is especially valuable to me. Given the reciprocal nature of esteem, this is a signal that I especially esteem them. In this way, secrecy affords me the capacity to give special signals of esteem. A related benefit of secrecy is that while those who belong with me in a secret society, or those I let know of my fame under a pseudonym, will be bound to respect the confidence involved, they need not be inhibited from speaking well of me more generally. Thus there can be a powerful benefit in the likely testimony that such people will give me for embracing the hidden bonds that bind me to them. And this benefit is the greater because the testimony thus offered is not seen to redound to their own glory. 5.2 S e g m e n t e d Audiences The case of segmented audiences invokes somewhat similar considerations to those already canvassed. Consider the case where A is a good performer in two separate arenas where those two arenas appeal not just to separate audiences but to somewhat opposing ones. I am a good author; and also Jewish. I recognize well enough that being Jewish is an object of general disesteem in the population in which I am located, or at least is so among some people. Actually, I despise such people for their prejudice. And I relish my Jewish identity. On the other hand, I want people to buy my books and to read them 'without prejudice' as we might put it. If it became known in the Jewish community that I was writing as if a gentile, then this would not be approved within that community and might be regarded as outrageous by some. In such a case, the logic of a pseudonym seems clear. I want to segment the relevant arenas in which I can earn esteem. On the one hand, I want to develop my talent and be appreciated simply as an author. On the other, I want to be recognized as a decent committed member of my cultural and religious community. This situation arguably describes the George Eliot case. Mary Ann Evans was a controversial individual, with a somewhat dubious public reputation. She was living out of wedlock with a married man; she had strongly expressed unconventional religious views. She was not exactly infamous—but might have become so were she to come to more extensive public attention. Better to avert such a risk by writing under a pseudonym. And better to do so under a pseudonym that does not declare itself immediately as such—so precisely not 'a Lady', or 'Boz' or 'Publius' or 'Mark Twain' (a name that would have declared its pseudonymous qualities at least to anyone familiar with Mississippi river boat calls). Now, it need not be the case that the 'natural audiences' in the two arenas disesteem each other. Perhaps the disesteem is only in one quarter. Or perhaps the disesteem is not general but only occurs within a minority of those involved in one activity or the other. Still, in these cases, it can be esteem-maximising to

154

Geoffrey Brennan/Philip

Pettit

segment the audiences; and the pseudonym provides a mechanism for securing that segmentation. It should also be recognised that audiences sometimes segment naturally. You are, let us suppose, a singer and a golfer, and the overlap between the groups who are expert in these fields and whose esteem is really worth something may be very small. Segmentation of audiences just occurs automatically. Nevertheless, your esteem will be magnified if your golfing colleagues are aware that you have some prowess as a singer, and vice versa. Positive esteem is likely to 'compound', in the sense stipulated in the introductory section. In this kind of case, you will not rationally work to keep the audiences separate unless some of the considerations we have explored earlier (secrecy effects, or risk management issues) come into play—or unless golfers tend to hold singers in contempt (and/or vice versa). On the other hand, segmentation in such cases is unlikely to cost you much. If experts count disproportionately for esteem and expert groups are totally disjoint, the 'compounding effect' does not generate much additional esteem—or more accurately, the additional esteem is not worth very much, coming as it does from a group that is uninformed about one or other of your accomplishments.

6. Back to The Future: The Internet Context The aim in this paper has been to explore some of the implications of a desire for esteem and for esteem-based reputation, for the operation of Internet relations. One of the features, we have said, of Internet relations is that they are often pseudonymous: many individuals conduct their most significant Internet transactions via e-specific identities. Perhaps in some cases, the adoption of such e-specific identity is necessary because off-line names contain too many characters or are not unique; but it is increasingly just an emergent convention. Many of the pseudonyms are clearly recognisable as pseudonyms: Names like '#9**ms' or even 'hotchick-in-fairfax' are not for off-line use. But there is scope for use of pseudonyms that might be interpreted as 'real' names, and in this respect scope for some measure of deception. Such deception possibilities may make one generally anxious about Internet relations. And there are of course contexts in which such anxieties may well be justified. In economic trading settings for example, the rewards to deception can be considerable. And some commentators fear for the long-term viability of e-trading precisely because of these sorts of difficulties with verifying identity. However, even in these cases, the problems can be overstated. Even in market contexts, agents seem to care about their reputations for trustworthiness and honest dealing as an end in itself (or more accurately, as a means to greater esteem), as well as for economically instrumental reasons. More generally, the Internet seems to be an especially fruitful source of possible esteem. It offers potentially large audiences of an appropriately fine-grained kind. What is crucial, of course, for the effectiveness of esteem on the Internet is that agents care about the reputations that their e-identities secure. The fact that such e-identities are often pseudonymous—and where not, are difficult to check—certainly moder-

Esteem, Identifiability and the Internet

155

ates the forces of disesteem for some range of actions and actors. The kind of anonymity involved means that e-identities that lack reputation have nothing to lose by acting in a disestimable manner. However, the same is not true for e-identities who have established a reputation already: they have esteem to lose. And even those without a (positive) reputation aspire to have one. Behaving poorly on the Internet always means positive esteem for behaving exceptionally well forgone. And there is no reason to believe that real actors do not care about their virtual reputations. That is something that we think the analogy with the use of pseudonyms in literature establishes quite clearly. There is no reason to think that pseudonymous reputations cannot be a source of esteem to the generator of the pseudonymous material. George Eliot has a reputation as an author; people esteem George Eliot in that connection; and Maryann Evans has every reason to care about that reputation and to act in whatever way she can to sustain and augment it. There may be a cost in esteem terms of the pseudonymous strategy, of course: namely, that in the normal case, the esteem that an individual derives from the multitude of activities she engages in tends to be more than the simple sum of the separate pieces. I will esteem you more when I observe your estimable qualities across a wide range of arenas. If this is so, then in the general case esteem considerations would encourage individuals to operate under a single identity. And the adoption of a separate e-identity might on such grounds be seen to take on a sinister cast. However, as we have been at pains to show in the foregoing discussion, there can be countervailing, esteem-based reasons for maintaining separate identities— and for a separate e-identity specifically. For example, having a distinct e-identity can be a risk-management strategy, in the sense that one's off-line esteem is protected from the consequences of eactions. If one's initial e-actions turn out to produce disesteem, the reputational cost is negligible. One can simply jettison that particular e-identity, and begin e-life anew with another. The resultant removal of downside risk can be highly liberating. One can be more adventurous, more speculative, in one's initial einteractions precisely because one's reputation elsewhere is protected. One's inhibitions are lowered. Rather like the eighteenth century masked ball—an institution that seems to have been invented precisely with the intent of lowering inhibitions and promoting general licence—participants do things they would otherwise not do. But in the Internet setting, at least, there seems no general reason to fear the effects of this removal of inhibition. If our reasoning is broadly right, individuals will be seeking positive esteem in their Internet transactions; they will not in most cases be using the cloak of pseudonymity to do outrageous things. But they may well experiment with things that they would be inhibited from pursuing in offline life, and some of those experiments will prove successful. In these successful cases, e-life can have a quality, and constitute a source of considerable positive esteem, that offline life lacks. And indeed, we would expect that most ongoing Internet relations will have these characteristics: those whose e-life does not have them will have less incentive to maintain their e-life. Of course, once one's e-life proves successful, there will be some incentive to

156

Geoffrey Brennan/Philip

Pettit

integrate reputations—and the maintenance of separate real and virtual identities seems on that basis to represent a residual puzzle. What we have tried to argue in the foregoing discussion, is that this is less of a puzzle than it might seem. Esteem maximization does not always call for identity-integration. And it will be a useful way of summarizing our argument to enumerate the reasons as to why it might not. First, if one's e-reputation is strong and one's off-line reputation lackluster, it may diminish one's overall esteem to integrate on-line and off-line identities. One's on-line reputation may be damaged by one's off-line mediocrity (and a fortiori if one's off-line reputation is a source of disesteem). Better then to keep one's e-reputation unsullied. Second, there is some presumption that those who develop on-line relations will think that on-line activities are estimable. Not all those who admire you in ordinary life will necessarily share that view. And perhaps not all those who operate on the Internet will think that being a successful stockbroker or a wellreputed clergyman is such a big deal. In all discretionary activities, there is a presumption that those who are involved think the activity to be worth-while and success in it to be highly estimable. But those not involved may have other attitudes. Perhaps it would be better to keep one's audiences segmented. Third, even in the contrary case where one's off-line and on-line reputations are rather impressive, declaration might seem to be self-aggrandising and therefore serve to diminish one's esteem. Although integration would add to your reputation and esteem overall if it were to occur by accident, deliberate action on your part to bring that integration about, runs the risk of seeming immodest. A little cultivated modesty may be esteem-enhancing. Finally, retaining separate identities in general allows me to share the secret in particular. I can admit specially selected e-persons to the inner circle of those who 'know who I really am'; I can reveal to special personal friends my online activities. This can be an added source of esteem to me in itself; but it also provides a source of esteem to them, and thereby is likely to induce some measure of reciprocal esteem. Effectively, one is creating a kind of small 'secret society' around the dual identities; and secret societies can be the source of special esteem benefits. The bottom line here is that Internet activities can be a significant source of esteem for those who operate 'well' in virtual space. Agents have reason to care about their e-reputations, even where those reputations attach to a pseudonymous e-specific identity. This being so, the desire for esteem will play a significant role in stabilising decent communication practices and supporting the operation of other social norms on the Internet. In this respect, there is no reason to think that the Internet context will be significantly different from interactions in the 'real' world.

Esteem, Identifiability and the Internet

157

Bibliography Brennan, G. /P. Pettit (2004), The Economy of Esteem: An Essay on Cimi and Political Society, Oxford Carty, T. (2000), A Dictionary of Literary Pseudonyms in the English Language, Chicago (2nd edition) Elster, J. (1983), Sour Grapes, Cambridge Hume, D. (1978), A Treatise of Human Nature, Oxford Lovejoy, A. O. (1961), Reflections on Human Nature, Baltimore Room, A. (1989), A Dictionary of Pseudonyms and their Origins, With Stories of Name Changes, Jefferson (3rd Edition) Smith, A. (1982), The Theory of the Moral Sentiments, Indianapolis Young, E. (1968), The Complete Works. Vol 1, Hildesheim

Analyse & Kritik 26/2004 ( © Lucius & Lucius, Stuttgart) p. 158-184

Chris Snijders/Richard

Zijdeman

Reputation and Internet Auctions: eBay and Beyond

Abstract: Each day, a countless number of items is sold through online auction sites such as eBay and Ricardo. Though abuse is being reported more and more, transactions seem to be relatively hassle free. A possible explanation for this phenomenon is that the sites' reputation mechanisms prevent opportunistic behavior. To analyze this issue, we first summarize and extend the mechanisms that affect the probability of sale of an item and its price. We then try to replicate the results as found in four recent papers on online auctions. Our analyses reveal that (1) it makes sense t o differentiate between 'power sellers' and the less regular users, (2) there are variables that have an effect on sales that are often not controlled for, (3) one should carefully consider how reputation is operationalized, (4) neglecting heteroscedasticity in the data can have serious consequences, and (5) there is some support indicating that effects differ across auction sites.

1. Introduction 1.1 Trust and Online Auctions Thousands of individuals sell and purchase items from one another on the Internet. These people meet each other on virtual market places or auction sites, such as eBay.com, ePier.com, or qxl.co.uk. Chances are that you are one of them, but in case you do not know how it works, the procedure roughly runs as follows: sellers promote goods they want to sell by giving a product description on an auction site. Buyers can compete with each other through an auction system to purchase the promoted product. The winner pays, usually upfront, the money to the seller and then the seller ships the item. Anybody with access to the internet is a potential seller or buyer on an auction site. An important aspect of these online C2C markets is that consumers can trade without direct physical interaction. In other words, consumers can engage in trade with others whom they do not meet. In a way, this means that the market gets closer to what a 'perfect market' is supposed to be. Unfortunately, this key aspect of online auctions is its Achilles' heel in trading as well. Lack of physical interaction means that careful checking of products and trading partners is hardly possible. Perhaps an even larger drawback is the time lag between purchasing and receiving goods. At least one party has to send in his good, either the item or the money, without strong guarantees of getting something in return. Nevertheless, some major online auction sites are quite successful. For instance,

Reputation and Internet Auctions: eBay and Beyond

159

in 2001 eBay.com counted 29.4 million registered users (van Swol 2001), while in 2003 eBay claimed to have 12 million items on sale in 18,000 categories on any given day. How can this be? To analyze this situation, previous researchers (e.g. Ba/Pavlou 2002; Diekmann/Wyder 2002) have already mentioned the obvious underlying trust issue. A buyer places trust by bidding on an auctioned item, whereas a seller may honor trust by shipping the auctioned item, or may abuse trust by either shipping an item of lower quality than advertised, or by not shipping the item at all (we disregard the related but smaller trust issues on the seller's side, such as buyers who win an auction but do not pay). Information often plays an important role in trust problems. In online auctions, two questions are typically of interest to the buyer. Can I trust this buyer? and Can I trust this item? If a buyer is uncertain enough about either one—the buyer's intentions or the product's quality—a potentially mutually profitable sale may not materialize (Akerlof 1970). Therefore, providing a buyer with information on the quality and value of a product and on the 'kind' of seller one is dealing with may help to decrease the trust problem, as auction sites obviously recognize. At least two mechanisms are in place that may provide a seller with such information: through an extensive product description the seller can try to reduce uncertainty about the quality of the good, and a reputation system exists as a measure of the reliability of the seller. Online auction sites provide a mechanism for their users to evaluate each other. Although not all reputation mechanisms are alike, all do provide information on a seller's past behavior on the auction site. For instance, after a completed auction on eBay.com, buyer and seller can (but need not) evaluate each other by 'leaving feedback' about the other party. Negative feedback is represented by a score of '-1', '0' resembles neutral feedback, while '+1' represents positive feedback. When a user first registers at eBay.com he receives a reputation score of zero. Over time, buyers and sellers leaving feedback leads to the reputation score increasing or decreasing according to the behavior in auctions the user participated in. A more extensive elaboration on eBay's reputation system is available at: http://pages.ebay.com/services/forum/feedback.html. Different reputation mechanisms have emerged on different auction sites. Most, if not all reputation mechanisms are at least partly quantitatively based. Every review left by a party results in feedback, and the summation of all points is a person's 'reputation score' or 'feedback rating'. In principle the reputation mechanisms can differ with respect to the kinds of feedback that can be given: positive, neutral, or negative. eBay's feedback mechanism—at least at the time of the data collection for this paper—allows positive, negative, and neutral feedback. The reputation score is calculated as the difference between the number of (unique) positive comments and the number of (unique) negative comments. Except for quantitative comments, most larger auction sites such as eBay and QXL allow for small bits of qualitative reputation as well. Seller and buyer can provide a couple of lines of text regarding the party they dealt with. What the reputation mechanisms have in common is that they are all acting as measurements of the trustworthiness of the people participating in the auction.

160

Chris Snijders/Richard

Zijdeman

1.2 Mixed Findings in the Literature A number of recent papers have considered the effects of reputation on sales and the results are typically mixed. In fact, the most straightforward effect one would expect, a positive reputation having a positive effect on both the probability of sale and the price, is not always found. Some researchers even come up with highly unexpected effects, such as a negative reputation having a positive effect on an auction's end price (Eaton 2002; Kauffman/Wood 2000). Dellarocas (2003) provides the most recently published overview; many working papers on the topic are available from the Reputations Research Network (http: / / databases.si.umich.edu / reputations). Lee, Im, and Lee (2000) give three possible reasons for the variation in the results found in previous papers by for example Kauffman and Wood (2000) and Ba and Pavlou (2002). First, they mention the differences in the types of products under study. Although they are not very concrete about how this might matter, a likely explanation is that the risk involved in purchasing an item depends on a number of product characteristics, combined with the assumption that a buyer's 'need' to deal with a seller of high reputation differs for different levels of risk. A buyer's risk is larger for items of larger value, so for the larger value items reputation might matter, whereas it has no impact when one considers items of smaller value. The need for information on a seller's trustworthiness may also depend on a product's condition. Ba and Pavlou (2002) believe that the fact that they did not find any significant effects for negative reputation is because they considered used and refurbished products, as opposed to only new products. The probability that an item malfunctions is lower for new products than for used products, the risk of purchasing a broken item is therefore higher for used products than for new products. Accordingly, buyers may be more sensitive to a seller's reputation when purchasing a used item. A similar argument holds for the complexity of a product. As Buskens mentioned, "complex products leave more opportunities for hidden defects" (Buskens 1999, 132). The more features a product has, the more features might be malfunctioning, increasing a buyer's risk of purchasing a product that is not perfect. Therefore, buyers may have greater need for information on the trustworthiness of the seller when buying more complex products than when buying less complex products. The second factor that might cause mixed results is the way the independent variable of reputation is operationalized. Eaton (2002), for instance, reports that he finds no significant effects for negative reputation unless negative reputation is being dichotomized. Other researchers log-transform the reputation score, include or exclude the negative comments in their measurement, and so on. Third, the way in which data was collected could contribute to the mixed results. While some researchers control for variables concerning product condition, others control for transaction related variables like shipment costs and means of payment, whereas still others control for hardly any other possible intervening variables. Finally, Resnick, Zeckhauser, Swanson, and Lockwood (2002) point to the effects of the different statistical methods that have been

Reputation and Internet Auctions: eBay and Beyond

161

used. Most researchers have used OLS regressions; some have used Heckmanlike selection models instead. 1.3 Research Questions To better understand behavior in online auctions, we employ the following strategy. First, we try to systematize and extend the theory as found in previous papers on reputation in online auctions. We then try to replicate the results of four of those studies using auction data we collected ourselves. In our data we tried to include all of the variables that were included in previous studies, and added some that we considered of possible interest based on our theoretical account. We then test the hypothesized effects using these data. The following research questions are at the core of this study: What is the effect of seller reputation on sales? What is the effect of product description on sales? What other factors affect sales in online auctions? Do effects of reputation differ across auction sites? The latter question needs some additional explanation. Since most research on online reputation has been conducted on eBay.com, it is wise to think about the extent to which results are 'eBay-specific'. There are indeed some arguments that suggest that effects of reputation may differ across auction sites. For instance, one alternative site we consider is ePier.com. This is a general auction site, largely similar to eBay, but with a different message. ePier.com claims to be "waiting to set internet free" and for this reason charges no listing fees (what you have to pay to list your item) and smaller final value fees (what you have to pay the auction site if the item is sold). In general, ePier.com tries to create a different identity than eBay.com, portraying an image of eBay as the big commercial auction site and of ePier.com as a 'happy community' where one comes to exchange goods with friends. In such a place, if it exists, a reputation system would add no value.

2. Auctions as Trust Problems Let us first make the resemblance between auctions and trust issues more concrete. Coleman (1990) mentions four elements that define trust situations between a trustor and a trustee. The first is the fact that placing trust provides a trustee with resources, that he otherwise would not have received. The second element is that a trustor is better off placing trust than not placing trust, provided that the trustee is trustworthy. If the trustee on the other hand is not trustworthy, the trustor is worse off placing trust than when not placing trust. Third, Coleman stresses that the placement of trust by a trustor is voluntarily and without formal safeguards. Finally, there should be a time lag between a trustor placing trust and a trustee honoring trust. As one can easily see, auction sales can be considered a trust problem according to this definition. Figure 1 shows the trust problem between seller and buyer in online auctions (the Trust Game: Dasgupta 1988; Kreps 1990). After a seller has opened the auction, a buyer can choose to either place a bid (cooperate) or not (defect).

Chris Snijders/Richard

162

Zijdeman

If a buyer chooses to defect, she has no chance of either receiving a potential gain (Rb), nor a potential loss (S). Her payoff is then lower than the potential gain, but higher than the potential loss. A seller will try to encourage a buyer to place trust, because if a buyer defects, the seller will receive the lowest payoff possible: Ps. However, if a buyer places trust, she is uncertain whether the seller will honor or abuse trust. The chance of receiving payoff Rb depends on the probability that a seller is trustworthy. Although Rb is the highest possible payoff for a buyer, by cooperating, a buyer runs the risk of receiving the lowest possible payoff S, which means that after a buyer has paid the seller, the seller defects and ships an item of lower quality than advertised or does not ship the item at all. Basically, if T is larger than Rs, the seller has an incentive to abuse trust. There can be reasons, however, why abuse by the seller is not rewarding after all, for instance because the seller can anticipate that abuse would lead to negative feedback on the auction site, which could frustrate future sales. The 'effective T', so to speak, could then be smaller than Rs. Though the seller is aware of his effective T, the buyer is uncertain about this value, and can try to infer something about it, for instance based on the seller's reputation score. Typically, the buyer is also uncertain about the value of Rb- Again, the buyer can try to decrease this uncertainty before buying. Note that instead of explicitly modeling future trade as new Trust Games, or uncertainty through moves of nature, we choose to define the rewards of future trade and uncertainty implicitly in the values of the payoffs (cf. Raub/Snijders 2001). Buyer

Seller

right

•

Buyer

Pb (20)

Ps (20)

(0)

T (80)

Rb (40)

Rs (40)

(hxrii

right

Seller

S

(tmvn

Figure 1: Trust Game. S < Pb < Rb and Ps < Rs < T. Example payoffs between brackets.

Reputation and Internet Auctions: eBay and Beyond

163

3. Explaining Trust through Seller Characteristics: Reputation Although seller characteristics other than reputation could well be of influence on sales and are included in our data collection and analyses (for instance, one could imagine that it helps if a seller has a label 'seller is shopowner' in his auctions), we first focus solely on the seller's reputation. Reputation:

Basic

Argument

It is straightforward to show how, at least in theory, reputation can help solving the trust problem. One could do with three assumptions: 1. a winning buyer always provides feedback on a seller's behavior after each completed auction; 2. a winning buyer always provides feedback honestly; 3. a seller's behavior is stable over time, in the sense that his past behavior is a strong predictor of his present behavior. If a seller succeeds in auctioning a product he receives a reputation rating as indicated by the first assumption. Prom the second assumption one can conclude that a seller's reputation is an accurate overview of the behavior of that specific seller. The third assumption allows to make inferences of the kind "if this is a high reputation seller, it is likely that my trust will be honored". Let us now examine to what extent these three assumptions are realistic. Judging from the high reputation scores and the large number of sellers with a (positive) reputation, it is certainly true that buyers (and sellers) provide feedback. However, Resnick and Zeckhauser report that on eBay.com only a little more than 50 percent of the buyers provide feedback and warn that it may be " ... dissatisfied customers [who] are substantially less likely to give feedback" (Resnick/Zeckhauser 2002, 3). There are at least two reasons why it is harder to give negative than positive feedback. First of all, giving negative feedback comes with a certain risk. A seller who is sanctioned by a buyer, could retaliate by giving negative feedback to that buyer. In addition, by giving negative feedback a buyer admits that she has been 'outsmarted' by a seller, which is to a certain degree loss of face for the buyer. The second assumption does not seem to be realistic as well. Those who want to give negative feedback when positive feedback would have been appropriate can do so. Lee et al. have already pointed out that the subjectivity of ratings can indeed lead to unfairly low or high ratings, though others have argued that this is a minor issue "a large volume of accumulated information about a seller might converge to a certain point that represents the level of trustworthiness of the seller" (Lee et al. 2000, 4). There is also an incentive to provide positive feedback regardless of the behavior of the other party. The reason for this is that providing negative feedback regarding a bad seller or buyer is a collective good. The negative feedback is useful for others who may want to engage in future transactions

164

Chris Snijders/Richard

Zijdeman

with the seller who cheated you, but comes at a potential cost to the one who gives it. One runs the risk of retaliation with negative comments, and could in fact be missing out on positive feedback one could have received in case of a positive review. The final assumption stated that a seller's behavior is stable over time, or at least an accurate assessment of a seller's honesty. As we argue later, the argument given by Lee et al. (2000) that the volume of accumulated information converges to the true level of trustworthiness has a double edge. For a seller, the incentive to earn a positive rating decreases as his reputation increases because one negative rating hardly affects a positive score of 1,000. Moreover, a seller might just be waiting and increasing his positive score in interactions of smaller value, only to deceive when stakes are high (in several fraud cases this seems to have been the strategy that was used). In sum, one might expect that a seller's behavior will not deviate for long periods of time, but that especially high reputation sellers have incentives to defect when the occasion presents itself. Although the argument supporting an effect of reputation on sales is straightforward and compelling ("higher reputation goes with more and better sales"), there are clear objections that raise doubts as to whether the conditions under which auctions are run actually support a useful implementation of this argument. Will the real Reputation

please

rise?

There are many ways in which one could incorporate the reputation score, and researchers have used different ways to measure what they feel is the most adequate representation of the seller's 'real reputation' (whatever that may be). Some have used the total number of (unique) positive feedback, others count only the (unique) negative feedback, and some use ratios such as the percentage of all feedback that is positive. Given that there is a wide range of positive reputation ratings—on eBay.com there are sellers with reputation scores well over 1,000—it may be a good idea to transform the scores. Based on the plausible idea that the added value of an extra 'reputation point' decreases as reputation increases, many researchers have considered not the reputation score, but its logarithm (Ba/Pavlou 2002; Diekmann/Wyder 2002; Resnick/Zeckhauser 2002; Houser/Wooders 2000). Including the neutral comments increases the number of ways to operationalize reputation even more. Moreover, on eBay and other large auction sites the feedback profile does not only provide an overview of all feedback ratings for a seller. It also provides an overview of feedback ratings separately for the last week, the last month, and the past six months. This allows users to subjectively value ratings differently depending on the time it has been since the ratings have been given. There are at least two arguments why information on the age of feedback ratings could influence the weight of each singular rating. First of all, because people's behavior might change over time, more recent past behavior could be a better predictor of near future behavior than less recent past behavior. This would imply that sellers with high positive feedback scores would have a smaller benefit from their oldest feedback ratings than from their more recent feedback ratings (which ac-

Reputation and Internet Auctions: eBay and Beyond

165

tually goes against the argument that the 501st positive rating adds less than the 11th). Another implication of that idea would be that larger time gaps between feedback ratings have a negative effect on the influence of feedback ratings itself. Related to that, one can imagine buyers are able to discern patterns in a seller's behavior. For example, consider a seller with a feedback score of 87 consisting of 90 positives and 3 negatives, which is not too bad a score. Then if the negatives were collected in the three most recent transactions, this seems a reasonable indicator of 'a seller gone sour' and could influence buyer behavior. Upfront we do not have very compelling thoughts as to which of the above is the best measurement, so we will treat this as an empirical question and get back to this issue when analyzing the data. We do note, however, that researchers often pick one without much argumentation, which may have given rise to the inconsistent results across researchers.

4. Explaining Trust through Buyer and Product Characteristics Basically, the elements of a sale on an auction site are the following: a buyer buys a product from a seller on an auction site (through a product description). Logically, one can expect differences with regard to trust that relate to precisely these different emphasized elements. Some characteristics of buyers may affect probability of sale, some items are perhaps more easily sold, some sellers are more likely to sell, and some auction sites might generate higher prices than others. In this section we consider these separate elements and review and elaborate on the arguments that may affect sales in online auctions. Buyer

Characteristics

Not all buyers have the same need for information to estimate the probability of a successful sale. Although buyer characteristics are hardly mentioned in the literature, whether trust is placed does not solely depend on the trustworthiness of the seller, but also on the trustfulness (the degree to which a person is willing to trust) of the buyer (Snijders 1996). When a buyer bids at an auction site for the first time, she has no experience, for instance, regarding the effectiveness of the site's reputation mechanism to keep fraudulent sellers out. As a buyer successfully engages in more auctions, past experiences reduce the uncertainty about the probability of a positive outcome. Therefore, a more experienced buyer (with positive experiences) will have a smaller need to reduce uncertainty through information from reputation and product description than a buyer who has no experiences. We therefore hypothesize that more experienced buyers are willing to pay higher prices, and that effects of reputation and product description on sales are smaller for experienced buyers (with positive experiences).

Chris Snijders/Richard

166

Product

Zijdeman

Characteristics

Part of the risk a buyer runs in online auctions is caused by the characteristics of the product. As the buyer's trust problem increases, she will demand more safeguards from a seller (cf. Snijders/Buskens 2001). In other words, a buyer's need for certainty increases as the risk of the transaction increases. Assuming that reputation indeed reduces uncertainty about a seller's trustworthiness, then a buyer's demand for reputation will increase as the level of risk involved in the purchase increases. We consider three characteristics of products that influence the risk involved for the buyer: the value of a product, it's condition, and it's complexity. Product value is the product characteristic that is most often controlled for in previous research. For instance, in Eaton (2002) one of the main reasons to choose Paul Reed Smith Guitars as the object of study was the relatively high price (typically over 1,000 US Dollar), combined with the fact that this is a good with a quality that is difficult to assess. The monetary risk a buyer runs is higher for high value products than for low value products. We therefore hypothesize that the effect of reputation on sales increases as product value increases. A second product characteristic that influences a buyer's risk is whether the product is used or new. Ba and Pavlou (2002) point out that for used or even for refurbished products, "[...] the quality variance might increase significantly, which means the risk level for the buyer also increases". Besides the objective variance in quality, subjective interpretation of the description of the quality matters as well. For example, the interpretation of a description 'in reasonable condition' is ambiguous, whereas the description 'item is unused' leaves no room for discussion. Given that purchasing used products is more uncertain than purchasing new products, one would expect reputation to have larger effects for used and refurbished products than for new products. Third, the complexity of the product is also likely to influence a buyer's risk. Complex products leave more opportunities for defects that are hidden in the description, or can at least be more easily denied afterwards ("I did not know that option 23 in menu 4 did not work"). The uncertainty about the value of the good, and thereby the value of Rb, decreases as the complexity of the product increases. We therefore hypothesize that for complex products, the effect of reputation on sales should be larger.

5. The Combination of Seller and Product: Product Descriptions The most basic information that is displayed in (almost) all product descriptions in auctions is a product's brand name and type. In fact, this is one way for the seller to convey quality and functionality of the product. In a similar context, Resnick and Zeckhauser (2002) refer more generally to standardization as one way of signaling information. Besides brand and type, one can try to reduce the inherent trust problem by including pictures. As Eaton puts it "[t]o reduce uncertainty concerning the condition of a product a seller can post pictures

Reputation and Internet Auctions: eBay and Beyond

167

of an item for sale" (Eaton 2002, 8). It makes sense to distinguish between two types of pictures: pictures taken from some catalogue, and home made pictures. Sometimes sellers add pictures to their description that are taken from the website of the original manufacturer (e.g., a picture of a Nokia phone, as it appears on Nokia's website). Such pictures can give a good idea about what the product looks like in general and might be more useful than just referring to the product type, especially when the number of different products with a similar type number is large (Nokia 5200, 3100, 3300). Such a picture does not give any idea about the product's actual condition, of course, which is why it can make sense to include home made pictures (i.e., pictures made by the seller) as well. The pictures can likewise signal information about the seller when the picture reveals something unrelated to the product itself, but possibly related to characteristics of the seller. Adding home made pictures showing a phone in pristine state but in a really messy room might do the seller more harm than good. Although they may prove to be very important, differences between kinds of home made pictures are (empirically) beyond the scope of this research. However, we do consider the difference between home made and catalogue pictures. In accordance with our earlier argumentation we hypothesize that adding a picture improves sales, and adding a home made picture helps more than adding a catalogue picture (given that the picture is of reasonable quality). In addition, and in line with Ba and Pavlou (2002) and the argument that a buyer's risk is higher for used products, we expect that the beneficial effect of pictures is larger for used products than for new products. In the previous paragraph we already pointed out that for certain information derived from the product description it is unclear whether the information concerns the product or the seller. For example, some sellers mention that the original packaging is included with the sale. In principle, the packaging has little or no value, but one could argue that a person who keeps a product's original packaging is a person taking more than average care in handling his products. The interpretation of such information is anything but straightforward and largely beyond the scope of this paper. An interesting example is a seller who indicates that a product has been slightly scratched, thereby giving the impression that he is honest by telling that the product is not in mint condition, which implicitly suggests that, besides the scratch that is mentioned, nothing is wrong with the product (or otherwise the seller would have told you). Obviously, at the heart of this problem lies the fact that the product description is not provided by some outside objective source but by the seller of the good. This is in itself already problematic for the buyer, since the seller is the person with whom the buyer has the trust problem to begin with. In principle, this implies that all hypothesized effects of characteristics of the product description are conditional on the buyer trusting the description. For this reason one could expect that the effect of characteristics of the product description are stronger for sellers with higher reputations. Unfortunately, it is just as easy to argue the opposite. For if a seller has a low reputation score, reputation cannot be used to reduce the trust problem, and only product information can help to reduce it. If this argument is correct, characteristics of the product description only

Chris Snijders/Richard

168

Zijdeman

have an effect when reputation scores are low, since these are the only cases in which there still is a trust problem to solve. Since we see no particular reason to favor one or the other reasoning, we leave the effect of product description as an empirical issue.

6. Contextual Factors: Type of Auction and Site Characteristics We now discuss some arguments rarely mentioned or measured in the literature that could partially explain mixed results found in previous research for the effects of reputation. The general argument is that a number of characteristics might have had an effect on sales, but have previously not always been controlled for. The first part of this section concentrates on auction characteristics such as whether or not the seller uses a 'reserve price' (i.e. a minimum selling price), and how long the auction period lasts. The second part focuses on site characteristics such as the ratio of demand and supply for a good and a site's reputation. Auction-type

Characteristics

Besides reputation and product information some researchers claim that other auction characteristics also influence sales, or at least correlate with the probability of sale and the end price. Since most of these characteristics are generally not controlled for, they might also form an explanation for the different findings with regard to the effects of reputation as found in previous research. This subsection discusses five auction characteristics that could affect sales. The first auction characteristic we consider are the available means of payment for a buyer. Eaton suggests that "[...] escrow services may actually send a negative signal to prospective buyers [especially if] a seller has negative feedback" (Eaton 2002, 18). Diekmann and Wyder report that the higher the reputation of a seller, the more likely that he will demand payment in advance (Diekmann/Wyder 2002, 20). Since payment in advance is more risky to a buyer, this could have a negative impact on sales and thus obscure an effect of reputation if it is not controlled for. The second and third characteristic are the bid increment and the reserve price. Strictly speaking the size of the bid increment should not matter: everybody bids up to his or her maximum, and the only difference with larger bid increments would be that bidding stops a bit sooner, which could lead to a somewhat lower price. Nevertheless Diekmann and Wyder find positive effects for bid increment on end price (and no effects on probability of sale). Perhaps there are other issues at stake, such as the subjective idea that when bid increments are larger, you are more likely to think that the good is valuable. Similar arguments apply to the minimum bid and to (sometimes secret) reserve prices. Lower reserve prices attract more potential buyers, thereby increasing the number of bidders. A fourth and perhaps a somewhat underrated characteristic that could explain differences in reputation effects are the shipment costs. In the (working) papers published up to April 2004, the size of the effects of reputation on end price are no larger than a few percent of the product's value

Reputation and Internet Auctions: eBay and Beyond

169

(Melnik/Alm 2002; Houser/Wooders 2000; Eaton 2002; Dewan/Hsu 2001). In our study, shipment costs ranged between roughly eight and twenty US Dollars. Estimating the mean price of all completed auctions in the research on 200 US dollars, the difference between the highest and lowest shipment costs are about six percent of the mean auction price. This implies that, given that the effects of reputation from previous research are rather small in size, forgetting to look for the shipment costs might prove a bigger mistake than forgetting to consider the seller's reputation. Finally, the auction duration could have an effect on the probability of sale or on an auction's end price. The longer an auction lasts, the more time people have to notice the product and be interested, which increases the probability of sale and the end price. On the other hand, a longer auction likewise implies there is a larger probability that potential bidders find an attractive other item on offer, so that the net effect is not clear. There is some evidence though that most of the bidding occurs close to the end of the auction. An extreme example of this is the bidder strategy 'sniping' (Roth/Ockenfels 2000). The strategy comes down to trying to squeeze in your bid as close to the auction deadline as possible to avoid other bidders from overbidding you and thus avoiding a price war amongst bidders. Roth and Ockenfels (2000) show that this strategy is used more often on eBay.com than on Amazon.com since Amazon uses a system with an open ending rule for the auction duration. (The auction does not close until a certain amount of time after the highest bid has been placed.) Since most previous research has been conducted on eBay.com one might argue that this strategy does not affect the mixed results found for reputation effects. However, Roth and Ockenfels (2000) report that even within eBay.com there is a large difference between categories of products with respect to the prevalence of sniping. Site

Characteristics

Apart from auction characteristics, site characteristics can also influence the effects of reputation on sales. Each auction site consists of a number of sections, for instance cars, real estate, electronics and art. Judging from the differences in the total number of auctions, not all sections enjoy the same amount of activity. Differences in the number of sellers and buyers within a certain auction section interfere with effects of reputation. First, demand for and supply of products is likely to influence sales, for obvious standard economic reasons. Some interference can also be expected from differences in mean scores of reputation across sites or across sections within sites. For instance, on eBay.com a reputation score of +50 is not bad, but a +50 on ePier.com is outstanding. These differences are all hard to control for. Measuring demand and supply for, say, a mobile phone is hard to do, because it is not clear what the reference category is (all identical phones? all similar phones? all phones?). Reputation mechanisms differ across sites, and even within sites over time. Some mechanisms only monitor negative reputation, while others distinguish positive, neutral and negative comments. A more subtle difference can be found between the reputation mechanisms of ePier.com and eBay.com. While ePier.com generates a reputation score from the total number of ratings, eBay.com

170

Chris Snijders/Richard

Zijdeman

only takes unique ratings into account (so that 10 positive feedback comments, all from person A to person B only count as +1 for B). Such deviations in reputation mechanisms and thus in the way reputation is calculated, may cause differences in reputation effects. Even if two sites have the same reputation mechanism, effects of reputation could differ per site because of other structural differences between sites. This becomes clear when one looks at the way in which auction sites advertise themselves in the 'about us'-section (examples below are taken from the sites in 2003). For example, eBay.com clearly markets their large number of members and auctions, trying to convey that eBay.com is where the action is.1 The Dutch QXL site, Ricardo.nl, does not display any figures about the number of auctions, but refers to two tests, one by a Dutch computer magazine, the other by an 'independent research bureau' using a 'research instrument developed by the University of Amsterdam'. Both tests call Ricardo.nl the best Dutch auction site, although it is not clear what that means. A somewhat different approach is chosen by ePier.com. Apart from an 'about us'-section, they have a special section called 'eBay vs ePier'. This section shows a price comparison for insertion and final value fees between eBay.com and ePier.com. EPier.com does not charge insertion fees and charges lower final value fees than eBay.com for all item values (at least at the time of data collection). EPier.com is trying to position itself as the friendly alternative to eBay.com, perhaps hoping that users with anti-eBay sentiments organize themselves in ePier.com and make it flourish. In any case, in principle this kind of marketing could lead to users identifying themselves with the auction community. A strong belief in the honesty of other members of an auction site and a strong moral imperative to treat other community members honestly could then make a reputation system superfluous. Table 1 provides an overview of our hypotheses.

7. Data and Replications We start by trying to replicate the results of four earlier conducted researches, albeit based on newly collected data. Our choice of papers to a large extent reflects the papers that were available to us at the start of our research; meanwhile several other (working) papers have been published (see Dellarocas 2003 for a recent overview). We chose to try to replicate Eaton (2002), Lee et al. (2000), Ba and Pavlou (2002), and Diekmann and Wyder (2002). Eaton (2002) analyzes the effect of reputation and product description on sales of Paul Reed Smith guitars. Lee et al. (2000) focus on the effects of negative feedback on the difference between retail and auction price for both new and used monitors 1 Interestingly, it is not that obvious that a larger site is necessarily a good thing for a potential user. A potential buyer knows there is a large probability that what he wants to buy is actually for sale, but also knows there is a lot of competition from other buyers. A potential seller knows there are lots of potential buyers, but also that there are a lot of rival sellers. Perhaps the reason for eBay.com to advertise with size of the site is that it wants to claim that the auction system works there, and on a grand scale.

Reputation and Internet Auctions: eBay and Beyond Table 1

171

Overview of hypotheses Probability of sale +

End Price + +

Interaction with reputation: Value of the good Good is used Complexity of the good

+ + +

+ + +

Picture included" Picture is home made" Description is included"

+ + +

+ + +

Reputation of seller Buyer is experienced

a

Possibly, interactions with reputation exist. An argument can be made

that these effects only surface when reputation is high (because this is the only case when a picture can be trusted) or only when it is low (because this is the only case in which one needs a picture to compensate for a lack of reputation).

Table 1: Other variables mentioned in the main text are variables about which we do not have any specific hypotheses. and printers. Ba and Pavlou (2002) examine the effects of feedback on what they call 'price premium' for 18 different products, such as Pokemon Gold cards, camcorders, music and software, and include (the log of) positive and negative feedback to study the effect of seller reputation on price premium. Diekmann and Wyder (2002) study the effects on sales of reputation and different auction characteristics, such as auction duration and bid increment. They use auctions of the Nokia 8310 from the Swiss QXL-site Ricardo.ch. Our data has been gathered (by hand) from November 2002 until February 2003 from four different auction sites. Two auction sites are Dutch: eBay.nl and Ricardo.nl. The other two auction sites are US-based: eBay.com and ePier.com. The number of cases varies between the four auction sites: 111 from eBay.nl, 125 from Ricardo.nl, 103 from eBay.com and 79 from ePier.com, adding up to a total of 418 cases from 266 unique sellers. We chose to consider auctions of different types of Nokia and Motorola mobile phones. Mobile phones are popular on auction sites and although standardized products, they have the potential to induce trust problems, given their vulnerability and complexity. Data were collected on both new and used phones, but only data on phones in working condition have been gathered. Also both simlocked and simlock free phones are in the data; sometimes accessories were included in the sale. Besides 'standard' variables concerning reputation, auction duration, number of bids, and number of pictures, we collected other possible intervening variables

172

Chris Snijders/Richard

Table 2

Overview of t h e variables collected in this a n d o t h e r studies.

SELLER Variable

Author

Reputation ( # ) Negative R e p u t a tion (d) Negative R e p u t a tion ( # ) M e m b e r s h i p : Duration Comments ( # ) Comments past 7 days Comments past month C o m m e n t s past 6 months Qualitative Reputation Removed from auction A u c t i o n s Deleted Stars (#,color) O n l i n e store (d) New Seller-tag ID-tag Me-tag BUYER Variable M e m b e r s h i p ; Duration Reputation Feedback ( # )

PRODUCT DESCRIPTION Author Variable

AUCTION Variable

[E][*] [Eiri

Picture ( # ) P i c t u r e (d)

Price P r e m i u m Auction Duration

m

H o m e m a d e Picture(d,#) Warranty (length) W a r r a n t y (d) Color (d)

pi [*i

pi

[E]P] [L][E][*]

Zijdeman

Author Men*]

pip]

[*]

S h i p m e n t Cost

P]

Bid I n c r e m e n t

[*] [*]

Price Net P r i c e

[*]

Bids ( # )

PI

pip] p]p] [E][L][P]P]

PI

[*]

Function Desc. (d) Reason of Sale (d)

[*]

Bidders ( # )

[*]

C o n d i t i o n (d)

PI

Buy now ($)

PI

H

Original Box (d)

{*]

Buy now p u r c h a s e

PI

[*]

ri n

[*i

M a n u a l (d) Url of p r o d u c t (d) Phone nr./email of seller

PIP]

(d) [*1

[*] [*]

[*] pi [*] Author

CONDITION Variable

PI

R e f u r b i s h e d (d)

pi pi

Used (d) Condition

ACCESSORIES Variable

Author

[L] n w

ri

PAYMENT Variable P a y in a d v a n c e Cash & Pickup C a s h & Delivery Cash W i r e & Delivery Pay with CC Escrow Service

Author p] p] p]

pi pi pi P][E] PJ

[E]

Author

C h a r g e Device ri Covers n Headset ri Beltclip [*] Accessories (d) pi C a p i t a l s between brackets i n d i c a t e d w h e t h e r a variable was included in t h e focal s t u d y , w i t h [E]=Eaton(2002), [L]=Lee,Im,Lee(2001),[P]=Ba& Pavlou(2002),[D]=Diekmann & Wyder(2002), and [*]=Snijders & Z i j d e m a n . ( d ) = d u m m y v a r i a b l e , ( # ) = v a r i a b l e r e p r e s e n t n u m b e r of ... ( $ ) = p r i c e in US Dollar or E u r o .

Reputation and Internet Auctions: eBay and Beyond

173

such as miscellaneous seller properties, characteristics of the product description and the means of payment. Table 2 gives an overview of the variables used in the four papers we try to replicate, and of our own data. Seller reputation (when defined as number of unique positive feedback comments minus number of unique negative feedback comments) in our data is as follows: -1(3 sellers), zero (88 sellers), + 1 (19 sellers), +2 (20 sellers), and a maximum reputation score of +1032 (1 seller). The mean of the seller reputation is 58. Reputation scores differ between auctions sites, as can be seen in Table 3. EBay.com is by far the site with the largest reputation scores and ePier.com by far the site with the lowest.

Table 3

Overview of reputation scores (number of unique positive minus number of unique negatives) across auction sites.

Site

mean

median

Ricardo.nl eBay.nl eBay.com ePier.com

15.6 42.0 131.6 2.3

1 3 40 0

75-percentile 10 27 145 0

maximum 519 867 1032 87

A picture is included in 80% of the auctions; a home made picture is included in 25% of the auctions. The buyer's are generally a bit less experienced than the sellers: 2 buyers with a reputation of -2, three with -1, 92 with zero, and a mean of about +13. Our auctions end in a sale in 51% of the cases, with prices ranging from about 10-25 Euro for the oldest and crappiest phones up to 500-900 Euro for the most expensive models. First a few general remarks on the process of replication. We have tried to copy the research methods of each paper as much as possible. To keep matters tractable, we tried to replicate at most two analyses from each paper, the ones that we thought were at the heart of the issue in the particular paper. Space constraints do not allow us to discuss the separate replications in detail, but a detailed account of the way in which the replications were run can be found in Zijdeman (2003). Here we summarize our replication results. The first thing to note is that the results from our replication vary considerably from the original results, and it seems unlikely to us that all differences can be attributed to differences in the way in which we performed the replications. A summary of the findings of the replication of previous results is shown in Table 4. These meager results show that the results on the effect of reputation are not only different across papers, but also across replications. Apparently, there is still much to be learned here. Some remarks on the most conspicuous replication results are in order. First, in some analyses on the probability of sale, seller reputation has a negative sign. In the replication of Ba and Pavlou this negative

174 Table 4

Chris Snijders/Richard

Zijdeman

Overview of the number of successfully replicated effects, compared to the total number of effects aimed to replicate (hence, 2:4 means "2 out of 4 replicated").

Probability of Sale Observations from: Eaton Diekmann et al. Ebay.com only 2:4 1:4° All Sits 3:4 1:4 Observations from both Dutch sites were used, not

End price Lee et al. Ba et Eaton 2:4 2:5 1:5 4:5 3:4 from eBay.com.

effect of reputation is even significant. It seems quite unlikely that the perceived trustworthiness of the seller decreases with increasing reputation. One possible explanation for this finding is that as a seller's reputation increases, his auction behavior changes in a way that has a negative influence on the probability of sale, and there is some evidence for this in the data. For instance, sellers with high reputation scores are more often 'power sellers', who steadily offer their goods on sale for relatively high prices, in the expectation that in the end someone will buy at this higher level. A related argument is that sellers with high reputation scores could be trying to receive higher prices as a bonus for their perceived trustworthiness. In any case, this is something to reckon with in the extended analyses on our data below. A second interesting finding is that the effects of reputation on an auction's end price are so diverse. For instance, Ba and Pavlou found a significant positive effect for negative reputation (!) on end price, but we could not replicate this result. A likely reason for this positive effect is that there is a high correlation between positive and negative feedback (in our data 0.85). Positive and negative feedback can then hardly be distinguished and may actually both measure something like 'eBay experience'.

8. Analyses on the Mobile Phone Data In the analyses of our mobile phone data we intend to test our hypotheses, as well as control for possible effects of additional variables that we collected. First we turn to our measurement of reputation. Reputation

Resnick and Zeckhauser (2002) report three ways in which reputation has been modeled in previous research. As we mentioned before, in some papers only negative feedback is used to represent a seller's reputation. In other papers the feedback rating or the 'net score' (positive feedback minus negative feedback, most of the times based on only the unique comments) is used as an indicator for a seller's reputation. Finally, in several papers the logarithm of feedback or the logarithms of positive and negative feedback are used to operationalize seller reputation. When comparing the results of previous research the question arises to what degree these measurements of reputation correspond. Table 5 shows the correlations in our data between the measurements of reputation based on

Reputation and Internet Auctions: eBay and Beyond

175

observations from all four auction sites. The table shows that correlations between different ways of operationalizing reputation are high. When we use the observations from each auction site separately, the correlation matrices based on the observations of eBay.com and eBay.nl show similar results, and the correlation matrix based on observations from ePier.com even shows somewhat higher correlation coefficients.

Table 5

Correlation matrix for several indicators of reputation based on observations from all four auction sites (N=418). Seller reputation

# positive 0.99* comments negative 0.84* # comments Net repu0.99* tation Log 0.74* positive reputation Log 0.90* negative reputation Negative -0.07 reputation (dummy) * indicates significance at p

# positive comments

# negative comments

Net reputation

Log postive reputation

Log negative reputation

0.85* 0.99*

0.83*

0.74*

0.54*

0.74*

0.90*

0.87*

0.90*

0.77*

-0.07*

-0.10*

-0.07*

0.23*

0.049

= 0.05 level.

Besides the total number of positive, neutral and negative feedback, the four auction sites also provide information on the number of feedbacks received in the last week, the last month (4 weeks on ePier.com) and the last 6 months. A comparison of these numbers could, in principle, give some insight into whether the effects of more recent feedback differ from the effects of less recent feedback. Again the strong correlations prevent such conclusions. First of all, 'feedback in the last month' is necessarily correlated with 'feedback in the last 6 months', simply because one is a subset of the other. We therefore subtracted the feedback from the last week from the feedback from the last month. A similar procedure was used for the feedback from the last month and the last half year, to get separate measures for feedback in the last week, the three weeks before that, and the five months before that. Even then the correlations remain strong, and using logarithms or comparing negative and positive feedback does not make these strong correlations disappear. The smallest correlation coefficient we found was between positive and negative feedback in the last week (r=0.375); most other correlations between different ways of measuring reputation are above 0.75. Though previous papers have operationalized reputation in various ways, our data show strong correlations between different measurements of reputation. On the one hand this suggests that the different ways of operationalizing reputation are not likely to be the reason for the different results in different papers. On the other hand this finding is problematic because it makes it hard to disentangle

176

Chris Snijders/Richard

Zijdeman

precisely what drives the effects of reputation on sales. In our analyses we chose to use the reputation score as used by eBay.com, the number of unique positive feedback items minus the number of unique negative feedback items. Characteristics

of Buyer,

Product,

and

Seller

The only buyer characteristic in the analyses is the reputation rating of the winning buyer, as a measure of buyer experience. We hypothesized that more experienced buyers (with positive experiences) perceive a smaller risk - for instance because over time they start to believe that the auction system actually works—and are therefore willing to pay higher prices, as compared to buyers with less experience. Another argument, with opposite implications, is that with experience buyers learn to better evaluate what is a bargain and what not. This would instead suggest that 'rookie buyers' are more likely to overpay than seasoned eBay professionals. We also argued that there may be a difference between pictures copied from a website and home made pictures. The number of catalogue pictures included in an auction ranges from one to seven, except for a single outlier with 14 pictures. The number of home made pictures included in an auction ranges from one to six, except for two outliers from a single seller with 18 home made pictures. After removing the outliers the total number of pictures correlates almost equally with the number of catalogue pictures and the number of home made pictures, both 0.65. There is a small negative correlation, -0.13, between the number of internet pictures and home made pictures. We use both variables in the analyses. Our Table 2 showed a number of variables representing elements in the product description. For example, whether the color of the phone was given and whether a function description of the phone was given. Although we controlled for all of these variables in various analyses, few of them turned out to influence sales, only and sometimes did. Therefore, only these two have been included in the analyses as controls. We start our analyses with a base model including seller reputation, the number of unique positive feedback items minus the number of unique negative feedback items, to measure reputation. With respect to the product description we start with whether or not one or more pictures were included (), the total number of pictures (), and whether home made pictures were included (). Buyer experience was measured with the buyer's reputation score (). We also include whether a good was used () and whether the good was in a good condition (), ranging on a five-point scale from poor ('1') to new ('5'). The complexity of goods, which we hypothesized to have an effect on the effect of reputation on sales, hardly varies in our data so we do not include it. We also control for the four different kinds of auctions (eBay.com, eBay.nl, ePier.com, and Ricardo.ch), for the different types of phones, and for the date (at the end of the auction), because prices of mobile phones decrease over time. More elaborate models include several variables related to the characteristics of the auction itself such as means of payment, kinds of bid increments, shipment costs, etc.

Reputation and Internet Auctions: eBay and Beyond

177

Explaining the Probability of Sale There is clustering in the data, because some of the auctions are from the same sellers. We therefore correct the standard errors of our logistic regression model based on Huber's procedure for clustered data (Huber 1981, as implemented in Stata). The base model with all observations shows a significant negative effect of seller reputation but the coefficient is very small: 10 extra reputation points lead to an decrease in the probability of sale of 0.5 percentage points. We think this negative effect largely arises because of high volume sellers using a strategy offering goods against relatively high reserve prices. If we include a dummy variable measuring whether the reputation score is larger than 50, then this variable has a substantial and significant negative effect (-1.06, p=0.035), the difference is more than 20 percentage points, which indeed corroborates the idea that high reputation sellers ('power sellers') tend to be more willing to wait until a buyer with a high need for the good comes along. It does not help much if a seller offers a headset or some other accessory, as can be seen from the absence of an effect of the number of items for sale. Including a picture helps (about a 20 percentage points increase if you have one included), and there is some indication that including more pictures increases the probability of sale further (p=0.06), but whether or not pictures are home made does not seem to matter. Moreover, whether the phone was used or new also has no effect. Compared to eBay.com the probability of selling a phone on any of the other auction sites is significantly lower. We find that during our data collection period, phones sell less well as time progresses. Note that we did not include the buyer reputation in this model. If we add the interaction of reputation with the site dummies, we find that the effects of reputation differ across auction sites, though the differences are not that large. Only on ePier.com the effect of reputation is positive (but not significantly different from zero) It is important to realize that sellers' reputation scores are higher on eBay.com than on the other sites. If we confine the analysis to the cases where reputation scores are smaller than 75, the differences between sites disappear. When we include the interactions of reputation with other variables, we find only one interaction effect of reputation that turns out to have a statistically significant effect: the positive effect of a picture increases with increasing reputation (or: the positive effect of reputation increases when a picture is present). For some other variables we have relatively large numbers of missing values (for instance for shipment costs and condition of item). Separate analyses show that neither of these show a significant relationship with probability of sale. Explaining End Price Next we consider the analyses on end price, using the same independent variables as in the previous analysis on probability of sale. The only exception is that we can now also include buyer reputation, as a measure for a buyer's experience in auctions. Obviously, we only consider auctions where the phone was eventually sold. As in the previous analyses, we control for the different

178

Chris Snijders/Richard Table 6

Zijdeman

Logit regression on whether a phone was sold.

Variable Seller reputation

Effect -0.0018**

Std. error 0.0008

Number of items

0.0488

0.1316

Home made picture included (dummy)

0.1642

0.3038

Picture included (dummy)

0.8364**

0.3659

Number of pictures included

0.1136*

0.0611

Function description included (dummy)

-0.0115

0.3141

Original box included

-0.0093

0.1504

Product is used (dummy)

-0.0946

0.3430

Ricardo.nl

-2.4622***

0.5195

ePier.com

-3.9085***

0.6293

eBay.nl

-2.5561***

0.4676

Auction's end date

-0.0048***

0.0012

Phone.categoryl

1.5937**

0.7494

Phone_category2

-0.2269***

0.4222

Phone_categroy4

-0.3189

0.5660

constant

77.0800

19.1986

*: p < .1, **: p < .05, ***: p < .001. Wald chi 2 (15) = 75.17, Prob.>chi 2 = 0.0000, standard error adjusted for clustering on sellers (N=416).

types of phones in the data. Closer inspection of the data shows that one of the assumptions of standard regression models, homoscedasticity, does not hold in our data. Loosely translated, the assumption that is violated in our data is that the regression line should fit about equally well for different values of the independent variables. For instance, we find that the model predicts much better for the cases where the seller has high reputation. We therefore estimate a model assuming multiplicative heteroscedasticity as in Harvey (1976). 2 Though this may appear to be a mere technicality, the results will show that precisely this issue, using standard regression models where this is not appropriate, is an important one. The basic idea of this alternative estimation method is that in the regression analyses one includes not only the predictors for the mean value of the dependent variable, but also predictors for its variance. The technical details are as follows. The estimated model is y, = E[yi] + e i with E[yi] = bo + biXu + . . - + bkXki, as in the usual regression model, but in addition we model logVar[ei] = a* E[yi](go + g\Z\i + ... + zmi). The e* are assumed to be standard normal. The b's, g's, and a are to be estimated. Roughly stated, the model is appropriate in cases 2

We thank Jeroen Weesie, who implemented the estimation of this model in Stata.

Reputation and Internet Auctions: eBay and Beyond

179

where it is reasonable to assume that the heteroscedasticity varies smoothly with predictors g. In our case the heteroscedasticity problem arises at least to some extent because the model fits better for less expensive phones. This is not so strange: it is easier to predict selling prices for phones of 100 dollars than it is to predict prices of phones of 500 dollars, simply because there is more room for error around 500 (taking the log of end price as the dependent variable does not solve this issue). In our base model, we also include the reputation score in the prediction of the variance in the residuals (the variable z\ then equals the seller's reputation score) and whether a product was used (the variable zi equals whether or not a product was used). The disadvantage is that we are now no longer able to control for clustering at the seller level, but given that the differences between the results with and without this option in previous analyses seem rather small, we will take that for granted. Table 7 shows the results of our analyses.

Table 7

Regression of end price, assuming multiplicative heteroscedasticity.

Variable Mean

Effect Seller reputation

0.0662*

Number of items

-10.6199**

Home made picture included (dummy)

0.03458 4.6258 9.4568

26.1201**

12.6432

Log of Number of pictures included

4.6822

10.6617

Buyer reputation

-0.1143

Picture included (dummy)

Function description included (dummy) Original box included

30.3045*** -4.5530

0.0977 10.2619 7.6250

Product is used (dummy)

-59.1183***

14.6938

Ricardo.nl

-50.1290***

12.5275

ePier.com

27.0364

18.4878

eBay.nl

-21.5475

14.3296

Auction's end date Phone.category 1

0.2461*** -51.6195***

0.0515 10.37952

P hone-category 2

-8.5019

17.64102

Phone.category4

48.1995

32.64756

constant Variance

-3.7688

Std. error

Seller reputation

-3641.094*** -0.0035***

802.9132 0.0010

Product is used (dummy)

2.3786

0.6198

constant

2.7273

1.2824

constant 0.0060 Alpha 0.0276 *: p < .1, **: p < .05, ***: p < .001. Multiplicative Heteroscedastic Regression, estimator: mle/Newton-Raphson, Wald chi 2 (19) = 2444.905, P r o b . > c h i 2 = 0.0000, corr2 fa, y fr) = 0.33, vwcorr 2 (y, y/t) = 0 . 5 7 ( N = 206)

In our base model, we now find that there is a positive effect of reputation

180

Chris Snijders/Richard

Zijdeman

on end price. This is all the more noteworthy because without controlling for heteroscedasticity we do not find such an effect. On average, every 10 additional reputation points yield an extra 60 dollar cents. People attending our presentation in Bielefeld at the Trust and community on the Internet conference of August 2003 might remember that we did not control for heteroscedasticity in the analyses we presented there and subsequently concluded—wrongly, as it turns out—that no effect of reputation exists. In fact, in none of the papers we considered in which no effect of reputation was found, heteroscedasticity is given any serious attention. For Diekmann and Wyder (2002) the issue is not likely to play a role, since they consider only one specific (new) phone and therefore a relatively homogenous set of end prices, and in fact they do find an effect of reputation (comparable in size). Reputation also plays a role in the variance component. It turns out that the variance in the residuals decreases with increasing reputation scores (controlling for the other independent variables, that is). This is again an indication that auctions from 'power sellers' are different from auctions of others. The image that pops up on the basis of our results is that the power sellers do not mind putting their goods up for auction as often as is necessary to get their wanted price, and that eventually their goods are indeed sold for around these prices. If we include the square of reputation in the predictors for the residuals, we get a significant quadratic effect of reputation with a maximum around reputation score 290. Besides the effect of reputation on the residuals, we also find the expected positive effect of the mean of end price (a=0.028, z=4.63): variance in the residuals increases with price. The other independent variables show the following results. Including a picture increases the value with on average 26 dollars, but adding more pictures does not help, and whether the pictures include home made ones is also irrelevant. Adding a function description increases the value with 30 dollars (p=0.003), and when a phone is not new this decreases the end price with on average 59 dollars. Prices on Ricardo.nl are lowest (an average difference of 50 dollars), followed by eBay.nl (21 dollar lower end prices). eBay.com and ePier.com have the highest end prices, with a statistically not significant edge for ePier.com of 27 dollars. When we add interactions of reputation with site dummies, we find that the positive effect of reputation is largely due to the eBay.com cases. For ePier.com and eBay.nl the net effect of reputation is even slightly negative. This is another reason to be cautious about generalizing the usual eBay.com results. Adding other interactions of reputation shows small effects: the effect of reputation slightly decreases when pictures are used, and the effect of reputation slightly increases when used goods are considered. Several variables hardly ever approached significance or had too many missing values to run meaningful analyses and are not included in the analyses we present here: the interaction of seller reputation with buyer reputation, the means of payment, the size of the bid increment, reserve price, and shipment costs. When we compare these results with our hypotheses, we see that the following hypothesized effects are supported by the data: the probability of sale increases with the inclusion of a picture and more pictures is better, and the probability of sale varies across auction sites. The end price is higher when the seller has

Reputation and Internet Auctions: eBay and Beyond

181

a higher reputation, when a picture is included, when a function description is included, and when the item is new. Hardly any of the hypothesized interaction effects with reputation emerged.

9. Summary and Conclusion The general aim of this paper was to gain some insight into the mixed results found in previous research on the effects of reputation in online auctions. First we summarized and extended the arguments in previous papers on reputation effects, using the Trust Game as a metaphor for online interaction. We then tried to replicate four recent papers using data gathered from four different auction sites and also analyzed this data somewhat more extensively using the additional variables we collected. The results of the replications are rather different from the original results, not only with respect to the effects of reputation. The only replication that seems to work out well is that of the analyses by Eaton (2002), although also there not all results match. We considered several reasons for these differences, but none of these seem completely satisfactory. The first possible reason for differences between the papers was that there are considerable differences as to which variables are included in the analyses. This is why we took the trouble to collect (almost) all variables as used in these papers, plus a large set of potentially useful others. Moreover, the papers differed in the way in which reputation was operationalized. Our analyses show that neither of these possible reasons for differences between papers can explain away the differences completely. Especially the different reputation measures are so highly correlated that it does not seem to matter much which measurement of reputation one takes. Another potential reason for differences is that the products under study varied across papers. In particular whether goods are used versus new, or standard versus complex, might be of influence. For this we also do not find much corroboration in our data. However, we do find strong evidence suggesting that in previous papers some problems may have gone unnoticed with respect to heteroscedasticity in the data (when considering the end price). When analyzed with ordinary least squares techniques, which implies ignoring the heteroscedasticity, our data show no effect of reputation on end price, just as most of the other auction papers do. When we do control for heteroscedasticity, we find a positive and significant, albeit small, effect of reputation. To replicate this result we collected an additional set of auction data, consisting of 1152 eBay.com auctions of five kinds of Ipod mp3-players (only end price, reputation, and a few product characteristics were collected). In these data we indeed found similar results: end price significantly increases with reputation, but this is found only when one controls for heteroscedasticity in the data; otherwise one finds no effect. Hence, with respect to our original hypotheses we can say that there is some evidence in our data that reputation helps sales, in the sense that it may get you a higher price. A high reputation does not help to get the goods sold. For this latter finding there is in fact another explanation, namely that high reputation

182

Chris Snijders/Richard

Zijdeman

sellers tend to sell only for higher prices (through establishing a high minimum price). Some additional corroboration for this explanation comes from the fact that the variance in end price is smaller for sellers with a high reputation. We also found some variables that have thus far not often been included in the analyses. Adding a picture to the product description helps: goods are sold more often and sold goods sell for higher prices when a picture is included. We find no evidence of extra benefits from the picture being home made. Moreover, there is evidence of higher end prices when a function desription of the product is included. No differences between experienced and inexperienced buyers are found. The experienced buyers neither buy at lower prices (there is some evidence of this in the data, but not significantly so) nor do they value the effect of reputation differently than less experienced buyers. Whereas it does not matter whether a good is used or new for the probability of sale, used goods obviously sell at lower prices. The variance in end prices is larger for used goods than for new ones. We have only scratched the surface of what could be going on on auction sites and it certainly is a surprising finding that both model replication as well as deduction of hypotheses that can stand a comparison with the data is so difficult. Especially since predictions in auction research should be where social science excels: virtually all the information that is known to the user of an auction site is available to the researcher, making it a setup under quasi-experimental conditions. Certainly auctions should be easier to analyze than topics where the only data one can use are based on field studies in noisy circumstances, as so often is the case. Apparently, it is not so simple. We conclude with some suggestions for researchers who wish to analyze auction and reputation data. First, include other variables besides sale, end price, and reputation in your data collection. Especially whether pictures are included and whether a function description is included has an impact, but so might other variables. At this point we are still not sure enough about which issues matter most, and additional empirical evidence is necessary to get some guidance with respect to theory development. Second, be sure to consider the different ways in which reputation can be operationalized and check for large correlations between them. For instance, what you think is an effect of negative reputation could be an effect of positive reputation (given that these tend to correlate strongly). Third, think carefully about whether you want to analyze auctions of different goods simultaneously. There is evidence that this leads to hctcroscedasticity, which in turn can produce inadequate results when analyzed with standard regression analyses. You should check for heteroscedasticity in any case, since there is evidence that other independent variables, such as the reputation score of the seller, correlate with the fit of the model. Fourth, think about analyzing the data, or perhaps even designing theory, separately for the occasional user and the professional user of an auction site. There is strong evidence that both groups show different behavior. Finally, some extra attention should be paid to the differences with respect to the auction sites themselves. Although we do not have enough data to make strong claims here, there are differences that are worthy of further investigation. Probability of sale and end prices differ between

Reputation

and Internet Auctions:

eBay and Beyond

183

the auction sites we considered (eBay.com, eBay.nl, Ricardo.nl, and ePier.com) and it is not clear whether these differences would disappear if one would control for supply and demand characteristics. Perhaps more important is that the site on which the effect of reputation seemed to work the strongest was eBay.com. The emphasis of researchers on eBay.com is logical, since it is by far the largest auction site, but given our results some caution as to whether these results carry over to other sites is in order.

Bibliography Akerlof, G. A. (1970), The Market for 'Lemons': Quality Uncertainty and the Market Mechanism, in: Quarterly Journal of Economics 84, 488-500 Ba, S./P. A. Pavlou (2002), Evidence of the Effect of Trust Building Technology in Electronic Markets: Price Premiums and Buyer Behaviour, in: MIS Quarterly 26, 43-268 Buskens, V. (1999), Social Networks and Trust, Amsterdam Coleman, J. S. (1990), Foundations of Social Theory, Cambridge Dasgupta, P. (1988), Trust as a Commodity, in: Gambetta, D. (ed.), Trust. Making and Breaking Cooperative Relations, Oxford, 49-72 Dellarocas, C. (2003), The Digitization of Word-of-Mouth: Promise and Challenges of Online Reputation Mechanisms (March), MIT Sloan Working Paper No. 4296-03. http://ssrn.com/abstract=393042 Dewan, S./V. Hsu (2001), Trust in Electronic Markets: Price Discovery in Generalist Versus Specialty Online Auctions, Working paper (http://databases.si.umich.edu/reputations) Diekmann, A./D. Wyder (2002), Reputation und Vertrauen bei Internet-Auktionen, in: Kölner Zeitschrift fur Soziologie und Sozialpsychologie 54, 674-693 Eaton, D. H. (2002), Valuing Information: Evidence from Guitar Auctions on eBay, Working paper, Department of Economics, University of Arizona (http://databases.si.umich.edu/reputations) Harvey, A. C. (1976), Estimating Regression Models with Multiplicative Heteroscedasticity, in: Econometrica 44i 461-465 Houser, D./J. Wooders (2000), Reputation in Auctions: Theory and Evidence, Working paper, Department of Economics, University of Arizona (http://databases.si.umich.edu/reputations) Huber, P. J. (1981), Robust Statistics, New York Kauffman, R. J./C. Wood (2000), Running up the Bid: Modeling Seller Opportunism in Internet Auctions. Working paper, presented on America's Conference on Information Systems (AMCIS), Long Beach Kreps, D. M. (1990) Corporate Culture and Economic Theory, in: J. E. Alt/K. A. Shepsle (eds.), Perspectives on Positive Political Economy, Cambridge, 90-143 Lee, Z./I. Im/S. J. Lee (2000), The Effect of Negative Buyer Feedback on Prices in Internet Auction Markets, in: Proceedings of the 21st International Conference on Information Systems, Brisbane, 286-287 Melnik, M. I./J. Aim (2002), Does a Seller's eCommerce Reputation Matter?, in: Journal of Industrial Economics 50, 337-349 Raub, W./C. Snijders (2001), A Reluctant Match: Models for the Analysis of Trust in Durable Two Party Relations, in C. Castelfranchi/Y.-H. Tan (eds.), Trust and Deception in Virtual Societies, Amsterdam, 195-220

184

Chris Snijders/Richard

Zijdeman

Resnick, P./R. Zeckhauser (2002), Trust Among Strangers in Internet Transactions: Empirical Analysis of eBay's Reputation System, in: M. R. Baye (ed.), The Economics of the Internet and E-Commerce. Volume 11 of Advances in Applied Microeconomics, Amsterdam, 127-157 — / — / J . Swanson/K. Lockwood (2002), The Value of Reputation on eBay: A Controlled Experiment. Working paper, available online from the Reputations Research Network (http://databases.si.umich.edu/reputations) Roth. A. E./A. Ockenfels (2000), Last Minute Bidding and the Rules for Ending SecondPrice Auctions: Theory and Evidence from a Natural Experiment on the Internet, in: American Economic Review 92, 1093-1103 Snijders, C. (1996), Trust and Commitments, Thela Thesis, Amsterdam — /V. Buskens (2001), How to Convince Someone that You can be Trusted? The Role of 'Hostages', in: Journal of Mathematical Sociology 24, 355-383 van Swol, L. M. (2001), EBay and the Return of the Town Square: How eBay uses Reputational Gossip to Build Trust, Working paper, Evanston Zijdeman, R. L. (2003), Reputation and Internet Auctions. EBay and Beyond, Master's Thesis, Utrecht University, ISCORE paper no. 214

Analyse & Kritik 26/2004 (© Lucius & Lucius, Stuttgart) p. 185-202

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

Trust among Internet Traders A Behavioral Economics Approach*

Abstract: Standard economic theory does not capture trust among anonymous Internet traders. But when traders are allowed to have social preferences, uncertainty about a seller's morals opens the door for trust, reward, exploitation and reputation building. We report experiments suggesting that sellers' intrinsic motivations to be trustworthy are not sufficient to sustain trade when not complemented by a feedback system. We demonstrate that it is the interaction of social preferences and cleverly designed reputation mechanisms that solves to a large extent the trust problem on Internet market platforms. However, economic theory and social preference models tend to underestimate the difficulties of promoting trust in anonymous online trading communities.

1. Introduction: Trust on Internet Market Platforms In a traditional market place, t r u s t plays little role in t h e exchange of a good of recognizable quality. Being physically together in time and space allows buyers t o inspect t h e item before buying, and permits a 'simultaneous' exchange of money and object. T h e situation is quite different in computer-mediated online m a r k e t s such as eBay's auction house and Amazon's market platform for used books. Transactions on these platforms are characterized by asynchronous (sequential) actions of anonymous traders, operating at spatially disperse locations. In such a setting, trust becomes a critical issue. It is not just t h e scattering of t r a d e in space and time t h a t pose a challenge to Internet exchange, it is also t h e medium of communication per se. C o m p u t e r mediated communication makes it more difficult to signal trustworthiness and to promote cooperation t h a n 'richer' communication media such as face-to-face communication (Prank 1988; Brosig et al. 2003). Trading via computer networks also allows buyers and sellers to freely choose a trader identity, different from one's t r u e 'offline identity', t h a t can be changed, often with ease. In addition, lasting personal relationships on Internet market platforms are infrequent; unlike in traditional market environments, one-shot interaction is t h e rule r a t h e r t h a n t h e exception (Resnick/Zeckhauser 2002). Thus, cyberspace makes it particularly difficult to develop social and economic bonding t h a t s u p p o r t s t h e emergence of t r u s t and trustworthiness in more traditional markets. 1 * We thank Bernd Lahno for very helpful comments. Ockenfels gratefully acknowledges the support of the Deutsche Forschungsgemeinschaft through the Emmy Noether-program. Bolton and Katok both gratefully acknowledge the financial support of the National Science Foundation. 1 There are more difficulties. For example, unlike in traditional auction markets, such as

186

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

To deal with the problem, many online market platforms implement electronic reputation mechanisms that collect, process, and distribute large amounts of information about past trading activities of the market participants. These reputation mechanisms attempt to emulate traditional word-of-mouth networks and thus promote trust and trustworthiness among strangers. On eBay's 'feedback forum', for instance, market participants continuously assess and reassess their transaction partners. Figure 1 provides a sample of the type of feedback that is left on the site. Each user's assessment is made public and so forms his individual reputation that others can condition their behavior on. As eBay's founder Pierre Omidyar writes in a letter to all users: "By creating an open market that encourages honest dealings, I hope to make it easier to conduct business with strangers over the net. Most people are honest. And they mean well. Some people go out of their way to make things right. I've heard great stories about the honesty of people here. But some people are dishonest. Or deceptive. ... It's a fact of life. But here, those people can't hide. We'll drive them away. Protect others from them. This grand hope depends on your active participation. Become a registered user. Use our Feedback Forum. Give praise where it is due; make complaints where appropriate." http: / / pages .ebay.com/community / news/founders-letter, html In this paper, we describe a simple experimental game that captures the key problems of trust and trustworthiness on Internet market platforms from an economics point of view. We present new experiments and survey other evidence indicating why and when people do or do not trust. We also compare the empirical evidence with economic theories of decision making and institutional design. The two critical findings from this work that we will emphasize are, first, that understanding trust and trustworthy behavior requires us to consider a mix of human motives, no one motive being sufficient, and second, that the expression of these motives is sensitive to the institutional and strategic environment. The important implication is that trust and trustworthy behavior is neither wholly a matter of social norms and morality, as it is sometimes popularly assumed, nor entirely a matter of institutional design, as economists sometimes assume. Understanding how trust and trustworthiness is, and can be, sustained in an environment like the Internet will require a coming to grips with the interplay between the two. The paper is organized as follows: In section 2, we motivate and describe the game. In section 3, we analyze one-shot encounters of the game from the point in Christie's and Sotheby's auction houses, transactions among private individuals conducted on online market platforms such as eBay and Amazon are typically not secured; that is, eBay and Amazon neither attest to the credibility of the private sellers nor do they guarantee item quality. Also, selling on the Internet is typically cheap, so that signaling trustworthiness through 'fancy store fronts' is not possible (although fancy homepages or elaborately designed offers may well signal professional commitment and competence). Finally, formal contracting and legal enforcement is more costly (at least relative to the item's value) than in traditional markets.

Trust among Internet Traders: A Behavioral Economics Approach An Feedback Recehred

From & * e r s

Fmrn Sutlers

187

Lett for Other«

14 feedback received by bsofensge from buyers Comment

From

0

Fajtdelivety and veiy pleased with the product - Thank you very much

Buyer eumlOOt ( 8 )

Q

beautiful vasB, fast shipment, recommendl

Buyer nageras ( 25Q A

0

Excellent transaction, smooth shipping. Item as described

Buyer jb a uer?< (139 £

Q

Oustanding packaging-some nice old Chinese antlctue on this e 9 a y - t h a n k s = A * * * * * * * *

Buyer rfwavnalnftis (ÌLÙ i r

Ç ) veiyfastandresponslble

) )

Buyer S70305 ( 2 )

Q

Great communication and always responded to e-mlals. Preffered Seller. A * * * * * * *

Buyer chemlcalsS ( l à & )

Q

Item recleved as descrtbed.Thanks

Ouyer rartanv f i "iq -¿f )

Q

Arrived faster than I had expected, good transaction

Buyer hetz8888 ( S )

0

Hianksl Good and friendly seivlcel

B

Q

wonderful selieri Great Items, professionally p a c k e d ! rapidly s h i p p s d / A + « « * *

B uyer

Q

Wired money, didnt receive merchandise. This seller is bogus and a

ripoff,

w

rj/nnasn ( 3 B

ft)

nekaptah ( ¿ 5 2 & )

Buyer mdavne ( 1 £ ft )

Reply by baofengge responsible.mv account has no money,I contact the bank end buyer uninterrupted 0

Seller was extremely helpful and went beyond expectations to deliver the product

Buyer | 3 f r i m 0 r e (>j2 & )

Q

Well-packed. Fast international delivery well-received. A pleasant transaction.

Buyer kmnesteele-smith ( S )

©

GREAT CARVING FAST SHIPPING A L S O NICE F R E E OIFT A A A A A A A A A A * * * * * * * * * * *

Buyer onentallrut M 91

ftl

Figure 1: An example of buyer feedback at eBay (left for a seller of Chinese antiques) of view of recent behavioral theories that seek to explain when trust and trustworthy behavior will arise. In section 4, we focus on institutional explanations of how trust and trustworthiness can be created even if traders are ultimately guided by their material self-interest; these explanations all pivot on repeated encounters. Section 5 summarizes the findings.

2. The Trust Dilemma Typically, in online trading, a seller posts an offer on the market platform that includes a description of the item and its condition, and a price at which he is willing to sell. Then, a willing buyer sends the money to the seller. Finally, upon receiving the money, the seller is supposed to ship the item to the buyer. 2 With no outside control, having received the money, the seller can profit from not sending the item or sending poorer quality than promised. Anticipating this moral hazard, buyers may not be willing to buy. As a consequence, trading that would make everybody better off would not take place. This is the essential trust dilemma that online markets need navigate. The game-tree in figure 2 captures this dilemma (for the moment, ignore the number in the brackets). Both the seller and the buyer are endowed with 35, which is the payoff when no trade takes place. The seller offers an item for sale at a price of 35 which has a value of 50 to the buyer. The seller's cost of providing the buyer with the 2

Here we assume that the seller fixes the price, such as on Amazon's used book market. However, all arguments in this paper hold equally if the price is endogenously determined, such as in eBay's auctions (in this case the auction winner is the buyer).

Gary E. Bolton/Elena Katok/Axel

188

Ockenfels

item - costs associated with executing the trade, shipping, handling, as well as production costs - is 20. So each successfully completed trade creates a consumer surplus of 15 and a net profit of 15 for the seller. If the buyer chooses to buy the item, he sends 35 to the seller, who then has to decide whether to ship the item. If the seller does not ship, he receives the price plus his endowment of 35 for a total of 70. If he ships, he receives the price minus the costs plus his endowment for a total of 50. If the buyer chooses not to buy the item, no trade occurs.

Buyer

buy [27%] Seller

not buy

skip [37%] Buyer earns Seller earns

50 50

0 70

35 35

Figure 2: The basic trust game The game captures the critical features of online trade. First, traders move sequentially because they are dispersed in space and time. Second, the seller's pecuniary motive dictates to keep the money along with his endowment and to not ship the item. In this case, the buyer would lose his endowment and end up with nothing. If the buyer anticipates that the seller does not ship, he will not buy and thus no transaction will take place. This is the prediction of economic theory assuming that traders are only guided by pecuniary concerns. Third, interaction is anonymous. In particular, the name attached to the seller does not reveal information about his trustworthiness. 3 Fourth, the trust-game is one-shot, meaning that there is no common history or common future among traders that could give them the opportunity to reward or punish each other for past behavior or to develop other kinds of social or economic bonding with each other. (As we will see later, however, even one-shot interactions can be strategically linked via information flows.) And fifth, since legal enforcement is typically too expensive to be a credible threat, prosecution when the item is not shipped is not an option in the game. 4 In the next sections, we discuss 3 Trustworthiness is no issue for known reputable sellers such as barnesandnoble.com. But we will concentrate here on trading among anonymous individuals on market platforms with large numbers of buyers and sellers. 4 The game abstracts away a couple of features that may be relevant to online markets. For instance, we assume that there is no uncertainty about the item's quality (and so there is no reason to have a first game stage in which the seller describes the item's condition). However, the seller's strategy 'not ship' can easily be reinterpreted as 'lying about the item's true quality'. Also, in principle one can easily introduce legal enforcement and courts (see, e.g., Giith/Ockenfels forthcoming, for an example), for, say, high-stake transactions.

Trust among Internet Traders: A Behavioral Economics

Approach

189

experimental evidence using the basic trust game as a vehicle for studying trust and trustworthiness.

3. Trust in One-Shot Games: Behavioral Explanations 3.1 T h e N a t u r e of Trust (a) Distributional

aspects

Prom the point of view of economic analysis, one can hardly think about a more trust-unfriendly environment than online market platforms; yet, many of these platforms flourish. So what makes these platforms work? One possible answer is that our basic trust game does accurately describe the market institution but that people behave in a different way than the economic analysis suggests. In particular, it is sometimes argued that people are not as selfish as economists typically assume, and that traders care about morals. In fact, in trust games and related anonymous one-shot games (like the prisoner's dilemma game and the ultimatum game), psychologists, sociologists, experimental economists and others have identified a couple of non-pecuniary motives that may drive behavior. Most prominently and actively discussed in the recent economics literature are concerns for fairness (Fehr/Schmidt 1999; Bolton/Ockenfels 2000) and reciprocity (Rabin 1993; Dufwenberg/Kirchsteiger forthcoming). Some authors also put forward an individual concern for efficiency (Charness/Rabin 2002). All these social preference models assume that traders care about their own monetary payoff but that some traders may additionally be concerned with the social impact of their behavior. Reciprocity models conjecture that people tend to be kind in response to kindness and unkind in response to unkindness, while fairness models posit that some individuals may have a preference for equitably sharing the efficiency gains from trade. That is, in our trust game, reciprocity models suggest that a seller ships because the buyer was so kind to buy, whereas fairness models suggest that he ships because otherwise the payoff distribution would be unfair. Models of efficiency striving behavior, finally, suggest that traders care about the sum of payoffs. That is, a seller in our trust game may ship because this increases the pie to be distributed—independent of reciprocal and distributive fairness. One important implication of all these models is that it is not the buyers' social concerns that drive the trading activities but the sellers' social motives. As long as the sellers are strictly selfish and rational, there will be no shipping and thus no buying by rational buyers. Even a perfectly fair-minded buyer, always striving for equitable outcomes, would prefer 35 for both players over 70 for the seller and zero for himself (this holds, e.g., regardless of whether we apply Fehr and Schmidt's or Bolton and Ockenfels' model here). Similarly, in reciprocity models, a buyer is only willing to buy if the seller reciprocates by shipping (again, this holds regardless of the specific reciprocity model). Finally, a buyer concerned about efficiency cannot increase efficiency by sending money to a seller who is unwilling to materialize the efficiency gain by completing the

190

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

trade. Thus, the only plausible reason to buy in our basic trust game is that the buyer sufficiently strongly believes that the seller will ship. By this view, the preferences of the buyer are largely irrelevant: shipping in our basic trust game induces a selfish, a reciprocal, a fair-minded or an efficiency striving buyer to trust. At first glance, the statement that trust is the belief that the seller ships (with a sufficiently high probability) may sound like a trivial statement, but in fact it is highly sensitive to the details of the trust game. To see why, let us consider a very simplified 2-trader version of "ERC", the fairness model developed by Bolton and Ockenfels (2000). In this model, each trader i is motivated by his pecuniary income yi (equivalent to the payoffs given at the end nodes of the trust game tree) but also prefers fair over unfair final payoff distributions. A payoff distribution is perceived as unfair if a trader either earns more or less than the trading partner, that is, in the simplest form, if (yi — yj)2 > 0. 5 Traders motivated this way can be characterized by a 'motivation function' or utility UiiVuVj) =Vi~

r

i{Vi - Vj)2

Utility is increasing in pecuniary income but decreasing in the inequality of income distribution. Not all traders suffer equally from unfairness; the parameter Ti > 0 measures the trade-off of trader i between the (sometimes) competing motives. Setting rl =0 yields the standard economic model of complete selfishness. The distribution of r^ in the population is supposed to follow some cumulative distribution function F(ri). In our basic trust game, a buyer i will only trust if his expected payoff from trusting, p(50 — 0r t ) + (1 — p)(0 — r,(—70) 2 ), is larger than his certain payoff from not trusting, 35, where p is the probability that a trusted seller will ship. 6 Equivalently, the condition for trust is: ri < (50p - 35)/(4900(l - p)). That is, buyer i is only willing to buy if shipping is likely (p is large) and if he is sufficiently selfish, that is, if rl is sufficiently small. Why? The buyer can avoid being exploited by the seller by not buying the item. Thus, if fairness (or exploitation aversion) is the main motive, he should not buy. He will only run the risk of being exploited and buy the item when the expected material gain of trade is sufficiently large, and if he is sufficiently motivated by the material gain relative to his inequity aversion. In other words, in our basic trust game, pecuniary motives drive trust. Buying is motivated by trust in shipping, not by non-pecuniary incentives. While the simple model we present here oversimplifies certain complexities (and the general ERC model addresses some of these complexities), it sharply 5

The general ERC model assumes that a trader may care about his own relative standing in a reference group, but the model formulation we use here works equally well in this particular 2traders setting. Also, the general ERC model does not restrict the functional form of inequality aversion in any way. In particular, it is not restricted to be linear, nor quadratic, nor symmetric around equality. 6 A seller j will ship if 50 > 70 - r,(70) 2 , or equivalently rj > .004; thus, p = 1 - F(.004).

Trust among Internet Traders: A Behavioral Economics

Approach

191

explains why the act of buying can be identified as trust in our basic trust game and is not confounded by other motives. By the same token, what has been called trust in other standard trust games studied in the literature may have little to do with trust in a reciprocal response. In sequential prisoner's dilemma games or in the well-known investment game by Berg et al. (1995) cooperating, respectively investing, increases efficiency, so that trusting may theoretically reflect a concern for efficiency. The following example demonstrates the difficulties of identifying trust when the basic game is even only slightly changed:

Figure 3: The asymmetric trust game The asymmetric trust game in Figure 3 is the same game as our basic trust game in Figure 2 with the exception that the buyer gets an additional payment of 70 regardless of the game outcome. While this small change has no effect on the standard economic analysis based on (marginal) pecuniary incentives, it may have a big effect on behavior in a world with fairness. Assume that some traders care about equitable outcomes as described in the model above. This model predicts that no seller in the asymmetric trust game, regardless of his Tj, will ever ship, because not shipping maximizes pecuniary outcomes and is equitable. 7 That is, this model predicts that there is no shipping, regardless of the extent to which the seller population is selfish or fair. The interesting point is that even if he anticipates that there is no shipping, a fair-minded buyer may still want to buy. By buying he sacrifices pecuniary payoff in exchange for a fair distribution. 8 In particular, a buyer i prefers to buy if 70 > 105 - rj(70) 2 , or equivalently if rl > .011. However, this buying cannot be interpreted as trust in shipping: no seller will ship and all buyers know that. Buying, in the asymmetric trust game of Figure 3, might, however, be motivated by fairness. Confounding motives for trust like this are largely avoided in the basic trust game. (b) Procedural aspects 7 Shipping yields for the seller 50 — rj(50 — 120) 2 ; not shipping yields 70 which is greater for all Tj > 0. 8 These effects have also been studied in the so called dictator game (see, e.g., Forsythe et al. 1991).

192

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

Trust necessarily involves the risk of being exploited. An interesting question is whether the risk of a bad outcome is treated differently when it is due to uncertainty of the seller's morals or whether it is just determined by a chance move (e.g., there is always a risk that the item is accidentally lost or damaged in the mail). Standard economic theory does not distinguish between different sources of risk when the probabilities for the different states of the world are equivalent. Bohnet and Zeckhauser (forthcoming) demonstrate in an experiment, however, that the risk is perceived differently in a social context than in a non-social context, something that is not captured by any of the social preference models mentioned so far. The game they study is similar to our basic trust game, but the shipping decision is made, not by a seller, but a chance move. The following figure illustrates their game within our simple framework:

Buyer earns Seller earns

50 50

0 70

35 35

Figure 4: The random-shipping game In their study, Bohnet and Zeckhauser (forthcoming) asked each buyer for the probability a such that he is indifferent between buying and not buying. They also asked each buyer in the basic trust game (where the sellers determine the shipping probability a endogenously) which probability of being matched with a trustworthy seller makes him indifferent between buying and not buying. By using an incentive-compatible experimental design, Bohnet and Zeckhauser could compare how much risk buyers are willing to take if the source of the risk is the morals of the sellers, as in the basic trust game, to the amount of risk they are willing to take in a lottery, as in the random-shipping game. In fact, Bohnet and Zeckhauser find that "individuals are much more willing to take risks when the outcome is due to chance, as opposed to an equivalent-odds situation where the outcome depends on whether another player proves trustworthy. Taking a chance on the latter risks incurring betrayal costs, costs shown to be above and beyond mere monetary losses." A related effect has been observed by Blount (1995) in the context of bargaining games. The same offer was more acceptable to a responder when determined by a lottery than when determined by a (human) proposer. A plausible interpretation for these effects is based on a notion of procedural fairness that has been first put forward and tested by Bolton, Brandts and Ockenfels (2003). The idea is that an unfair outcome chosen by a fair lottery is more acceptable than the same unfair outcome chosen by another trader. This notion of procedural fair-

Trust among Internet Traders: A Behavioral Economics

Approach

193

ness is not only consistent with Blount's bargaining experiment but also with Bohnet and Zeckhauser's trust game study. It is, however, inconsistent with both standard economic models and social preference models in behavioral economics that are all invariant to the fairness of procedures (see, however, Bolton et al. 2003, for a sketch of a model that extends the social utility approach to procedures). 3.2 T h e Nature of Trustworthiness (a) Distributional

aspects

Trust in shipping drives buying. But what drives shipping? As mentioned before, all social motives described so far—fairness, reciprocity and efficiency— may trigger shipping. Here we examine some empirical evidence on this issue. Let's look first at the basic and the asymmetric trust game in Figures 1 and 2. Buying can, for both games, be equally interpreted as 'kind' behavior and thus, since the seller payoffs are unchanged, reciprocity models would predict no difference of shipping probabilities across the two games. 9 Furthermore, observe that efficiency-seeking behavior would also make no distinction, because an additional endowment to the buyer in the asymmetric trust game does not affect the amount of potential efficiency gains across outcomes. As explained before, however, fairness models predict that while some sellers may be willing to ship in the basic game, nobody is willing to ship in the asymmetric game. Thus, while reciprocity and efficiency models predict no change in seller behavior, fairness models predict less shipping in the asymmetric trust game. We studied these games in classroom experiments. 10 While in the basic trust game 37 percent of the 30 sellers were willing to ship, in the asymmetric trust game only 7 percent shipped (see the numbers in the brackets in Figures 1 and 2). Thus, the data highly significantly (p < .01) reject reciprocity and efficiency as the predominant explanation of trustworthiness; fairness, on the other hand, is largely confirmed (though the two sellers who ship are inconsistent with the fairness motive but consistent with both other motives considered here). Interestingly, even though far fewer sellers are willing to ship in the asymmetric game, significantly more buyers are willing to buy (27% in the basic trust game and 46% in the asymmetric trust game; p < .05). This suggests that fairness, and not trust in shipping, is the main driver for buying in the asymmetric trust game. Similar conclusions about the impact of distributive fairness preferences in trust and related games are drawn by a number of other experimental studies. 11 9 See, for example, Bolton and Ockenfels forthcoming. Since the intuition is quite plausible, we will make no attempt here to go more into the details of the reciprocity models' mechanics, which can be quite complex. 10 In this and the other one-shot experiments, we have thirty independent observations, respectively, for each choice separately (buying and shipping). Each subject played only one of the studied games in only one role (buyer or seller). For sellers we applied the strategy method, that is each seller was asked whether he would ship if the buyer chose to buy. Half of all encounters were randomly chosen to be paid out according to the rules of the game. A payoff of ten in the game tree translated to 1 Euro cash. 11 See Bolton/Ockenfels 2000 and forthcoming and the references cited therein. More re-

194

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

(b) Procedural aspect To what extent do sellers condition their behavior on the history of play? The following game is the same as the basic trust game with the exception that the buying decision is made by a random procedure rather than by the buyer himself. From the view of distributional fairness and standard economic models, nothing has changed for the seller, who still has exactly the same options over payoff distributions. From the view of reciprocity models, however, since the buyer has no choice, buying cannot be attributed to the kindness of the buyer, and thus sellers have no reason to be 'kind in response'.

probability 1/2

Nature

Seller ship [33%] Buyer earns Seller earns

probability 1/2 \

not ship

50 50

Figure 5: The random-buying game In a classroom experiment study of the random-buying game, the shipping probability was with 33% only slightly (and not statistically significantly) smaller than in the basic trust game (Figure 5). This indicates that distributional concerns play a major role while the history of play, that is 'intentionality' and lottery procedures, only seem to play a minor role in determining trustworthiness. This result too is in line with a number of earlier studies. In particular, the first experimental economics study of distributional versus reciprocity-based preferences in Bolton, Brandts and Ockenfels (1998) suggested, in the context of a sequential two-players cooperation game similar in the incentive structure as the one we study here, that the second mover's behavior too was invariant to the history of play and to the 'intentions' of the first mover. A number of subsequent studies (e.g., Brandts/Charness forthcoming; Giith et al. 2001) confirmed this invariance property of behavior. Charness and Rabin (2002) distinguish negative reciprocity games, games where other-regarding behavior has a punishment flavor to it (such as ultimatum and other bargaining games), from positive reciprocity games, games where other-regarding behavior has a reward flavor to it (such as dictator and trust games). Attribution-based behavior was found to be significant only in negative reciprocity games (also see Bolton et al. 1998; Offerman 2002). Some studies conclude that the response to fair or unfair behavior is substantially invariant (Blount 1995; Dufwenberg et al. 2001) cently, a couple of authors also found evidence for efficiency-seeking behavior in simple, nonstrategic payoff allocation games—games that however cannot easily be related to trust problems (Charness/Rabin 2002; Engelmann/Strobel forthcoming; see Bolton/Ockenfels 2003, for a discussion of some of these results).

Trust among Internet Traders: A Behavioral Economics

Approach

195

or only weakly sensitive (Charness forthcoming) to whether the action is taken by a disinterested third party or by the actual game partner. Overall, it seems fair to conclude from the behavioral economics literature that, at least for the kind of trust games that we are considering here (positive reciprocity games), distributional fairness is a major motive in virtually all studies, while reciprocity and intentionality are sometimes found to be a secondary concern. 12 Summing up, our basic trust game is an appropriate vehicle to study the motives underlying trust and trustworthiness among online traders. There is no trust without trustworthiness. Trustworthiness appears to be mainly driven by distributive fairness concerns. Risk, a key ingredient of trust, is treated differently according to the procedure by which the uncertainty is resolved. On the other hand, trustworthiness appears to be largely invariant to the history of play (e.g., intentions and lottery procedures).

4. Trust in Repeated Games: Combining Behavioral with Institutional Explanations In expected monetary terms, the probability of a trustworthy seller needs to be at least 70% in order to make buying in the basic trust game profitable. In none of our one-shot games, was the shipping probability substantially greater than half the required amount. Thus, if trust rests solely on behavioral propensities, such as social preferences, there is little hope that trust among strangers can be stable. 13 However, when we embed the trust game in a repeated setting, the interaction of social behavior with institutional design such as reputation mechanisms can multiply the impact of the 'truly intrinsic' trustworthiness, because cleverly designed institutions can create strategic incentives to be trustworthy even for selfish and rational sellers. In this section we discuss these incentives, relate them to online market platforms and study their empirical content. Suppose, for the moment, that the same seller and the same buyer play the basic trust game repeatedly, with an infinite time horizon, and so have no expectation of a stopping round of play. In such a setting, even if all traders are completely selfish and rational, there exist equilibria in which the buyer always buys and the seller always ships. The equilibria can be supported by triggerstrategies that call for a buyer, for instance, to trust as long as the seller ships. If the seller defects only once, he will never be trusted again. If future payoffs are sufficiently important, this gives an incentive to the seller to be trustworthy all the time, and thus for the buyer to trust all the time (Kandori 1992). There are, however, two problems with this kind of simple equilibria in our 12 We note that the fact that the implementation of a lottery affects behavior in the randomshipping game but not in the random-buying game is in line with Bolton et al.'s (2003) concept of procedural fairness. In the random-shipping game, an unfair outcome chosen by a fair lottery may be perceived as procedural fair; however, in the random-buying game an unfair outcome chosen by the seller cannot be justified as procedural fair within their model. 13 While, of course, payoffs, framing and context are different on computer mediated market platforms compared to our simple classroom game, we believe that an analogous conclusion also holds for actual online market platforms.

196

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

context. First, the trading horizon in online market platforms is typically finite. If either the buyer or the seller believes that there will be some upper boundary of items to be traded, cooperation among completely selfish, rational traders will unravel. Second, buying and shipping in infinite game equilibria do not capture trust and trustworthiness because, in equilibrium, there is no uncertainty about each others' behavior; in equilibrium, sellers can have a material incentive to ship, so there is no risk of being exploited. 14 Since this paper is concerned with trust (characterized by a risk of being exploited) in online markets (where traders trade a finite number of items), we prefer to study finitely repeated games. In finite games, however, trust may only emerge when there is some (possibly small) amount of 'truly intrinsic' trustworthiness within the seller population (Wilson 1985). 15 That there is intrinsic trustworthiness has been demonstrated in our experimental studies of the one-shot trust game. 4.1 Strangers' Market Suppose the market transactions take place over a series of rounds. At the beginning of each round, a potential buyer is matched with a potential seller and they then play the basic trust game in Figure 2. In the simplest version of such a repeated setting, each game is played with a different transaction partner 1 6 and no information about trade outcomes leaks from one encounter to another one. This market is basically a sequence of one-shot games. Thus, because there is not sufficient intrinsic trustworthiness to make trust profitable, we hypothesize that all trading activities will collapse.

Buyer

buy [37%] Stranger-seller

not buy

ship [39%] Buyer earns Seller earns

50 50

0 70

35 35

Figure 6: Average behavior in the Strangers market 14 There are also more subtle equilibria in which cooperation in any given round is uncertain, but, in our view, trust is also not satisfactorily described as an equilibrium selection problem. 15 If not, all trust and trustworthiness would unravel. We will not make an attempt to show the details of the equilibrium analysis of trust and trustworthiness in finitely repeated games in a world with fairness. The mechanics of these equilibria are relatively complex (see Bolton/Ockenfels 2004, for a theoretical and experimental treatment within a trust game environment). However, we note here that these equilibria in fact capture the risk of being exploited. 16 In fact, empirical evidence suggests that buyers and sellers on Internet market platforms such as eBay only occasionally come back to the same trading partner (Resnick/Zeckhauser 2002).

Trust among Internet Traders: A Behavioral Economics Approach

197

In an experiment by Bolton, Katok and Ockenfels (forthcoming), we had subjects play in precisely this format, over 30 rounds. No buyer met the same seller more than once, so we call this experimental treatment the Strangers market. All interaction was computer-mediated and anonymous; subjects sat in cubicles in front of computers not knowing the true identity of their trading partners, capturing an important aspect of online trading. The rules and that all rounds were actually paid was common knowledge. Figure 6 shows the average buying and shipping (conditioned on buying) behavior over all rounds. There is almost the same amount of trustworthiness as in the one-shot version of the trust game in Figure 2. This reflects that the Strangers market does not create additional incentives to be trustworthy as compared to the one-shot game. On the buyer side, there is, on average, more trust in the Strangers market than in the one shot version of the game, possibly reflecting the hope that repeated action may support more cooperation. But the dynamics reveal that buyers respond to the fact that, on average, this expectation was disappointed: They started out by trusting quite a lot but trust quickly collapsed. In fact, the percentage of last round trust was only 0.04%, much less than in the one shotgame, indicating that buying in the one shot game is mainly due to inexperience. 4.2 Reputation Market From an economic theory perspective, reputation mechanisms of the sort we describe in the introduction can solve many of the trust problems associated with the Internet. In fact, simple theory suggests that information is the key prerequisite to the emergence of trust, because it allows buyers to avoid sellers who are known as fraudulent and to buy only from sellers who have been proven to be trustworthy in the past. Conditioning trust on the seller's history creates incentives for sellers to build up a reputation for being trustworthy, at least when the end of the market is not too close and maintaining a good reputation is still valuable. A reputation of being trustworthy can be developed and sustained even by completely rational and selfish sellers—as long as the probability of being matched with intrinsically trustworthy sellers is strictly positive. That intrinsic trustworthiness exists is demonstrated in our one-shot classroom experiment and in many other experiments. That it is not enough to sustain a trading platform that has no reputation system, however, is demonstrated in our experimental Strangers market. So does a feedback system help promoting trust and trustworthiness as suggested by theory? In Bolton et al. (forthcoming) the Strangers market was compared to a Reputation market, played over 30 rounds, in which, as before, a buyer never met the same seller more than once. However, in this market we introduced a reputation system that, similar to eBay's feedback forum, informs buyers about all past actions of their current seller (see Duffy/Ochs 2003; Bohnet/Huck forthcoming, for related experimental work). Outside the lab, there are a couple of problems with feedback systems having to do with the voluntary provision and strategic manipulation of feedback, and manipulation of online identities etc. (see Dellarocas forthcoming; Bolton/Katok/Ockenfels forthcoming b; Ockenfels 2003).

Gary E. Bolton/Elena Katok/Axel

198

Ockenfels

In our experiments, however, feedback on the seller's past behavior is always shared and reliable (because it is not given by the buyers themselves but by the experimenter), and sellers had no way to change their online identity. This way, the experiment studies the impact of feedback information on trading behavior when an ideal, frictionless reputation mechanism is available. Buyer

buy [56%] Reputation-seller

not buy

ship [13%] Buyer earns Seller earns

50 50

0 70

35 35

Figure 7: Average behavior in the Reputation market On average, there is economically and statistically significantly more buying (56 vs. 37 %; p < .05) and shipping (73 vs. 39 %; p < .01) in the Reputation market than in the Strangers market. In fact, the shipping probability is slightly higher than the threshold of 70% for trust being profitable. As a consequence, the trade dynamics too look quite different than in the Strangers market; trading starts at about the same level as in the Strangers market but then the level remains stable until the very last rounds, when the strategic value of having a reputation for being trustworthy vanishes and virtually all cooperation collapses. We conclude that introducing a feedback system in a market with strangers has a strongly positive impact on trust, trustworthiness and trading efficiency. Both buyers and sellers respond strategically to the information provided. 4.3 Partners Market The positive impact of electronic reputation mechanisms relative to a Strangers market has also been demonstrated in various field studies (see, e.g., Dellarocas forthcoming, for a recent survey). However, field studies cannot measure trust and trustworthiness on Internet markets relative to offline markets that are typically characterized by repeated interaction between the same trading partners. Theoretically, offline partners-relationships should not do better than online markets that implement reputation mechanisms that reliably share all relevant information about the sellers' past behavior. That is, from a theoretical perspective matching is irrelevant for trust to emerge as long as buyers are sufficiently informed—either through their own experience or through the experience of third parties. For the seller to have an incentive to be trustworthy, he need only expect that a future buyer will punish or reward his behavior; whether these punishments or rewards come from the same or from different buyers is irrelevant. The buyer, to induce this trustworthiness, need only be equipped

Trust among Internet Traders: A Behavioral Economics

Approach

199

with sufficient information about the sellers' histories; whether this information comes from one's own experience or from different sources is irrelevant. This is the basic message that derives from the game theoretic models: it is the information, not its source or its dissemination, that matters (Kreps et al. 1982; Ellison 1994; Bolton/Ockenfels 2004).

Buyer

buy [83%]

Partner-seller

not buy

ship [87%] Buyer earns Seller earns

50 50

0 70

35 35

Figure 8: Average behavior in the Partners market So, does theflowof reputation information make no difference? We tested this by comparing the Reputation market to a Partners market where the same two traders were matched over all 30 rounds, keeping everything constant including the anonymity of trading partners, communication channels, etc. Our experiment demonstrates that the simple theory is quite misleading when it comes to the impact of matching and information flows (Figure 8). There is substantially and statistically significantly more buying (83 vs. 56 %) and shipping (87 vs. 73 %) in the Partners market than in the Reputation market. Thus, even an ideal reputation mechanism cannot substitute for a partnersrelationship.17

5. Conclusions Normative economic theory, assuming that rationality and selfishness are common knowledge, does not capture trust in our basic game, because there is either never trustworthiness (in a one-shot or finitely repeated version of the trust game) or there is no risk in 'trusting' (in simple equilibria of the infinitely repeated game). This all changes when traders are allowed to have social preferences. Then, uncertainty about one's seller morals opens the door for trust, reward, exploitation and reputation building—even in the one-shot basic trust game, but especially in the (finitely) repeated game. In our experiments, sellers' intrinsic motivations to be trustworthy are not sufficient to sustain trade when not complemented by a feedback system. Translated to Internet market platforms, it seems likely that eBay or Amazon's market 1 7 As we explain in Bolton et al. forthcoming, this is because a feedback mechanism exhibits a kind of public goods problem in that, unlike in the partners market, the benefits of trust and trustworthy behavior go to the whole community and are not completely internalized.

200

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

for used books would quickly collapse without a reputation system. On the other hand, a reputation system without any individual social concerns cannot work either. If no seller would ever ship in a one-shot encounter, and if this were common knowledge among buyers, any cooperation would unravel, because the promise of being trustworthy will always be broken. However, clever institutional design multiplies the positive impact of 'truly intrinsic' trustworthiness that exists even without such systems. It is the interaction of social preferences and reputation mechanisms that solves to a large extent the trust problem on Internet market platforms. The strategic aspects of this interaction is in important parts captured by modern behavioral economics. At the same time, equilibrium theory and social preference models tend to underestimate the difficulties of promoting trust in anonymous online trading communities. For one, in one-shot games, economic theory does not capture the observation that the source of risk, independent of its stochastic properties, affects behavior. In terms of our basic trust game, the procedure by which the uncertainty of whether the item is shipped or not is resolved is critical for the buyers' decision to trust—even if the objective risk parameters are the same. The risk of being exploited by an untrustworthy seller makes buyers more cautious than (the same) risk that the item is accidentally lost by the postman. As a consequence, the trust problem is more severe than models of rational risk behavior would suggest. Similarly, in repeated games, the role of information as provided by electronic reputation mechanisms is overrated by theory and the role of matching underrated. Information problems can theoretically be solved by appropriate information dissemination, but, in fact, the matching scheme limits what can be maximally reached in anonymous online trade communities. Economic theory does not capture the observation that the source of information, independent of its content, affects behavior. As a consequence, trust problems are more severe than models of rational processing of information would suggest. We think these results suggest that a satisfactory explanation of trust and trustworthiness in Internet markets will require a model that grapples with both the mixed motives of the traders and how these motives inter-play with the institutional design of the market. In our experimental markets, trading behavior— trust and trustworthy behavior included—is clearly strategic but nevertheless differs in important ways from existing strategic models. By the same token, any model that assumes that trust and trustworthy behavior is entirely a matter of social norms or morality will probably be unable to capture the important behavioral response traders have to other traders strategic options as well as their own.

Trust among Internet

Traders: A Behavioral Economics

Approach

201

Bibliography Berg, J . / J . Dickhaut/K. McCabe (1995), Trust, Reciprocity, and Social History, in: Games and Economic Behavior 10, 122-142 Blount, S. (1995), When Social Outcomes Aren't Fair: The Effect of Causal Attributions on Preferences, in: Organizational Behavior and Human Decision Processes 63, 131-144 Bohnet, I./S. Huck (forthcoming), Repetition and Reputation: Implications for Trust and Trustworthiness When Institutions Change, in: American Economic Review (Papers and Proceedings) — / R . Zeckhauser (forthcoming), Trust, Risk and Betrayal, in: Journal of Economic Behavior and Organization Bolton, G. E./J. Brandts/A. Ockenfels (2003), Fair Procedures: Evidence from Games Involving Lotteries, Working paper, Penn State University — / — / — (1998), Measuring Motivations for the Reciprocal Responses Observed in a Simple Dilemma Game, in: Experimental Economics 1, 207-219 — / E . Katok/A. Ockenfels (forthcoming), How Effective are Online Reputation Mechanisms? An Experimental Investigation, in: Management Science — / — / — (forthcoming b), Cooperation among Strangers with Limited Information about Reputation, in: Journal of Public Economics, College of Business Administration, Penn State University — / — /R. Zwick (1998), Dictator Game Giving: Rules of Fairness versus Acts of Kindness, in: International Journal of Game Theory 27, 269-99 Bolton, G. E./A. Ockenfels (forthcoming), A Stress Test of Fairness Measures in Models of Social Utility, in: Economic Theory — / — (2004), Reputation, Information and Matching: Some Theory and Evidence, Work in progress — / — (2000), ERC: A Theory of Equity, Reciprocity and Competition, in: American Economic Review 90, 166-193 Brandts, J./G. Charness (forthcoming), Do Market Conditions Affect Preferences? Evidence from Experimental Markets, in: Economic Journal Brosig, J./A. Ockenfels/J. Weimann (2002), The Effect of Communication Media on Cooperation, in: German Economic Review 4, 217-241 Charness, G. (forthcoming), Attribution and Reciprocity in a Experimental Labor Market, in: Journal of Labor Economics — /M. Rabin (2002), Understanding Social Preferences with Simple Tests, in: Quarterly Journal of Economics 117, 817-869 Dellarocas, C. (forthcoming), The Digitization of Word-of-Mouth: Promise and Challenges of Online Reputation Mechanisms, in: Management Science Duffy, J . / J . Ochs (2003), Cooperative Behavior and the Frequency of Social Interaction, Working paper, University of Pittsburgh Dufwenberg, M./G. Kirchsteiger (forthcoming), A Theory of Sequential Reciprocity, in: Games and Economic Behavior — /U. Gneezy/W. Giith/E. van Damme (2001), Direct and Indirect Reciprocity: An Experiment, in: Homo Oeconomicus 18, 19-30 Ellison, G. (1994), Cooperation in the Prisoner's Dilemma with Anonymous Random Matching, in: Review of Economic Studies 61, 567-588 Engelmann, D./M. Strobel (forthcoming), Inequality Aversion, Efficiency, and Maximin Preferences in Simple Distribution Experiments, in: American Economic Review

202

Gary E. Bolton/Elena

Katok/Axel

Ockenfels

Fehr, E./K. Schmidt (1999), A Theory of Fairness, Competition, and Cooperation, in: Quarterly Journal of Economics 114, 817-868 Frank, R. (1988), Passions within Reason. The Strategic Role of the Emotions, New York Güth, W./H. Kliemt/A. Ockenfels (2001), Retributive Responses, in: Journal of Conflict Resolution 45, 453-469 Kandori, M. (1992), Social Norms and Community Enforcement, in: Review of Economic Studies 59, 63-80 Kreps, D. M/P. Milgrom/J. Roberts/R. Wilson (1982), Rational Cooperation in the Finitely Repeated Prisoners' Dilemma, in: Journal of Economic Theory 27, 245252 Ockenfels, A. (2003), Reputationsmechanismen auf Internet-Marktplattformen, in: Zeitschrift für Betriebswirtschaft 73, 295-315 Offerman, T. (2002), Hurting Hurts More than Helping Helps, in: European Economic Review 46, 1423-1437 Rabin, M. (1993), Incorporating Fairness into Game Theory and Economics, in: American Economic Review 83, 1281-1302 Resnick, P./R. Zeckhauser (2002), Trust Among Strangers in Internet Transactions: Empirical Analysis of eBay's Reputation System, in: M. R. Baye (ed.), The Economics of the Internet and E-Commerce, Advances in Applied Microeconomics 11, Amsterdam, 127-157 Wilson, R. (1985), Reputations in Games and Markets, in: A. Roth (ed.), Game Theoretic Models of Bargaining, Cambridge, 27-62

Analyse & Kritik 26/2004 ( © Lucius & Lucius, Stuttgart) p. 203-219

Werner Gùth/Hartmut

Kliemt

The Evolution of Trust (worthiness) in the Net*

Abstract: The main results of our indirect evolutionary approach to trust in large interactions suggest that trustworthiness must be detectable if good conduct in trustrelationships is to survive. According to theoretical reasoning there is a niche then for an organization offering a (possibly) costly service of keeping track of the conduct of participants on the net. We compare traits of an organizational design as suggested by economic reasoning with those that actually emerged and ask whether institutions like eBay will increasingly have to 'economize on virtue' although so far they could rely on its spontaneous provision.

1. Introduction and Overview Order can emerge spontaneously if an interaction is repeated and such mechanisms as reputation formation can run their course. Platforms like eBay that organize social cooperation over the Internet are a case in point. They produce 'order without law' (see Ellikson 1991) or without commanding the fundamental coercive powers to raise taxes, to collect information and to punish misbehavior but definitely not without incurring costs. Therefore we must ask why individuals are willing to bear the costs of organizing social cooperation over the Internet, why others are willing to pay them for rendering such organizational services as there are and what the prospects for the survival and development of service providers are. Addressing such questions in a somewhat indirect way we will first sketch (2.) some results of our former work that show that even in large (almost) anonymous interactions as today prevailing on the Internet trustworthiness can survive if it can be detected with some reliability. This contradicts some common views to the effect that 'large' interactions cannot conceivably be organized without the fundamental coercive power of the state and therefore could in particular not persist on the international level where a common state organization is lacking. It, however, presupposes that individuals have characteristics that as a matter of fact restrict their opportunistic choice making in ways akin to boundedly rational behavior. In the next section (3.) we lay out some 'economists" ideas concerning desirable or likely properties of a net platform—Big Brother or BB—for organizing bilateral exchanges. Comparing this reference model with eBay is instructive since it shows that eBay chose a route relying more strongly on community feel* Critical comments by Bernd Lahno were very helpful in improving the paper; of course, he is responsible for all remaining errors.

204

Werner Guth/Hartmut

Kliemt

ings and intrinsic motivation than on extrinsic incentives like reputation scores as suggested by 'more economic' approaches to human behavior (4.). Starting from the contrast between eBay and BB we raise additional questions, indicate possible lines of future research, and finally issue a cautionary remark on the scope and limits of eBay as a platform of trade (5.).

2. The Indirect Evolutionary Approach to Trust in Large Transactions' Systems 2.1 T h e Basic Trust Problem and its Representation as a Simple Game Imagine a situation in which there are so many potential interaction partners for bi-lateral transactions that in repeated interaction none of them can keep track of the actions of all others. The behavior of each of the actors directly and significantly affects her or his partner but is individually insignificant for the collective result. We refer to such a transactions' system as a large transactions' system. Given this definition it is obvious in particular that whether or not there is a 'climate of trust' in a large transactions' system is not a matter to be influenced strategically by individual actions. Individual actions affect the partner of a transaction only. The overall 'trust climate' characterizing the ongoing interaction is the result of all the individual actions but none of the individual acts is significantly responsible for the collective result. Interaction partners are concerned with the results of the particular interactions in which they participate. They are also interested in their own reputation. If there is a reputation effect they will take it into account. But as far as there is no reputation effect what may be called the 'trust predicament' emerges: every participant of interactions taking place in a large transactions' system can act opportunistically behind the veil of individual insignificance. In pursuit of their common interest both actors can and in view of their extrinsic profit motives should always deviate from agreements and go for their private interest. When engaging an interaction with a partner both individuals expect to be better off if both act as agreed. However, each knows that showing trust in moving first is risky because the second mover can exploit the first moving actor's trust. The mutually advantageous deals will not be realized if the actor in the first mover role does not trust or the actor in the second mover role does not reward trust. The lack of trust or the presence of the risk of exploitation can thus stand in the way of what is in the best interest of the actors. The following graph presents a rational choice explication of the trust predicament in unilateral trust problems (the prisoner's dilemma or exchange being the paradigm example of bilateral trust problems): Player i starts by deciding between iV(o-trust) and T(rust). After N the game ends with player i earning s and player j earning 0. After T the game continues with j's choice between ¿J(xploitation) and .R(eward).

The Evolution of Trust(worthiness)

in the Net

205

1

0,0)

(0,1-111)

(U)

Figure 1: The game of trust with payoff parameters 0 < r ,

s 0) otherwise it affects merely the (subjective) payoff without being behaviorally relevant. If the motivation to act honestly is strong enough then exploitation in the second mover role will be avoided. However, this will further cooperation only if it can be known by the first mover. The first mover will trust only if he knows that the second mover is motivated to act fairly—or at least he must expect with sufficient probability that the second mover is trustworthy. Quite obviously, actors who manage to show trust and to receive the reward for trust in the first mover role will be better off than those who do not trust or are

206

Werner Giith/Hartmut

Kliemt

exploited. Likewise the trustworthy individuals who are recognized as such will be trusted and do better than the untrustworthy if otherwise no trust will be shown—this will apply at least under suitable conditions. 2.2 Basic Ideas of Indirect Evolution We studied the basic game of figure 1 quite extensively within the framework of an indirect evolutionary approach (Giith/Kliemt 1994, 2000; Giith/Kliemt/Brennan 2003; Giith/Kliemt/Peleg 2000). In this approach rational choices that are made by a rational forward looking actor in view of his 'subjective expectations' or 'subjective payoffs' are embedded in an evolutionary process which fixes success of decision determinants in terms of 'objective' payoffs that are brought about by the choices. The duality of subjective payoffs guiding behavior and objective payoffs determining success is at root of the indirect evolutionary approach in general and of the corresponding account of 'trustworthiness' in particular. As an illustration it may be helpful to think of the evolutionary model of firm competition and innovation that has been suggested by Armen Alchian a long time ago (see Alchian 1950). In this seminal paper it is argued that under suitable competitive conditions the more profitable firms will drive out the less profitable ones regardless of whether or not those who are running the firms are consciously aiming at profit maximization. The subjective aims as represented by the subjective utility function of the staff of a firm may be almost anything. For instance the staff might aim at 'providing the customer with the best quality' available, 'furthering the common weal', 'maximizing market share', 'having an adequate share of the market' or whatever. If a strategy is as a matter of fact objectively profitable then firms that adopt it will survive and otherwise not. Survival is determined by objective success of choices and thus independently of the subjective motives leading to the choices. Firms that pursue for what reasons ever strategies that lead to losses will eventually be eliminated from the game. As Alchian argued, even if initial strategies would be assigned completely randomly to firms, in the end those firms survive that are endowed by chance with a successful strategy. This will hold good as long as there is a growth, birth and, decline as well as death mechanism that will let the objectively more successful flourish and the others not. Alchian knew Darwinian arguments—that had a long tradition in social theory anyway—very well. But he wrote before more precise concepts of evolutionary game theory like that of an evolutionary stable strategy (ESS) were developed. Later results that rely on such concepts show, that once profit maximizing strategies are established (as a monomorphic population) other strategies cannot invade such a population anymore. The result is evolutionarily stable if interaction forms a competitive market where no seller's profit depends on another seller's behavior (see Giith/Peleg 2001). Real world examples of objectively successful strategies that are not meant to be profit maximizing might form those (in particular German and Japanese) companies run by engineers that were pursuing a 'quality first strategy'. The

The Evolution of Trust(worthiness)

in the Net

207

strategy was economically doubtful and often seemed foolish in the short perspective but obviously was selected by market forces as the objectively profit maximizing strategy over the long haul. Even though the heads of companies were acting against maximization of subjectively perceived profits, what they did for other reasons was sufficient to be ahead of the average performance of the market as measured in 'true' or objective profits. 2.3 Indirect Evolution in Large Groups The indirect evolutionary approach to the evolution of trust in large groups (on small interactions see Giith et al. 2002) considers basically two second mover strategies for the game represented in figure 1. For the first strategy the parameter m is sufficiently large to affect behavior. The individual actor is endowed with an intrinsic motive that is strong enough to make her behave in trustworthy ways. This makes her a trustworthy actor. She is not only strategically behaving as if she were one but is making strategy choices because she is trustworthy. For the second strategy the parameter m is not large enough to affect behavior. The individual actor is not endowed with a sufficiently strong intrinsic motive such as to induce her to reward the trusting initial move of the first mover. She is not trustworthy and therefore will not behave in trustworthy ways unless there is a strategic extrinsic motive to do so. Social theorists almost across the board have uttered skepticism about survival prospects of such forms of good conduct as 'trustworthiness'. 'The elicitation of good conduct' (see Klein 1997) in large groups seems to be ruled out by the conditions of large scale social interaction. They were even more skeptical when it came to the survival prospects of genuine trustworthiness or the intrinsic motivation to behave in trustworthy ways. People would behave as if trustworthy for strategic reasons for instance in an effort to maintain their good reputation in repeated interactions but there would not be the genuine 'thing' (though even in Kreps et al. 1982 the genuine thing and the 'crazy' belief in its presence are needed to induce others to behave as if trustworthy). Relating this to the preceding discussion in general and in particular to the concept of a large transactions' system as introduced before the issue to be addressed is whether or not trustworthy individuals can survive in large transactions' systems: Are there and if so what are the conditions under which trustworthy individuals will fare well enough to avoid extinction? 2.3.1 A Simple Model of Large Transactions'

Systems

To model a large transactions' system we make some assumptions about the scope and kind of the interaction (for formal derivations underlying the following discussion see for instance Giith/Kliemt 1994; 2000): Imagine infinitely many rounds of play of infinitely many individuals (representing a very large group interacting under conditions of individual insignificance). Let the individuals be matched randomly to play the game of trust of figure 1. Half of the time they end up in the first and half of the time they end up in the second mover role. Some of the individuals may be trustworthy in the second mover role. After

208

Werner Giith/Hartmut

Kliemt

each round of play objective relative success fixes the population composition on the next round of play. Whether the choice of trust is more advantageous than showing no trust depends on the population composition, i.e. on the presence of trustworthiness. As long as sufficiently many trustworthy individuals are around the payoff expectation of showing trust can be higher than that of showing no trust even without the possibility of screening. Therefore it is worthwhile to incur the risk of showing trust and rational individuals will show trust in the first mover role. If undetected, untrustworthy individuals can exploit first mover trust, however. As long as objective success on each round of play determines how many players of each type will be present on the next round of play relatively higher success of the untrustworthy will drive out the trustworthy until a lower threshold of the population share of the trustworthy is reached from which on it will not be advantageous to trust. This holds good under all plausible dynamics which correlate higher objective payoff with higher population share. Once there are not sufficiently many trustworthy individuals around to make it worthwhile to incur the risk of showing trust no first mover will ever deliberately choose to show trust. However, as long as trust is chosen by mistake occasionally, the untrustworthy will once in a while be presented with the opportunity to exploit a trusting first mover. They will then fare better than the trustworthy and fare equally well as the trustworthy in all other instances when no trust is shown. In this mistake driven process the trustworthy will still be eliminated over the long haul though much more slowly than in the case of rational trust. Whether the trustworthy or those who are not trustworthy succeed more generally depends on whether or not trustworthiness can be detected by potential first movers. If first movers can trust the trustworthy and avoid trusting those who do not deserve to be trusted the trustworthy individuals will flourish. If the abilities to discriminate between types are perfect then only trustworthy types can conceivably survive. The reasoning is obvious: the trustworthy will be trusted and therefore receive a higher objective payoff than the untrustworthy in each and every transaction in which they take part. The untrustworthy will not be trusted and therefore fare less well than the trustworthy. They never will find a chance to exploit first mover trust. Going to the obvious opposite extreme in which all type discrimination skills are lacking and type information is completely private the untrustworthy will fare better if trusted and fare as well as the trustworthy if nobody ever trusts. In that case the trustworthy will go extinct if only slowly so beyond the threshold from which on trust will be shown only by mistake. 2.3.2 Enriching the Model by a Costly Detection Technology The model becomes richer if we allow screening to be costly. Consider what may be called a 'costly detection technology of limited reliability' or a 'detection technology' for short. Such a detection technology provides a signal to the actor in the first mover role. Those who pay the price for using the technology can thereby know with higher probability whether the partner in the second mover role is a trustworthy or an untrustworthy type. Whether investing the price or

The Evolution of Trust(uiorthiness)

in the Net

209

incurring the costs is worthwhile, of course, depends on the reliability of the technology and the type composition of the population. If almost all individuals are trustworthy or if almost all individuals are untrustworthy, it will not pay to invest in screening. This and some additional insights are, in a way, summed up by figure 2 which shows the population share p of the trustworthy on the horizontal and the cost C for type detection on the vertical axis. If the reliability of the technology is very low then the realm in which it is advantageous to rely on it will be small. In figure 2 the triangle would shrink by lowering its top corner C" while the two other corners on the horizontal axis might be approaching s. If the costs are higher the use of the technology will be advantageous for fewer population compositions. Whenever the population composition parameter p falls for some C' into the interval {w, II), whose boundaries 7r and II vary with C', the population share of trustworthy individuals will grow due to the availability of screening (after costly investment in type detection). For given C' the population share of trustworthy individuals will shrink outside that interval, i.e. for p < TT or p > II where no type information is bought. For other values of C basically the same argument applies. The size of the angles of the triangle depends on the reliability of the detection technology. The sides of the triangle show for each C the lower limiting value p(C) from which on up to p(C) it pays to invest in detecting the trustworthy and the upper limiting value p(C) above which it does not pay to invest in screening.

c

c

C"

/

c _

\p(C)

i\

/

9.(C). 0

7i

s

n

\

Figure 2: Dynamics of trustworthiness The more reliable the detection technology the higher its costs can be without preventing its use. If the detection technology works with perfect reliability and its costs are zero then, as previously indicated, the full interval [0, 1] of initial

210

Werner Giith/Hartmut

Kliemt

population compositions will lead to a population of trustworthy individuals only. The solid arrows show dynamics that are driven by deliberate choices while the dashed arrows show dynamics that are merely mistake driven. For C > C" there is only one evolutionarily stable type composition, namely p * = 0 , resp. the m < 1-r-monomorphism, whereas for C = 0 and perfect reliability of screening only p * = l , resp. the m > 1-r-monomorphism is stable. In all other cases (i.e. for 0 < C' < C" or C = 0 with unreliable detection) there are two evolutionarily stable population compositions p, namely p * = 0 and p*=p(C). Figure 2 also illustrates the basins of attraction of these two: whenever the initial composition pQ satisfies p 0 < 7r, the population composition p converges to p * = 0 whereas for p 0 > 7r it will converge to p*=p(C)—either by an increase of p if p 0 < p(C) or a decrease of p if p 0 > p(C). 2.3.3 Relating Results to eBay To relate the findings of the indirect evolutionary approach to trust in large interactions' systems to eBay it may be helpful to sum up some general insights of the previous discussion first. If type discrimination is sufficiently inexpensive and sufficiently reliable then: • there is a niche in which trustworthiness can survive in evolutionarily stable ways; • there is a niche for those providing at reasonable costs the detection technology or organizing an interaction 'platform' on which trustworthiness may survive; • if detection is imperfect and/or detection costs are positive then a sufficient degree of trustworthiness must be initially present to stabilize its presence endogenously (there will always be a niche for some untrustworthy individuals but the trustworthy can survive only if there is initially a sufficient number of them). For successful Internet auction platforms like eBay, this suggests the following conclusions: • the founders of eBay were lucky in the sense that initially the situation p 0 > 7r (see figure 2) prevailed; • the relative carelessness of behavior when transacting over the net seems to suggest that participants perceive the population share of trustworthy individuals as p > II and therefore in general do not incur (idiosyncratic) costs of type detection; • no efforts at screening will be worthwhile as long as the perception of p > II prevails and therefore the share p of trustworthiness on the Internet will tend to decline and chances for as well as instances of fraud will tend to increase.

The Evolution of Trust (worthiness) in the Net

211

Without any doubt eBay is a great success. Its market capitalization is remarkable. It is a company going increasingly international and it might even seem to be a first step towards a global market in which individuals transact increasingly across borders without specialized intermediaries. The Internet is thereby bringing about what may be called spontaneous globalization, not only of information but also of transactions. Nevertheless, in view of the precarious nature of a general climate of trust in large transactions' systems it is by no means a sure thing that eBay will survive. Interpreted in terms of our general model eBay could work rather well as long as p > II and people were showing trust. Unless the general climate of trust and trustworthiness be eroded the share p of trustworthy individuals must be protected. The safeguard provided by eBay is its reputation mechanism. However, from a rational choice point of view the reputation feedback mechanism of eBay is rather peculiar. The exclusion from interaction that would drive the efficacy of reputation mechanisms in other contexts does not play a central role for eBay. Without possibilities of proper verification of identity a trader in eBay cannot be detected if switching to a new identity. Moreover, the mechanism of building up a 'good' reputation is not cheat proof either. Reputations may be built up strategically for exploitation purposes. What may be called 'eBay brand names' can be traded. So, how to prevent good reputations being bought up by bad guys? True enough, the higher the premium that is carried by positive reputation on the net (but there is not too much evidence for that anyway) the more costly it would become to send a false signal by buying into a good reputation or by willing to 'burn' a good one for the 'big kill'. Still, costs would not preclude all forms of such fraud among strategically rational traders who seek and might find ways to exploit market platforms like eBay and the bigger and the more global transactions become the higher the risks. We submit that eBay will have to change its character and become more 'knave proof' in the process.

3. The Example of Big Brother With the preceding results on the 'evolutionary niche for trustworthiness' in hand let us imagine an institution that may be called big brother. Let us think of 'BB' as providing a net platform and the essential information on the trustworthiness of potential transaction partners that might get into contact through use of that platform. In view of the anarchic character of the net BB is lacking that fundamental coercive power that goes along with the modern state. In particular BB cannot check whether information whispered into his ears is true or not. At least he has no independent coercive means of finding out the truth. BB must rely on whatever he is told and try to distill from this some more reliable information. Can such a powerless BB implement a useful yet costly detection technology and get paid for it if he merely prudently compares information from different sources? Asking this question is, in a way, asking for something akin to eBay. But let

Werner Giith/Hartmut

212

Kliemt

us stay a bit with BB. BB stands more or less for an economist's 'natural way' to set up a large transactions' system that can facilitate exchange and reputation formation over the Internet. Obviously a typical economist should be expected to operate under the assumption that " . . . people would act economically; when an opportunity of an advantage was presented to them they would take it" (Hicks 1979, 43). But restless maximization and taking of all opportunities is perhaps too far removed from real world behavior to be taken seriously. Some form of merely boundedly rational opportunity taking behavior must, however, clearly be taken into account. Even if not all people take all opportunities all the time some will do. Presumably in an environment like eBay they will do so rather frequently and with relative impunity. Since an economist would anticipate that this might undermine the whole transaction system we would expect B B to be set up in a way that prevents such problems as well as possible. Obviously it is crucial that we specify what B B would do with the information that he might acquire. The simplest way and a method that comes to one's mind immediately is a check on consistency of reports. Lacking fundamental coercive power and without any backing by such power through state enforced legal institutions this may be almost all B B might be able to do. In forming a reputation for arbitrary individuals i, j that have enlisted as potential transaction partners BB cannot do much more than comparing what the two partners i, j tell him, i.e. by checking whether both parties conceive of a deal as properly completed. Counting the numbers of transactions and to relate them to the number of good, bad, consistent or inconsistent reports seems more or less all B B can do. To see more specifically what may go on here, let us for all i and all times t call the following vector a 'counting type': c\ = (m\,n\,ml2,n2);

where

m\ is the number of participations as seller in successful transactions, n\ is the number of participations as seller in transactions, m%2 is the number of participations as buyer in successful transactions, n\ is the number of participations as buyer in transactions. How a counting type develops through time depends on how the counting is done. It seems reasonable to assume that in all cases in which both participants in a transaction report the same result, counting would just follow the reports. Whenever participants reported differently B B can either count the transaction as unsuccessful, successful or count it not at all. BB must make a strategic decision here. As a most simple yet plausible mechanism imagine the following: • Initially all counting-types of all individuals i are CQ = (m\, n\, mx2, n\) = (0,0,0,0).

• BB asks i and j to report simultaneously and - reduces the reputation of non-reporters i, by increasing n\ or n\ but neither m\ norml2;

The Evolution of Trust (worthiness) in the Net

213

- counts according to report if i and j report the same result, i.e. r(i)=rO);

- increases n\ orn\ but neither m\ norm%2, for both if r(i)^r(j), unless one of the self-reports is unfavorable leading to no alteration of count for any party (in that case both were presumably honest and a mistake occurred preventing successful completion of the interaction). BB does not command the power to tax. But BB cannot provide his services for free either. He must get paid for his services. Therefore to act reasonably, BB should keep all reputation information strictly to himself—or so it seems. He would offer information on the trustworthiness of a potential transaction partner only when asked and he would answer only if a fee for service would be paid. BB as conceived in our thought experiment would continue his dealings with customers only if those customers were willing to identify themselves credibly and would report the results of transactions to him. As far as this is concerned BB would and could require positive proof. (The latter features are clearly different from eBay.) In 1996 a study group at the University of Duisburg (Stefan Dreckmann, a programmer, Markus Gruner, a simulation expert from physics, and Hartmut Kliemt) simulated a club-like large transactions' system run by BB. As suggested by the former results on the evolutionary niche for trustworthiness we started with quite a high proportion of trustworthy and truthful individuals. These individuals would be trustworthy in second mover roles. Being intrinsically motivated 'to do the right thing' they would truthfully report the results of transactions to BB and thereby give BB access to correct information depending on the number of intrinsically motivated types around. Concerning the other simulated individuals we allowed for arbitrary strategies of cheating and lying that could possibly crowd the trustworthy and truthful out of the population and scrutinized whether or not these strategies would be more successful than those of individuals showing good conduct. BB relied on counting types trying to extract information from the reports he got and updating the types accordingly then informing potential customers about the trustworthiness of potential transaction partners on condition of a feed back report and a modest fee (counting non-feedback as cheating). According to simulations with more than 50 per cent of trustworthy and truthful individuals in the population good conduct would drive out bad conduct. Since the objective success of strategies on each round of play would fix the proportions of trustworthy and truthful individuals in the next generation the relatively more successful individuals were present in the population in ever higher proportions after every round of play if starting with favorable initial conditions or an exogenous supply of trustworthiness and truthfulness. To put it slightly otherwise, starting out with sufficiently many trustworthy and truthful individuals and sufficiently low costs a BB who could only check on consistency of reports when updating his estimates of the trustworthiness of the transaction partners could make the trustworthy and truthful flourish. And BB could do so by means of a reputation mechanism only.

214

Werner Giith/Hartmut

Kliemt

Insufficient computing power (1996 is a long time back in computing) prevented to check on the long run stability of such results. Still, the simulations did at least indicate that even a powerless BB with very simple means of detecting the trustworthy and truthful might do the trick. This was and presumably still is interesting to some extent. However, BB did not turn into eBay or something akin to that great success. Even though some of us toyed with the idea of trying to exploit its potential, we would certainly have encountered a disappointment. For in fact BB did not have the potential for success while eBay did. It seems to us that eBay succeeded because it was less of an economist vision than BB. It was less concerned with cheating than BB would ever have been. BB appealed to interests, tried to build in checks on truth and hedged his information where eBay was trustful and openly appealing more to community feelings of those transacting over the net on its platform. It could do so since addressing originally only members of a single company it started in a kind of community environment. We speculate that it was triggering thereby a kind of co-operative heuristic in participants. When transcending the borders of the original 'community' additional transaction partners were taken into an already co-operative environment. Again certain heuristics recommending 'communitarian' types of behavior must have thereby been triggered. So the initial success of eBay perhaps was possible only since it was not so much of an economic institution as BB. Nevertheless the long run success may depend on eBay becoming more 'knave proof' or economic. It seems worthwhile to illustrate such speculative reasoning somewhat further by a short comparison between BB and eBay. We will then be in a position to address again the question of whether what was good for eBay initially will be good for it over the long haul or whether eBay might have to move somewhat closer to BB if it is to survive its own success on the several national and even global levels. In doing so we neglect the problems that might arise for both eBay and BB if these platforms are—as seems to be the case—used to perform transactions in illegally acquired goods.

4. A Comparison of eBay and BB In particular the fact that reputation information is publicized on the net and thereby provided as a collective good to all members of eBay seems to be significant. Distributing reputation information freely facilitates seeking out partners with certain reputation characteristics. More generally speaking eBay is only in a much weaker sense a club than BB. In particular lying about one's identity is possible quite easily in eBay and has been excluded in the environment created by BB. There is also no incentive provided by eBay at least not such a strong one as in conceivable BB environments to report the results of dealings. Where BB would diminish the reputation of those who are not reporting the results of transactions, eBay does not do any of the kind but relies on an intrinsic motivation to report. People must be motivated to report and to do so truthfully. Here as in other regards eBay seems at least implicitly to appeal to feelings of

The Evolution of Trust (worthiness) in the Net

215

belonging to a 'community' and the like. To put it slightly otherwise eBay seems to rely more on communitarian trust and on retributive emotions than commercial platforms of the BB type as imagined in the corresponding economically motivated thought and simulation experiment. Emotional trust is fragile in particular in a large transactions' system under conditions of individual insignificance. It may be that human individuals who cannot really shift their behavioral gears will often act in the new environment of eBay as if participating in a small numbers' interaction. However, it seems doubtful that we can 'trust' in the prevalence of such a systematic error over the long run. Bearing in mind the basic results of the indirect evolutionary approach to trust in large transactions' systems we must ask whether or not eBay really can create evolutionarily stable co-operation in the following sense: Is type detection in eBay sufficiently reliable and cheap to prevent innovative untrustworthiness from crowding out trustworthy behavior? Will net platforms have to acquire stronger club characteristics and have to impose stricter rules of admission in view of strategic reputation building (e.g. accumulating a reputation for trustworthiness incrementally with small stakes while going for the 'big kill' in a fraud eventually)? BB must make strategic decisions when setting up his reputation formation algorithm as well as other rules of the game that he provides. Likewise eBay must consider modifications of its rules if it intends to be prepared for future challenges to its viability as might emerge rather soon (again increasing trafficking in illegally acquired commodities being a crucial issue). It is fruitful to bear in mind BB when addressing such issues for eBay. Somebody who would not know anything about eBay would certainly imagine that a reasonable BB would not disclose the full counting type to the participants for free. He would rather insist on giving aggregate information on trustworthiness as seller or buyer respectively and would disclose it only on demand and against payment. If BB discloses all information and perhaps also allows the presentation of qualitative information, e.g. qualitative assessments by the parties to a transaction, then BB gives up control over how the information is used and aggregated by customers. Customers might like that in the first place but it might undermine the viability of the platform since there would not anymore be clear reputation signals applying to all and fixed by the platform providers. Now, compare that with eBay. eBay partly presents the reputation information in aggregate form. So it seems that it follows the same policy. But quite surprisingly yet also quite in line with its policy of being transparent eBay also allows to check the past transaction records of participants. eBay provides more information than it needs to and than is used most of the time by participants. Clearly transactions can fail for reasons other than deliberate cheating or deliberate exploitation. Things may go wrong by chance. If we include that possibility then there should be some tolerance in the formation of reputation. The unforgiving approach according to which those who cheated once are 'out of the game' so to say might be self-defeating. But forgiving strategies may be subject to exploitation. For instance if people know how close they are to the threshold beyond which they will be trusted or not trusted in a forgiving strat-

216

Werner Giith/Hartmut

Kliemt

egy they could respond to this information strategically. The most dangerous transaction partners should then be those who have never cheated before but might cheat without risking their reputation since the built in tolerance would protect them. On the other hand those who would fall below the threshold of acceptability as a trustworthy partner of interaction should they fail to fulfill their part of a specific bargain might be particularly trustworthy or rather reliable for opportunistic reasons in that deal. If the full counting type were revealed, actors could themselves draw conclusions. Some actors might be unforgiving, others might be very forgiving still others might act in flexible ways responding to additional information as might be available. As indicated, specialized services judging the trustworthiness of participants might be expected to emerge as secondary organizations on such a platform. This may in one way improve the reputation mechanism but it may also reduce the financial basis of the platform and the willingness of transaction partners to report—in particular if they have special counseling deals with secondary services. In view of this it does not seem to be obvious that the strategy of eBay to disclose more or less all information to all transaction partners is the optimal one or even one that can sustain eBay over the long haul. The issue is not merely whether or not eBay is sufficiently robust against increasingly professional efforts of cheating, it is also whether an informal set up as of eBay might not destabilize its own basis by providing its services too freely and in ways that invite secondary services to enter transactions. Secondary services already offered by eBay as costly devices, but rarely used might act as 'trustees' who, for instance, guarantee both payment and delivery of items as promised and in a quality as promised. So-called 'pick up points' (typically gas stations and the like) are used nowadays already to facilitate Nettransactions of other providers. The goods are delivered there and after inspection paid typically in cash. Firms offering new goods offer to testify the quality of used products of their own brand that are meant to go for sale on the net etc.—This all may be helpful in containing risks of transactions but it is costly to some extent and widely spread use of such services will render the process much more clumsy. It might in particular also hurt community feelings and as a consequence crowd out the intrinsic motivation to report the results of transactions on which eBay as of now relies.

5. Where Will and Where Should eBay Go from Here? Bearing in mind the lessons from the indirect evolutionary approach to trust in large transactions' systems it is not at all obvious that eBay in its present form might be a sustainable success. Success may breed new success, though, in that the present platform may develop into a new, different one in ways dependend on the stage already reached on the development path. It seems central to inquire whether or not good conduct can be evolutionarily stable in the environment created by eBay. Our skepticism may be due to our outsider point of view but

The Evolution of Trust(worthiness)

in the Net

217

we would certainly be less skeptical if some of the following issues could be dealt with convincingly: • Is it possible to go beyond general propositions about reliability and costs and to say something more specific about those reputation mechanisms that lead to evolutionarily stable positive population shares of trustworthy individuals? • If there is a population of basically trustworthy and truthful individuals, can it be sustained under a counting type characterization of trustworthiness as buyer or seller that is public on the net? • What are the relative merits of tolerant as compared to intolerant reputation mechanisms and is there an optimal degree of tolerance as well as of information about distance to thresholds of tolerance? • Would weighing with the monetary value of transactions render counting type mechanisms more stable in some precise sense of evolutionary stability against subversion by strategic reputation building in small deals followed by 'hit and run'-exploitation whenever a big deal comes up? Some more or less experimental issues: • Can we design economic experiments to identify how boundedly rational individuals respond to different forms of information in particular to counting type reputation? • Can we present experimental evidence that and why the inclusion of qualitative ("he is a good eBayer") assessments by the parties into the information vector as practiced by eBay is helpful? • Would the future availability of personal pictures of transaction partners (and other forms of personalizing relations) on the net affect behavior? • Is it true that 'thicker' communication will further co-operation in general? Of course, there has been quite a bit of experimentation on eBay notably by Karen Cook, Alvin Roth and Axel Ockenfels, Chris Snijders and, also mentioned here last but not least, by Toshio Yamagishi. For the present discussion some of Yamagishi's work (see 2002) seems to be particularly relevant. His theoretical argument that positive reputation in an open system may be superior to negative reputation over the long haul seems to be quite convincing. It should however also be factored in that reputation building affords no guarantee that positive reputations can not be strategically manipulated and sold. In particular if operations become more globalized there may be quite some efforts to cheat along that dimension. Whether in the end we will reach a kind of global eBay that would extend auctions to global levels as some might hope seems open to serious doubt. It seems rather likely that there will be countervailing forces that would threaten the evolutionary stability of such a set up.

218

Werner

Giith/Hartmut

Kliemt

It should be kept in mind that behaving as if trustworthy and true trustworthiness are two different things. The first will not work without the second over the long haul if something like eBay is to work on as smoothly as it does as of now. As the indirect evolutionary approach suggests the reliable traders must be distinguishable from 'fortune hunters' who behave only as if genuinely trustworthy with some 'reliability'. As far as this is concerned eBay uses its reputation score and provides—for those who are willing to go through the somewhat tedious exercise of retrieving it—additional information on the value and kind of transactions. To reduce information costs weighted measures of performance may be provided such as to prevent strategies of accumulating a good reputation by many small transactions and then to go for the 'big kill'. It may also be true that the untrustworthy suffer from higher discount rates, 'weakness of the will' problems or a lack of the 'indirectly' profit-maximizing intrinsic motivation to realize the long run gains of a good reputation because they just cannot resist the temptation to cheat on small things (see Frank 1988). Still, there will be opportunists who specialize in seemingly good conduct for exploiting trust by big kills. Again there may be remedies by not allowing too many transactions of the same individual of higher value at the same time on eBay. However, going through all the possible remedies, in the end the big deals will depend on other measures to facilitate exchange (guarantees, hostages etc.). A global eBay will be a global flea market but not a global big deal market.

Bibliography Alchian, A. A. (1950), Uncertainty, Evolution, and Economic Theory, in: Journal of Political Economy 58, 211-221 Ellickson, R. C. (1991), Order without Law: How Neighbors Settle Disputes, Cambridge/M A-London Prank, R. (1988), Passions within Reason. The Strategic Role of the Emotions, New York Giith, W . / B . Peleg (2001), When Will Payoff Maximization Survive?—An Indirect Evolutionary Analysis, in: Evolutionary Economics 11, 479-499 — / C . Schmidt et al. (forthcoming), Fairness in the Mail and Opportunism in the Internet—A Newspaper Experiment on Ultimatum Bargaining, in: German Economic Review — /H. Kliemt (1994), Competition or Co-operation: On the Evolutionary Economics of TYust, Exploitation and Moral Attitudes, in: Metroeconomica 45, 155-187 — / — (2000), Evolutionary Stable Co-operative Commitments, in: Theory and Decision 49, 197-221 — / — / B . Peleg (2000), Co-evolution of Preferences and Information in a Simple Game of Trust, in: German Economic Review 1, 83-100 — / — / G . Brennan (2003), Trust in the Shadow of the Courts, in: Journal of Institutional and Theoretical Economics (JITE) 159, 16-36 Giith, S./W. Giith, W./H. Kliemt (2002), The Dynamics of Trustworthiness Among the Few, in: The Japanese Economic Review 53, 369-388 Hicks, J. (1979), Causality in Economics, Oxford Klein, D. B. (ed.) (1997), Reputation. The Elicitation of Good Conduct, Ann Arbor

The Evolution

of Trust(worthiness)

in the Net

219

Kreps, D./P. Milgrom et al. (1982), Rational Cooperation in the Finitely-Repeated Prisoners' Dilemma, in: Journal of Economic Theory 27, 245-252 Yamagishi, Toshio (2002), The Role of Reputation in Open and Closed Societies: An Experimental Study of Internet Auctioning, http://ccs.mit.edu/dell/reputation/YamagishiMIT.pdf

Analyse & Kritik 26/2004 (© Lucius & Lucius, Stuttgart) p. 220-240

Coye Cheshire/Karen

S. Cook

The Emergence of Trust Networks under Uncertainty - Implications for Internet Interactions

Abstract: Computer-mediated interaction on the Internet provides new opportunities to examine the links between reputation, risk, and the development of trust between individuals who engage in various types of exchange. In this article, we comment on the application of experimental sociological research to different types of computermediated social interactions, with particular attention to the emergence of what we call 'trust networks' (networks of those one views as trustworthy). Drawing on the existing categorization systems that have been used in experimental social psychology, we relate the various forms of computer-mediated exchange to selected findings from experimental research. We develop a simple typology based on the intersection of random versus fixed-partner social dilemma games, and repeated versus one-shot interaction situations. By crossing these two types of social dilemma games and two types of interaction situations, we show that many forms of Internet exchange can be categorized effectively into four mutually exclusive categories. The resulting classification system helps to integrate the existing research on trust in experimental social psychology with the emerging field of computer-mediated exchange.

1. Introduction We investigate recent experimental work on social exchange networks, uncertainty, commitment and trust to analyze emerging forms of social interaction that are mediated by computers including systems like the Internet that link people across the globe. In particular we examine different types of exchange networks and the role of reputations and trust in these networks, commenting on the application of this research to various types of computer-mediated social interaction. We focus attention on the emergence of what we call 'trust networks' (Cook/Rice/Gerbasi forthcoming) to resolve some of the difficulties involved in anonymous exchanges of different types of goods and services. Trust networks are the egocentric networks of those one views as trustworthy. Computer-mediated exchange allows individuals to exchange with various degrees of anonymity through electronic networks. Since anonymity creates uncertainty and risk, resolving this uncertainty and developing ways to avoid risk are central to creating ongoing exchange environments. Although reputation systems are shown to be a well-supported solution to the inherent uncertainty and risk in this type of exchange situation, we conclude that reputation systems are only one mechanism for attempting to secure trustworthiness and reliability. We comment on other mechanisms that have been developed in efforts to ensure reliability, particularly when reputation information is not available.

The Emergence of Trust Networks under Uncertainty

221

To understand the emergence of trust networks in computer-mediated communication settings such as those on the Internet, we develop a typology based on the intersection of random versus fixed-partner social dilemma games, and repeated versus one-shot interaction situations. By crossing these two types of social dilemma games and two types of interaction situations, we show that many forms of Internet exchange can be categorized effectively into four mutually exclusive categories. Since similar categorization systems have been used in experimental social psychology, we relate the various forms of computer-mediated exchange to the selected findings from experimental research. We begin with a review of the pertinent findings from experimental social psychology on uncertainty and commitment in exchange relations.

2. Uncertainty and Commitment in Exchange Networks Experimental research on exchange relations in social networks indicates that under varying degrees of uncertainty actors begin to form commitments to their exchange partners when they believe there is risk in the environment, typically the risk of exploitation (i.e., Cook/Emerson 1978; Kollock 1999a; Yamagishi/Matsuda 2003; and Rice unpublished). This is a form of interpersonal commitment even though the actors may know very little about each other except their own history of exchange revealed over time through their interactions. This behavioral history and its veracity may be the most important source of reputation in networks of mediated interaction. Uncertainty concerning the quality of the goods at stake is one reason for commitment formation. When a buyer finds a supplier who is reliably trustworthy in terms of providing goods of high quality the buyer may try to enter a committed relationship with that particular seller. As Kollock's (1994) experimental research suggests those who locate partners who are trustworthy tend to form committed relations. In markets where quality can be determined more easily by the buyer (as is the case with the rice market, unlike the rubber market) actors are less likely to form commitments and thus they remain open to new exchange partners. Cook/Rice/Gerbasi (forthcoming) explore the consequences of commitment formation and the emergence of trust networks for the transition to open market economies. In Kollock's (1994) study of opportunistic uncertainty and commitment, actors exchanged goods in two different environments. In one environment (low uncertainty) the true value of the goods being exchanged was known, while in the other (high uncertainty) environment the true value of goods was withheld until the end of the negotiations. Actors had a greater tendency to form commitments in the higher uncertainty environment. They chose to continue to transact with the partners who had established their trustworthiness by not misrepresenting the value of their goods rather than to engage in potentially more profitable exchanges with new, untested partners. Yamagishi/Cook/Watabe (1998) further explored the connection between uncertainty and commitment. In their experiment, actors were faced with the de-

222

Coye Cheshire/Karen S. Cook

cision of remaining with a given partner or entering a pool of unknown potential partners. They employed several modifications of this basic experimental design, but in each instance the expected value of exchange outside the existing relation was higher than the returns from the current relation. They found that actors were willing to incur sizeable opportunity costs to reduce the risks associated with opportunism. Moreover, they found that uncertainty in either the form of an uncertain probability of loss or an unknown size of loss promoted commitment between exchange partners. Given high levels of uncertainty in most systems of Internet trade, commitments are likely to form in markets for certain types of goods and services. The most problematic trades for buyers are the one-shot interactions in which either there is no need for a repeat trade in the near future or it is unlikely that the actors involved would ever engage in repeat trade for a number of reasons (unrelated to the type of good or service involved).

3. The Emergence of Trust Networks under Uncertainty Our research indicates that commitments tend to form under uncertainty and that trust is higher among those who develop such behavioral commitments. However, it is also clear in our research that when trust is measured as an independent variable (as a predisposing factor) high trusters and low trusters tend to behave differently in the same environments. Under high uncertainty it is the low trusters who are the quickest to form interpersonal commitments. High trusters are more likely to be risk takers and to remain in the 'market' for new partners longer. In our most recent cross-cultural study of trust formation in the U.S. and Japan, we find that cooperation among individuals is significantly higher when individuals are given the opportunity to have repeat interactions with the same partners instead of being randomly paired with partners (Cook et al. forthcoming). In these experiments, individuals are given the ability to 1) control how much to entrust to a partner and, 2) decide whether to cooperate with the partner by returning (or not returning) the amount that was entrusted to them. When individuals only make a decision about whether to return coins or not (and the amount is decided by the computer), we consider this to be a basic prisoner's dilemma. When individuals have both of these choices, however, we call this a P D / R game, or, a prisoner's dilemma with risk. We found that participants significantly entrusted more money in the P D / R game (see Figure 1) and their cooperation rates were higher (see Figure 2) when they were paired with fixed partners instead of random partners (Cook et al. forthcoming). In the same study, we also examined the effect of signaling trust when matched with random partners. Thus, we randomly matched partners on every trial of the experiment, and in one situation individuals could control the amount to entrust to their partners and in the other situation they could not control this amount. We found that allowing individuals to signal their trustworthiness seems to have a positive effect on cooperation, but this positive effect is

223

The Emergence of Trust Networks under Uncertainty

• American fixedpartner PD/R • American randompartner PD/R • Japanese fixedpartner PD/R • Japanese randompartner PD/R

8

9

10

11

12

Trial Block Figure 1: Average number of entrusted coins across trial blocks in fixed versus random partner exchange using the prisoner's dilemma with risk (PD/R) in the U.S. and Japan (Cook et al. forthcoming).

minor and short-lived. In fact, the effect is not significantly stronger than in interactions between random partners who cannot signal each other (Cook et al. forthcoming). Furthermore, we found that there is no cross-cultural difference in levels of cooperation between American and the Japanese participants when they participate in random-partner exchange. In real-world situations in which trade partners are often initially unknown, having some system for establishing their trustworthiness is a primary concern, especially in settings in which it would be easy to cheat. Institutional backing, law and other devices for managing distrust or the breach of trust are crucial in the development of systems that manage trades, auctions and other forms of interaction involving the transfer of goods, services and money reliably. Computer-mediated exchange on the Internet, for example, is an environment in which trading and auctions have flourished—despite concerns about anonymity and the selection of trade partners at random initially. One reason for the success of many Internet exchange systems is that these systems have either adopted reputation systems, or such systems have evolved out of necessity. In the following sections, we discuss the emergence of reputation systems and the conditions under which different types of reputation systems are likely to work for mediated interactions (i.e. Internet trades). We also discuss alternatives to reputation systems, especially in exchange situations in which it is impractical or impossible to obtain reputation information.

224

Coye Cheshire/Karen

S. Cook

1

American fixed-partner PD/R « American random-partner PD/R Japanese fixed-partner PD/R — Japanese random-partner PD/R 0.2 0.1 -

0 1

2

3

4

5

6

7

8

9 10 11 12

Trial Block

Figure 2: Average cooperation rates (proportion of coins returned) across trial blocks for both American and Japanese participants using the prisoner's dilemma with risk (PD/R) with fixed and random partners (Cook et al. forthcoming).

The Role of Reputation

Systems

in

Exchanges

Yamagishi/Matsuda (2003) demonstrate that the role of reputations varies depending upon whether the social system in which the actors are embedded is closed or open. In closed communities reputations can be effective because negative reputations can lead to exclusion (Cook/Hardin 2001). In more open societies negative reputations are less effective because they are limited in the extent to which they are transferred to all those in the network (or social system) since information flows only across those actors that are linked directly or indirectly. More importantly, actors can alter their identities in ways that make it easier to reenter trade networks without being recognized as having had a negative reputation. The focus of Yamagishi/Matsuda's (2003) experimental study is an Internet trading network in which both the level of honesty in the trades and the price can be tracked. Positive—reputation mechanisms are more effective than negative—reputation mechanisms in open networks. In both cases (positive and negative reputation conditions) actors are rated on their honesty and thus accumulate reputation points that are published in the network after each transaction. In this research reputation mechanisms are investigated by Yamagishi and Matsuda (2003) as solutions to the 'lemons' problem in Internet markets, that is, the tendency for the goods on the market to be low quality especially when there is asymmetric information between buyers and sellers. Typically, the lemons market (Akerlof 1970) emerges when the buyers have much less valid or reliable information about the quality of the goods on the market than do the sellers (as is frequently the case in the used car market). Buyers must rely on the repu-

The Emergence of Trust Networks under

Uncertainty

225

tation of the sellers to determine how much confidence they have in the seller (i.e. how trustworthy the seller is). In this world of trade Yamagishi and Matsuda (2003) demonstrate that the lemons problem is exacerbated when actors can change their identities and reenter the market with a new identity (thus canceling their negative reputation). They also demonstrate that positive reputations affect behavior differently since actors develop an investment in their reputations and want to protect them. Traders lose their positive reputation and their investment in it if they alter their identities. In fact, the negative reputation mechanism designed to reveal dishonest traders is particularly vulnerable to identity changes, whereas the positive reputation mechanism designed to identify honest traders is not so vulnerable to identity changes. The results concerning the quality of goods produced in settings using these two different reputation mechanisms (with and without the possibility of changing identities in the exchange network) are presented in Figure 3 (see Yamagishi/Matsuda 2003 for details). The findings clearly demonstrate the superiority of positive reputation mechanisms especially when identities can be easily changed as is typically the case in Internet trading.

Figure 3: Average quality level of goods exchanged in positive versus negative reputation systems (Yamagishi/Matsuda 2003) To test the prediction that positive reputation devices are more effective than negative reputation mechanisms, Yamagishi/Matsuda (2003) conducted one experiment in which the quality level of the goods could be compared between exchange systems involving only positive or only negative reputations. The findings show that there is a significantly higher quality level, at least initially, when the negative reputation mechanism is used. Yet, as negative reputations grow, so do the tendencies to create new identities to shed the old negative reputation and begin anew. The positive reputation device proved to be better for producing higher quality goods in the long run when compared to a purely negative reputation system. Individuals became invested in their positive reputations.

226

Coye Cheshire/Karen

S. Cook

The negative reputation device may work best under the condition that someone who develops a negative reputation can be effectively excluded from trade by the group acting as a whole. This requires a closed network with information sharing.1 Norms of exclusion work only in groups and are not effective in open networks. They function to eliminate those who have 'cheated' from the system of trade (Hardin 1995; Cook/Hardin 2001). Kollock (1999a) views the auction houses that have been created for trade on the Internet as a laboratory for the study of the management of risk when there is little or no access to external enforcement mechanisms (the typical way we manage risk). He studies the emergence of endogenous solutions to the problems of risky trade in cases in which no guarantees, warranties, or other third-party enforcement mechanisms exist. The reputation systems that emerge to manage this risk are the primary focus of his research. Yamagishi and Matsuda (2003) explore the differences documented by Kollock between negative and positive reputation systems. Kollock (1999a, 111) notes that "a particularly disturbing strategy for fraud ... in a number of online markets is the person who works hard at establishing a trustworthy reputation, and then sets up a whole series of significant trades and defaults on all of them, disappearing into a new identity after the fact." In a quantitative study of auctions on eBay, Kollock (1999a) provides evidence that at least for some high-value goods, the price buyers paid was positively and significantly affected by the seller's reputation. In a simulation study of the effects of positive reputation systems that are similar in character to the positive reputations studied experimentally by Yamagishi and Matsuda (2003), Whitmeyer (2000, 196) finds that the effects of different types of positive reputation systems depend to a large extent on the proportion of cooperators (as opposed to non-cooperators) in the population. He examines the effects of reputation systems on general confidence gains in the society. If the proportion of non-cooperators is low, any type of positive reputation system will work because the non-cooperators will be more easily detected, especially if it is hard to obtain a positive reputation.2 If the proportion of noncooperators is high, then a tough reputation system will mean lost opportunities for cooperation because some potential cooperators will not be detected. Alternatives to Reputation Systems for Securing Trustworthiness In some exchange systems, it is not possible to establish reputation information, or it would at least not be feasible to do so. If an exchange system involves many one-shot interactions between random partners, it may not be possible to build a reputation (or the obstacles associated with regulating such a system may be too great). In such cases, there may be alternative types of mechanisms that emerge to provide security. The extent to which there is institutional backing for failed trust is what 1 See also Greif 1993 for a discussion of the difference between closed associations and open trade networks among the Maghribi. 2 Whitmeyer varied the degree to which it was easy or hard to get a positive reputation in his simulation.

The Emergence of Trust Networks under

Uncertainty

227

matters most in these environments independent of the quality or veracity of the reputation information involved in the decision of whom to trade with (or buy from), especially when such information is unreliable or impossible to obtain. For example, the Maghribi traders of the l l t / l century conducted trades across the Mediterranean despite the high level of uncertainty about their trading partners and, ultimately, the quality of the goods being traded (Greif 1989; 1993). Despite the incentives to cheat each other in these trades, the Maghribi were able to sustain successful trades. Greif (1989) explains this result by emphasizing the fact that the Maghribi traders were able to use closed networks that eliminated traders from their coalitions if they tried to cheat other traders. Community responsibility norms, as a kind of monitoring and policing device, developed as a way to secure trustworthiness within a circumscribed set of actors. The social contexts in which exchanges are embedded are thus crucial for understanding how trust and cooperation can develop, especially in the absence of reputation information. Social context refers to the nature of the situations in which social interactions take place (Cook/Cooper 2003). Factors such as group size and network density, for example, can have an independent effect on the development of trust. Systems that are too large can reduce the effect that trust has on cooperative behavior (Sato 1988). Furthermore, the very emergence of trust and cooperation may depend on locally embedded 'neighborhoods' that act as the source of cooperative and trusting behavior (Macy/Skvoretz 1998). Another important factor that affects cooperation and the emergence of trust is the ability to communicate with partners. The ability to communicate with one's partner before playing an experimental social dilemma game has a strong effect on specific trust of one's partner (Mori 1996). Other social context variables have been shown to affect levels of trust and cooperation, including time pressure, the presence of third parties, and culture. 3

4. Implications for the Internet and Other Forms of Computer Mediated Interaction Trust networks can take several different forms in computer-mediated interaction and on the Internet. Although the Internet is certainly the largest computermediated network, we argue that the processes by which trustworthiness is established in such networks can be informed by the experimental work on trust (e.g., Dasgupta 1988; Kreps 1990; Snijders 1996; Berg/Dickhaut/McCabe 1995; Yamagishi/Kakiuchi 2000). The experimental work in this tradition uses social dilemma games (such as the classic prisoner's dilemma) as tools for understanding the processes by which trust and trustworthiness are established under conditions of risk and uncertainty. A simple social dilemma game such as the prisoner's dilemma can be carried out as a one-shot game (in which the participants interact only one time), or as a repeated/iterated game (in which the participants interact several times). In 3 For a more complete review of experimental studies of social context and trust, as well as a review of other experimental research on trust and cooperation, see Cook/Cooper 2003.

228

Coye Cheshire/Karen

S. Cook

addition, an iterated social dilemma game may have fixed partners (i.e, the same partner on every turn), or random partners (a new, random partner on every turn). Since one-shot games only involve one interaction, the fixed/random partner distinction does not apply. Thus, the intersection of these two dimensions produces a 4-cell matrix (with one empty cell) which is useful for categorizing the various types of interaction situations found in the Internet environment. In the following sections, we describe these different types of interaction situations and some of the issues related to establishing trustworthiness under varying conditions. Table 1 includes examples of Internet interactions in each of the three main cells of our typology.4

Table 1: Examples of types of Internet situations with random and fixed partners 4 We view this typology as a useful tool for discussing the many types of interactions that take place on the Internet, yet we acknowledge that many of the examples in our typology can, and often do, take on different forms of interaction depending on the situation. As a result, the examples are not necessarily always one kind of interaction situation, nor do they always have fixed or random partners. It is precisely because of the malleability of interaction situations on the Internet that we find this kind of typology useful for describing any given situation.

The Emergence of Trust Networks under One-Shot Interactions with Random

Uncertainty

229

Partners

In an experiment involving a one-shot situation with random partners, individuals must simultaneously make decisions whether to cooperate. Because the interaction only occurs once (or there is only one trial), both parties can either cooperate or not, but trust would rarely, if ever, emerge since there would be no repeat interactions. An act of cooperation in a one-shot situation can be viewed as involving taking a risk, rather than as an act of 'trust' (see Hardin 2003). In a one-shot game there is no ability to send a signal to the partner by cooperating initially on several trials to signal trustworthiness. In computer-mediated communication, some types of Email interactions take this one-shot, random partner form. For example, suppose a professor receives an email from an unknown person requesting an unpublished draft of a current research paper. The professor does not know the person, so how does she know if the person is not going to plagiarize the paper, or quote it against the wishes of the author? A different example of this sort of uncertainty exists when an individual receives an email from an unknown person with an attached file. How does the recipient know that the attachment is not a computer virus? In both situations, social context cues (Sproull/Kiesler 1986) such as the email address of the sender can become essential to establishing trustworthiness. If we assume that the recipient does not know the sender, then seemingly trivial things like the domain of the sender (e.g., .edu, .com), the lack or presence of a personalized address in the email or the tone of the message can have a profound influence on the response of the recipient. In the absence of personal reputation information, such cues act as a proxy for reputation information when attempting to establish the trustworthiness of one's partner (see also Gambetta/Hamill forthcoming). Iterated Interactions with Random

Partners

Computer-mediated interactions on the Internet that involve iterated interactions give participants multiple chances to cooperate with their partners. However, if the interactions are always with new partners sometimes selected at random, then it is not possible to establish personal reputation information because the individuals do not have previous knowledge of the level of the other's willingness to cooperate. In other words, in such situations individuals do not have a 'shadow of the future' (Axelrod 1984) or any past that helps them determine what to expect from future interactions. Kollock (1999b) argues that reciprocity in online communication is encouraged when four conditions are met: ongoing interaction, identity persistence, knowledge of previous interactions, and strong group boundaries. While individuals may occasionally be purely altruistic, Kollock (1999b) argues that this is not a necessary assumption. In particular, some individuals may just want to cooperate out of a sense of self-efficacy, an expectation of future rewards, or as a way to improve their own reputation. However, when interactions are mostly random (i.e., identity is not persistent), then the last two explanatory factors (i.e., to obtain future rewards and to attain or produce a reputation) do not

230

Coye Cheshire/Karen

S. Cook

apply because there is no way to identify oneself to receive rewards from others and to build a positive reputation. On the Internet, one type of iterated interaction with random partners occurs in peer-to-peer exchange systems. In many popular peer-to-peer systems, individuals decide whether to share digital information goods, or just take such goods from others without sharing in return (Kollock 1999b). These digital information goods might take the form of digital media such as music or movies, or digital copies of written works such as electronic books or journal articles. Digital goods challenge many for-granted assumptions about the exchange of goods because digital goods can be copied from one person to another (i.e., the original owner does not lose value from the good during the exchange) and many people can benefit from the digital good without decreasing its value. Kollock (1999b) argues that digital goods produced by a single person on the Internet can become instant public goods because so many can benefit from them. These goods are easy to provide partially because they do not require coordination costs that usually prevent individuals from producing a public good on their own. The exchange of digital goods in peer-to-peer systems can be represented as a simple social dilemma in which an act of sharing digital goods is defined as 'cooperation' and taking digital goods without sharing is defined as 'defection'. Individuals can be thought of as random partners because a given provider is determined by what a user is looking for in the peer-to-peer network. For example, if an individual wants to find a specific song title, when she searches for this title she is given a list of individuals who have that item and are currently sharing it. Because there are often hundreds of thousands of users, and there is no specific reputation information available, each contributor can be thought of as a random partner. When an individual decides to retrieve a digital good from someone else, how does she know that the item will be what the provider says it will be? Furthermore, how can the person retrieving the digital good know that the item is not a computer virus or just bad data? In other words, just because an individual shares these goods, does not mean that she is trustworthy by default. There is some risk involved in downloading a digital good, such as a popular song, but the risk may be very small (i.e., involving only the time associated with downloading the file in question). As with many types of real-world exchange situations, a clear way to establish trustworthiness in an iterated set of online interactions is through reputation information. As Xiong and Liu (2003) argue, trustworthiness in peer-to-peer systems can be determined if a transaction-based feedback system is used. In their case, a 5-point trust model is used to create reputation scores that help establish trustworthiness. While such reputation systems are desirable, they are also not always practical and cannot be used in systems in which users are anonymous, or can easily change their identities (as in many existing peerto-peer systems). Thus, assuming that reputation systems cannot be used in iterated interactions among random partners; individuals must look for other devices for ensuring trustworthiness or reliability.

The Emergence of Trust Networks under

Uncertainty

231

One way individuals infer the trustworthiness of providers without reputational information is by observing how many other people are also retrieving the digital good from that same provider. Since many peer-to-peer systems (i.e., Kazaa, Morpheus, and the original Napster) allow users to see how many other anonymous individuals are retrieving digital goods from any given provider, each individual can potentially infer the trustworthiness of the provider from the actions of others. Such information provides a kind of 'virtual reputation' even if it is not specific enough. Of course, this may essentially be a conformity issue, because individuals may assume that if everyone else 'trusts' the contributor or finds her reliable, the good is probably legitimate. In the absence of personal reputation information, social conformity can end up providing a proxy for true reputation information. This type of conformity may represent a kind of herd behavior (sometimes seen in investment decisions), when a potentially infinite set of individuals make a similar decision (Smith/S0rensen 2000). Individuals who act in this manner do not necessarily ignore personal information. Instead, the behavior of other individuals may be the only information that is available to the individual. If an individual assumes that others may have more information than she does, then such 'herd' behavior is rational. Another way that individuals can infer the trustworthiness of a potential partner is through a third party that hits already become established as trustworthy. A contemporary example would be when well-known and established companies set up online stores for the buying and selling of used items. Amazon.com, for example, allows individuals to buy and sell used goods directly from their website—yet Amazon.com does not handle the inventory or the actual exchange of used goods. Instead, Amazon.com acts as a reliable or trusted third party intermediary that markets items for sale, and handles the collection of payments. Stefik (1999) argues that 'trusted systems' might be a solution to the problem of establishing confidence in digital goods and media. In this case, a trusted system might work like a bonded carrier, insuring the good and thereby reducing the risk that individuals face when exchanging digital goods (Lessig 2000). The key issue is that instead of the reputation of the individuals who are involved in a given exchange, it is the reputation of the third party that establishes trustworthiness or reliability. Thus, the risk associated with trusting other individuals is significantly reduced if some third party can either vouch for an individual, or possibly even underwrite or guarantee the transaction. Credit card markets work in this way in the consumer arena. Iterated Interactions with Fixed (or Consistent)

Partners

Iterated interactions among fixed (or consistent) partners on the Internet make up much of the existing research on Internet interactions and trust. Online auctions (Kollock 1999a; Yamagishi/Matsuda 2003) and virtual communities (Rheingold 1993; Curtis 1991; Abdul-Rahman/Hailes 2000) are often characterized by repeated interactions with fixed partners. In such cases, fixed partners have stable identities that allow individuals to have repeated interactions that

232

Coye Cheshire/Karen

S. Cook

can be compared to previous interactions. As a result, having reputation information is a clear advantage for individuals who want to determine the trustworthiness of partners in such systems (Kollock 1999a; Abdul-Rahman/Hailes 2000; Yamagishi/Matsuda 2003). An example of evolving virtual communities with iterated interactions among fixed (or consistent) partners is extremely large multiplayer games on the Internet. These games involve tens of thousands of players in a persistent online world. Players fight, trade, talk, and share experiences in these virtual online worlds. Raph Koster, a game developer for several of the world's largest multiplayer games argues that game developers should, "make sure that players have a reasonable expectation of future interaction. This means persistence of identity and limited mobility." (Koster 2003) The rationale is that players need reputation systems so that they know who they can and cannot trust. Because computer-mediated exchanges take place without the benefit of face-to-face communication, it becomes especially difficult for real people to know how how much to trust others through digital avatars (i.e., an icon or representation of a real person in a shared virtual world). Citing Yamagishi's and Matsuda's (2003) research on positive and negative reputation systems, Koster (2003) argues that these massively multiplayer games would be better served with positive reputation systems because, unlike negative reputation systems, they promote the persistence of identities. Another example of Internet exchange with partners who are fixed (or who are at least assumed to be fixed), are Usenet newsgroups. The Usenet describes a system of organized newsgroups, or bulletin boards, that are available to almost anyone with Internet access. The Usenet does not provide assurance of reliability or authenticity—anyone with an account can post to a newsgroup under any name she chooses. The stability of one's identity is often a key part of ongoing interactions in these newsgroups. It is argued that stable identities establish a sense of community (Beniger 1987), group affiliation and support (Sproull/Kiesler 1991; Wellman/Gulia 1999), as well as the basic motivation to actively participate in ongoing conversations (Donath 1999). The unregulated nature of Usenet newsgroups, however, also opens the possibility of identity switching and deception. As Donath (1999) describes in detail, a Usenet letter posted to any given newsgroup is filled with opportunities for deception and identity fabrication. A reader must take in several clues, including the email address of the sender, the content of the message, the language and tone of the message, and the signature of the sender. Is a given message really from a doctor if a message about medical advice is signed 'Dr. Smith'? Is it more or less credible if the supposed doctor's email address is from an official sounding domain? Unfortunately, there is little way to be sure that any message is reliable or legitimate. For example, the account name in the header or letterhead can be fake, identity claims can be completely false, social cues can be deliberately misleading or faked (see Donath 1999, 44; Gambetta/Hamill forthcoming). Although the kinds of Internet interaction situations described above can be viewed as repeated interactions with fixed partners, it is important to note that

The Emergence of Trust Networks under

Uncertainty

233

there are certainly exceptions to all of these examples. In fact, there is at least some research that shows that many types of Internet interaction situations are not always as 'iterated' as we might initially think. In a study on the quality of online social relationships, Cummings, Butler and Kraut (2002) show that some kinds of virtual communities such as newsgroups often have very little interactive communication at all between most members. Among the newsgroups that the researchers examined over a 130-day observation period, over half of the members did not contribute at all, and only a small number of the members contributed the majority of the messages (Cummings/Butler/Kraut 2002). So, while these kinds of Internet communities can (and often do) sustain ongoing interactions between individuals with fixed identities, a substantial number of interactions in these communities may actually consist of one-shot interactions between individuals who never communicate again.

5. Experimental Research on Trust and Implications for Research on the Internet Current findings in experimental studies of trust formation are useful for understanding how networks of trust can form in computer-mediated systems such as the Internet. A series of risk taking behaviors is central to building a trust relation (Blau 1964; Holmes/Rempel 1989). In a typical trust building scenario, two people realize they can potentially gain from engaging in social exchange. Each person knows that she will gain from engaging in an exchange with a potential partner if that partner turns out to be trustworthy. However, there is not always a way to know if a given partner is actually trustworthy. On the Internet, for example, one does not know for certain if the person one is chatting with is who they say they are (i.e., male or female). Similarly, if two individuals engage in a trade over the Internet, there may be limited assurances that the other person will honor the terms of a trade. Given such a risky situation, reputation becomes a useful and important tool for determining the trustworthiness of another person. This reputation may come in the form of either experience-based information (i.e., past experiences with this same individual), or third party-referenced information (i.e., reputation scores assigned by others who have interacted with the same individual) as noted above (Yamagishi/Matsuda 2003). Risk-taking and One-Shot Interactions with Random

Partners

If individuals have only one-shot interactions with one another and they are always new, random partners, trust cannot emerge. Even reputation information may not be possible to obtain in such a system because the interactions are always random. One solution is the use of formal or informal controls (Heckathorn 1993), yet such controls may not be possible either due to the costliness of maintaining them, or the impracticality of establishing them in open systems like those found on the Internet. For example, in the earlier example of an unso-

234

Coye Cheshire/Karen

S. Cook

licited email request, there are often no informal or formal mechanisms by which monitoring can be provided. This fact makes such interactions highly risky. In a general study using simulations, Macy/Skvoretz (1998) show that cooperation and 'trust' can emerge, despite the lack of formal or informal control mechanisms, but only under certain conditions. Although the authors use the term 'trust' to refer to the willingness to participate in a risky activity (such as playing a PD game with a random partner), this is better referred to as risktaking behavior (Hardin 2003) rather than trust. This research underscores the importance of distinguishing risk-taking from trusting behavior, especially in one-shot random partner situations where actual trust-building is not possible. Macy/Skvoretz (1998) find that the emergence of cooperation among random partners in one-shot interactions ultimately depends on two structural conditions. First, individuals must have the option to not participate (or exit). Not only does an exit strategy better approximate the real world (since real people often have the ability to not interact at all if they so choose), but it also tends to encourage risk-taking if the option to exit is relatively costly. The second condition they identify is the importance of embedded social networks. Macy/Skvoretz (1998) demonstrate that cooperation emerges locally among 'neighbors' in an exchange system. In this case, the term "neighbors" refers to the proximity of potential interaction partners in the system. Only after individuals cooperate in these local networks does cooperation disseminate through weak network ties. Thus, even if trust is not theoretically possible in this kind of one-shot random partner interaction situation, their research shows that the development of cooperative behavior is still possible under certain structural conditions. Trust-building Among Fixed Partners in Repeated

Interactions

In many situations there is often no regulatory agency to control behavior and little initial information about the reputations of possible exchange partners. Instead, individuals must determine the trustworthiness of their partners through experience. The process of building experience-based reputations is itself, risky—because each experience (i.e. interaction) offers a chance for one of the partners not to cooperate, or to betray the partner's trust. For example, in an online discussion among potential dating partners, each person takes a risk when they share increasingly personal information about themselves with the other person. Should they reveal their ages? Should they use their real names? Each decision is based on the results of previous interactions. As mentioned above, our recent cross-cultural experiments on trust formation in the U.S. and Japan show that cooperation among individuals is significantly higher when individuals are given the opportunity to have repeated interactions with fixed partners instead of dealing with random partners (Cook et al. forthcoming). We also found that participants significantly entrusted more money when they were paired with fixed partners instead of random partners, and the cooperation rates were also higher when they were paired with fixed instead of random partners.

The Emergence of Trust Networks under

Uncertainty

235

In the preceding example, individuals are simply not given the opportunity to exit the relationship. So, how does the situation change when individuals interact with a system of participants rather than a set of dyadic relationships? Yamagishi and Matsuda (2003) explore several questions about the trustworthiness of exchange partners in computer-mediated interactions by creating an Internet 'auction' in a laboratory setting. The authors find that both experience-based and third-party reputations are successful in promoting trustworthy behavior in an exchange system. 5 However, these reputations are only effective so long as the identity of the individuals is permanent. The ability to change one's identity (a common feature in many forms of exchange on the Internet), undercuts the effectiveness of reputation systems by allowing individuals to cheat each other and then to start over with a new identity once their reputations become degraded. In sum, both experience-based and third-party reputation information are essential for building trust relationships in computer-mediated interactions. While such reputation information is clearly important in the real world as well, it seems particularly crucial in online interactions that lack many of the cues that accompany real-world interactions. Trust-building Among Random Partners in Repeated

Interactions

What if individuals are either randomly matched or their identities change so that reputation information cannot be established as with repeated, fixedpartner exchange? When individuals are unable to use experience-based or third-party reputation information to secure their interactions, it becomes particularly difficult to ascertain the trustworthiness of their partners. And, it is especially difficult to facilitate the development of trust relationships through risk taking when there is no "shadow of the future" (Axelrod 1984). Trusting random partners, or those who can change their identity, begins to seem more like gambling than rational decision making. According to the encapsulated interest view of trust (Hardin 2002), such behavior would be defined as risk-taking, not trust. We have mentioned that peer-to-peer exchange with digital goods represents one kind of Internet exchange situation where random partners can have repeated interactions. However, the apparent Utopia of cooperative sharing in peer-topeer systems may largely be artificially created by forcing users to share their own digital goods (Bricklin 2001). There are very real costs (i.e., time, network bandwidth, security) that might lead individuals to refrain from sharing digital goods if they had the choice. However, all of the most popular online fileswapping systems have this built-in feature: users are not asked whether they want to share their files. Instead, the files are automatically shared whenever someone downloads a digital good. As a result, the problem of coordinating cooperation among file-swappers is essentially eliminated. 5 Yamagishi and Matsuda 2003 measured trustworthy behavior by examining the degree to which 'sellers' in their laboratory auction would sell goods at higher prices than they were worth. Since the true value of the goods was not known until after the transactions, these exchanges were analogous to many of the existing Internet auctions in which individuals buy and sell goods without inspecting the goods first.

236

Coye Cheshire/Karen

S. Cook

If peer-to-peer systems did not artificially create this "cornucopia of the commons" (Bricklin 2001), would individuals still share digital goods? The answer to this question is not yet clear—but the long history of research on public goods and exchange does indicate that individuals do not, on the whole, contribute without some incentive for doing so. One possible motivation could be social psychological selective incentives that further reduce the already minor costs associated with sharing digital goods (Cheshire 2004). Selective incentives have been very successful as an economic solution to free riding in the production of a public good because they offer additional motivations that make cooperation rational even when the initial conditions of the social dilemma make cooperation irrational (e.g., Olson 1965; Oliver 1980; Heckathorn 1996). Individuals may be encouraged to cooperate (i.e. share digital goods) if they receive social psychological rewards such as social approval and a sense of group solidarity when they contribute. In an experimental study of digital goods exchange, social psychological selective incentives are shown to have a significant influence on the likelihood of sharing, despite the costs associated with sharing and the lack of any specific reputation information about others (Cheshire 2004). In peer-to-peer Internet exchange and other types of situations in which random partners have repeated interactions, one of the key problems is the lack of reputation information. However, just because individuals cannot determine the reputation of a given partner does not mean that they cannot use other information or other tools to try to build trust. For example, we argue that the signaling role of trusting behavior is still important for building trust—even when individuals exchange with random partners. By acting in a trusting manner, one can signal her intention to cooperate. In a given system, this ability to signal may be particularly noticeable because it would be unexpected. For example, individuals who would probably prefer to cooperate with an exchange partner rather than to defect on her will often defect anyway because they expect that others are not willing to cooperate. These individuals may defect simply out of a fear of being exploited by others, not necessarily because of their own greed (cf. Pruitt/Kimmel 1977; Yamagishi/Sato 1986). When individuals have random partners, there can also be a kind of 'secondorder' fear that others will be similarly fearful and thus they might defect for the same reason. It is this 'second-order' fear that trusting behavior can lessen. For example, if two individuals meet on an Internet message board for bookswapping and decide to exchange books with each other, how do they know that they will each follow up on their agreement to mail the books (e.g., trusting behavior)? Furthermore, how can they be sure that the other person will return the book when they are finished (i.e., cooperative behavior)? When one of the individuals sends a book, it acts as a signal to the other person. By taking a risk and acting in a trusting manner, an individual signals to her partner that she expects cooperation, or, that she is not afraid that her partner will defect on the exchange. Thus, the act of trusting may help eliminate this second-order fear in her partner (see also Cook et al. forthcoming). In our cross-cultural experiment on trust formation, we found that allowing individuals to signal their trustworthiness seems to have a positive effect on

The Emergence of Trust Networks under Uncertainty

237

cooperation, but the effect is small and does not seem to last very long. In addition, we found that there is no cross-cultural difference in cooperation between the American and the Japanese participants when they participate in randompartner exchange. Thus, we do not find strong support for the argument that signaling is a mechanism for increasing cooperative behavior and trust in systems with random partners (Cook et al. forthcoming), but this is a fairly hostile environment for trust at the dyadic level. Further research varying the nature of the situation is needed to assess the importance of signaling at the population level.

6. Conclusion Computer-mediated exchanges offer many new ways to look at social exchange and the development of cooperative behavior and trust. Many recent experimental studies on social exchange networks, uncertainty, commitment and trust are particularly useful for understanding the emerging forms of social interaction that exist in computer-mediated situations like those found on the Internet and other mediated systems of trade. Since computer-mediated exchanges provide new possibilities for anonymity in different types of exchange situations, resolving the uncertainty and risk that are created by this anonymity is essential for creating ongoing, relatively secure exchange environments. As recent experimental research has shown, the development of trust networks (i.e. networks of those one views as trustworthy) can help resolve some of the difficulties involved in anonymous exchanges of different types of goods and services, but such networks have relatively high information requirements. Across the various types of exchange networks, reputational information has consistently proven to be a key factor for establishing cooperation and facilitating the development of trust. However, other information such as the social context, community responsibility norms, or social psychological selective incentives can play an important role in trust-building and insuring reliability even when personal or third party reputation information is not available. As exchange on the Internet and other forms of computer-mediated interaction continue to evolve, ongoing experimental research in sociology and related disciplines offers insights into the structural and behavioral factors that influence the development of trust in these environments. We have offered a classification system that helps to link the types of exchange situations that are found in computermediated systems such as the Internet with the interaction situations that are the subject of much experimental research on cooperation and trust formation in the laboratory. Future research on the Internet and the many types of social and economic exchanges now being mediated through computerized connections will help inform not only theories of exchange, but also theories of the role of trust in the economy. In this case the real world has become the laboratory.

238

Coye Cheshire/Karen

S. Cook

Bibliography Abdul-Rahman, A./S. Hailes (2000), Supporting Trust in Virtual Communities, in: Proc. of SSrd Hawaii Int'l Conf. on System Sciences. 6007 Akerlof, G. A. (1970), The Market for Lemons: Quality Uncertainty and the Market Mechanism, in: Quarterly Journal of Economics 84, 488-500 Axelrod, R. (1984), The Evolution of Cooperation, New York Beniger, J. (1987), Personalization of Mass Media and the Growth of Pseudo-Community, in: Communication Research 14, 352-371 Berg, J . / J . Dickhaut/K. A. McCabe (1995), Trust, Reciprocity, and Social History, in: Games and Economic Behavior 10, 122-42 Blau, P. M. (1964), Exchange and Power in Social Life, New York Bricklin, D. (2001), The Cornucopia of the Commons, in: Andy Oram (ed.), Peer-toPeer: Harnessing the Power of Disruptive Technologies, Sebastopol, 59-66 Cheshire, C. (2004), A Sociological Analysis of Generalized Information Exchange, Ph.D. Dissertation Cook, K. S./R. M. Cooper (2003), Experimental Studies of Cooperation, Trust, and Social Exchange, in: E. Ostrom/J. Walker (eds), Trust and Reciprocity: Interdisciplinary Lessons from Experimental Research, New York, 209-244 — / R . M. Emerson (1978), Power, Equity and Commitment in Exchange Networks, in: American Sociological Review 43, 721-739 — / R . Hardin (2001), Norms of Cooperativeness and Networks of Trust, in: M. Hechter/K. D. Opp (eds.), Social Norms, New York, 327-347 — /E. R. W. Rice/A. Gerbasi (2004), The Emergence of Trust Networks under Uncertainty: The Case of Transitional Economies-Insights from Social Psychological Research, in: S. Rose-Ackerman/B. Rothstein/J. Kornai (eds.), Problems of Post Socialist Transition: Creating Social Trust, New York, 191-221 — / T . Yamagishi/C. Cheshire/R. Cooper/M. Matsuda/R. Mashima (forthcoming), Trust Building via Risk Taking: A Cross-Societal Experiment, in: Social Psychology Quarterly Cummings, J. N./B. S. Butler/R. Kraut (2002), The Quality of Online Social Relationships, in: Communications of the ACM 45, 103-108 Curtis, P. (1991), Mudding: Social Phenomena in Text-Based Virtual Reality, Electronic manuscript (FTP: parcftp.xerox.com), 1-21 Dasgupta, P. (1988), Trust as a Commodity, in: D. Gambetta (ed.) Trust: Making and Breaking Cooperative Relations, Oxford, 49-72 Donath, J. (1999), Identity and Deception in Virtual Community, in: P. Kollock/M. Smith (eds.), Communities in Cyberspace, New York, 29-59 Ekeh, P. (1974), Social Exchange Theory: The Two Traditions, Cambridge Foa, U. G./E. B. Foa. (1974), Societal Structures of the Mind, Springfield Gambetta, D./H. Hamill (forthcoming), Streetwise: How Taxi Drivers Establish Customers' Trustworthiness, New York Greif A. (1989), Reputation and Coalitions in Medieval Trade: Evidence on the Maghribi Traders, in: Journal of Economic History 49, 857-882 — (1993), Contract Enforceability and Economic Institutions in Early Trade: The Maghribi Trader's Coalitions, in: American Economic Review 83, 525-548 Hardin, R. (1995), One for All, Princeton — (2002), Trust and Trustworthiness, New York

The Emergence

of Trust Networks

under

Uncertainty

239

Heckathorn, D. D. (1993), Collective Action and Group Heterogeneity: Voluntary Provision versus Selective Incentives, in: American Sociological Review 58, 329350 — (1996), The Dynamics and Dilemmas of Collective Action, in: American Sociological Review 61, 250-277 Holmes, J. G./K. Rempel (1989), Trust in Close Relationships, in: C. Hendrick (ed.), Close Relationships, Thousand Oaks, 187-220 Kollock, P. (1994), The Emergence of Exchange Structures: An Experimental Study of Uncertainty, Commitment, and Trust, in: American Journal of Sociology 100, 313-345 — (1999a), The Production of Trust in Online Markets, in: Advances in Group Processes, 99-123 — (1999b), The Economies of Online Cooperation: Gifts and Public Goods in Cyberspace, In: P. Kollock/M. Smith (eds.), Communities in Cyberspace, London, 220-239 Koster, R. (2003), Small Worlds: Competitive and Cooperative Structures in Online Worlds, conference talk, Game Developer's Conference, San Jose Kreps, D. M (1990), Corporate Culture and Economic Theory, in J. E. Alt/K. A. Shepsle (eds.), Perspectives on Positive Political Economy, 90-143, Cambridge Lessig, L. (1999), Code and Other Laws of Cyberspace, New York Macy, M. W . / J . Skvoretz (1998), The Evolution of Trust and Cooperation Between Strangers: A Computational Model, in: American Sociological Review 63, 638-660 Mori, K. (1996), The Effects of Trust and Communication on Cooperative Choice in a Two-Person Prisoner's Dilemma Game, in: Japanese Journal of Experimental Social Psychology 35, 324-36 Oliver, P. (1980), Rewards and Punishments as Selective Incentives for Collective Action: Theoretical Investigations, in: American Journal of Sociology 85, 1356-1375 Olson, M. (1965), The Logic of Collective Action. Public Goods and the Theory of Groups, Cambridge Pruitt, D. G./J. J. Kimmel (1977), Twenty Years of Experimental Gaming: Critique, Synthesis, and Suggestions for the Future, in: M. R. Rosenzweig/L. W. Porter (eds.), Annual Review of Psychology 28, 363-392 Rheingold, H. (1993), The Virtual Community, New York Rice, E. R. (2002), The Effect of Social Uncertainty in Networks of Social Exchange, Ph.D. dissertation Sato, K. (1988), Trust and Group Size in Social Dilemmas, in: Japanese Psychological Research 30, 88-93 Smith, L./P. Sorensen (2000), Pathological Outcomes of Observational Learning, in: Econometrica 68, 371-398 Snijders, C. (1996), Trust and Commitments, Interuniversity Center for Social Science Theory and Methodology, Utrecht Sproull, L./S. Kiesler (1986), Reducing Social Context Cues: Electronic Mail in Organizational Communication, in: Management Science 32, 1492-1512 Stefik, M. (ed.) (1999), The Internet Edge: Social, Technical, and Legal Challenges for a Networked World, Cambridge/MA Takahashi, N. (2000), The Emergence of Generalized Exchange, in: American Journal of Sociology 105, 1105-1134 Wellman, B./G. Milena (1999), Virtual Communities as Communities: Net Surfers Don't Ride Alone, in: P. Kollock/M. Smith (eds.), Communities in Cyberspace, New York, 167-194

240

Coye Cheshire/Karen

S. Cook

Whitmeyer, J. M. (2000), Power Over Groups Through Effective Monitoring and Sanctioning, conference presentation: American Sociological Association Xiong, L./L. Liu (2003), A Reputation-Based Trust Model for Peer-to-Peer eCommerce Communities, Unpublished Manuscript, 2003 IEEE Conference on E-Commerce (CEC'03), Newport Beach Yamagishi, T./M. Matsuda (2003), The Role of Reputation in Open and Closed Societies: An Experimental Study of Online Trading, Center for the Study of Cultural and Ecological Foundations of Mind, Working Paper Series 8 — /K. S. Cook (1993), Generalized Exchange and Social Dilemmas, in: Social Psychology Quarterly 56, 235-248 — /K. Sato (1986), Motivational Bases of the Public Goods Problem, in: Journal of Personality and Social Psychology 50, 67-73 — / R . Kakiuchi (2000), It Takes Venturing Into a Tiger's Cave to Steal a Baby Tiger: Experiments on the Development of Trust Relationships, in: W. Raub/J. Weesie (eds.), The Management of Durable Relations, Amsterdam, 121-123

Analyse & Kritik 26/2004 ( © Lucius & Lucius, Stuttgart) p. 241-257

Anabel Quan-Haase/Barry

Wellman

Local Virtuality in a High-Tech Networked Organization*

Abstract: What are networked organizations? The focus of discussions of the networked organization has been on the boundary-spanning nature of these new organizational structures. Yet, the role of the group in these networked organizations has remained unclear. Furthermore, little is known about how computer-mediated communication is used to bridge group and organizational boundaries. In particular, the role of new media in the context of existing communication patterns has received little attention. We examine how employees at a high-tech company, referred to as KME, communicate with members of the work group, other colleagues in the organization, and colleagues outside the organization to better understand their boundary-spanning communications.

1. The Debate about New Organizational Structures Are new forms of 'networked organization' arising, driven by the proliferation of information and communication technologies (ICTs)? Do employees have the trust to communicate over boundaries using ICTs? We use evidence from a case study of a high-tech networked organization to see if communication in the organization largely stays within work groups (as in traditional organizations) or crosses boundaries to elsewhere within the organization or to other organizations. Recent writings on organizational structure emphasize the emergence of new forms of work (Monge/Contractor 1997; 2003). Generally, two interrelated trends are seen as leading to these changes. The first trend is a shift from an emphasis on systems of production and related manufacturing processes to an economy based on information and the management of intellectual capital toward the development of services and innovation (Andersen/Howells/Hull/Miles/Roberts 2000; Drucker 1993; Quinn 1992). The second trend is the increased reliance on ICTs for information transfer at * Research underlying this chapter was supported by Communication and Information Technology Ontario, the Centre for Urban and Community Studies, the IBM Institute of Knowledge-Based Organizations, and Mitel Networks. Anabel Quan-Haase acknowledges assistance from the Alumni Research Awards Program, Faculty of Social Science, the University of Western Ontario. Barry Wellman acknowledges assistance from the Social Sciences and Humanities Research Council of Canada. We thank Lynne Howarth and Joseph Cothrel for their advice and Julie Wang for her editorial assistance. We especially want to thank all the employees at KME who completed the survey, and even more so, those employees who also participated in the interviews and observations.

242

Anabel

Quan-Haase/Barry

Wellman

all levels of the organization. ICTs, and in particular computer-mediated communication (CMC), are serving as a technological driver to create and maintain electronic networks of information exchange. Indeed, Castells (1996) argues that the high reliance on networks has revolutionized the economy and the structure of society creating a social order with the capacity to connect as a unit in real time on a global scale. Even though we do not feel as strongly, we believe that the ubiquity of ICTs is a key feature of organizations that are part of the new economy and see their mandate as providing services based on information and innovation. These two trends have led to organizations where information represents a key asset and the flow of information becomes critical for success (Choo 1998a; 1998b; Davenport/Prusak 1997; 2000). In these new forms of work, have the structure of organizations, and in particular the communication structure, fundamentally changed? Analysts have suggested that these trends result in networked organizations: communication structures based on electronic networks where information flows flexibly and spans group and organizational boundaries (e.g., Heydebrand 1989; Jarvenpaa/Ives 1994; Miles/Snow 1992; Nohria/Eccles 1994; Sproull/Kiesler 1991; Ward/Wamsley/Schroeder/Robins 2000; Wellman 1997). They contrast such networked organizations with bureaucratic organizations based on preInternet forms of information exchange. Bureaucratic organizations are efficient in an environment where the transfer of information is slow and problems are well-defined and fairly routine (Baker 1992; Heckscher 1994; Heydebrand 1989; Miles/Snow 1992; Nohria/Berkley 1994; Nohria/Eccles 1994; Ward et al. 2000). By contrast, the networked organization may have better means than hierarchical, place-based organizations for pooling decision-making and problem-solving resources (e.g., Castells 1996; Nardi/Miller 1991; Sproull/Kiesler 1991). Networked organizations fit the paradigm of the information society because they have the flexibility to acquire information from anywhere and at any time, as barriers of space and time are no longer perceived as impediments for communication (Castells 1996). In the networked organization, loosely coupled structures form, with ties to people outside the work group as well as to outside the organization (Ahuja/Carley 1999; Alstyne 1997; Jarvenpaa/Ives 1994). Peter Monge and Noshir Contractor define the networked organization as "a collection of organizations along with the linkages that tie them to each other, often organized around a focal organization. There are numerous variations on the network organizational form including joint partnerships, strategic alliances, cartels, R&D consortia, and a host of others." (Monge/Contractor 1997, 463) Similarly, for Manuel Castells, the key feature of the networked organization is the linkages that tie them to other organizations through ties such as alliances, partnerships, and information-exchanges (1996). These boundary-spanning ties make them more open systems whose boundaries are more permeable to information from the outside. They function as interconnectors between multiple networks, providing access to new information and more creative problem solving (Jarvenpaa/Ives 1994). For example, Robin Teigland (2000) has shown that

Local Virtuality in a High-Tech Networked

Organization

243

boundary-spanning information exchanges in a Scandinavian high-tech firm led to higher levels of creativity. The information that these high-tech workers obtained from online communities of practice increased their performance at work. The underlying assumption in these discussions of the networked organization is that geographical proximity, collegiality, and group membership does not bound communication. In the networked organization, groups are one special type of social network, but not necessarily the dominant structure for communication (Wellman 1997; 2001). For example, cross-functional teams or brainstorming groups are created for only a limited period of time and then cease to exist (Hargadon/Sutton 1997a; 1997b). For accomplishing their tasks, employees rely on information from outside their group and outside their organization. For Castells, the shift from firms as bounded units to the networked enterprise is a necessity because "[coopération and networking offer the only possibility to share costs, and risks, as well as to keep up with constantly renewed information. ... Inside the networks, new possibilities are relentlessly created. Outside the networks, survival is increasingly difficult." (Castells 1996, 171) While there has been much optimism around the value of the networked organization for information flow, collaboration, and innovation, few studies have documented the boundary-spanning nature of today's organizations. The metaphor of the networked organization does not provide further detail as to the extent to which group-based communication has changed and the nature of these changes. Are employees bridging group and organizational boundaries as the networked organization perspective predicts? Is boundary-spanning communication occurring at the expense of local, group-based communication? Extant research has focused on the use of online communities of practice for obtaining information but has not directly compared communication and media use at various distances. One exception is Hinds and Kiesler's examination of the technical and administrative communication patterns of employees in a large organization (Hinds/Kiesler 1995). They studied employees' communication frequencies at three distances: "departmental within the chain of command" ; "departmental out of the chain of command" ; and "extradepartmental". More than half of all communication in this study was extradepartmental supporting the notion of networked organizations, where employees contact individuals outside the work group for advice, information, and problem solving. Hinds and Kiesler also investigated what technologies were used for communication at each distance. The authors compared the use of the telephone, voice mail, and email for communication. When employees communicated outside the chain of command and outside the department, they relied more frequently on the telephone than on email. By contrast, email was used primarily for communication within the department and within the chain of command. Hinds and Kiesler's study provides support for the shift from traditional, densely knit organizations toward the sparsely knit, networked organization. As organizations become more globally oriented and deal with customers and part-

244

Anabel

Quan-Haase/Barry

Wellman

ners worldwide, we expect the trend toward networked organizations to continue. Although Hinds and Kiesler found that employees used the telephone and not CMC for boundary-spanning communication, they suggest that " . . . comfort and convenience would be expected to shift the use of technology towards asynchronous modes as people communicated with others further away and across time zones" (HInds/Kiesler 1995, 389). Their study suggests that boundaryspanning communications will continue to rise, and that CMC will play a key role in supporting these communications. However, Hinds and Kiesler did not collect data on face-to-face (FTF) exchanges and suggest that future research should directly compare FTF exchanges with CMC. Furthermore, as organizations introduce new technologies such as instant messaging (IM) in the work place, employees might come to rely more heavily on CMC for boundary-spanning interactions. In this paper, we examine how high-tech employees communicate at different distances to determine the extent to which boundary-spanning communication is occurring. This examination provides a more comprehensive view of groupbased and boundary-spanning communication in the networked organization. Instead of defining communication boundaries and distance in terms of hierarchy and location (Hinds/Kiesler 1995), we focus on communication within the work group, elsewhere in the organization, and outside the organization. In this way, we focus specifically on communication distance as a structural boundary inherent in the functional division of the organization. In the networked organization information flows easily across boundaries and the work group is of less relevance for getting work done. We treat communication distance as proximity because work group members are located more closely to one another than they are to people elsewhere in the organization, and much more distant from communication partners outside the organization. We believe that a shift has occurred toward sparsely knit, networked organizations. Hence, we hypothesize that a large proportion of communication will take place with people outside of the work group. Furthermore, we investigate the role of CMC in supporting boundary-spanning communication. We build on existing research by comparing CMC with FTF and telephone exchanges. Existing studies have mainly focused on email and little research has compared the use of email to IM. In many organizations employees are now collaborating via IM, either as a complement to email or a replacement (Handel/Herbsleb 2002; Herbsleb/Atkins/Boyer/Handel/Finholt 2002). Hence, we include IM sis one more means of communication that allows for convenient and cost effective exchanges at a distance. While there is agreement that CMC supports boundary-spanning communication, theories of social presence and media richness suggest limitations inherent in CMC. CMC has less ability to convey information about a person's characteristics (e.g., gender, age, and social status) and no ability to convey information about a person's bodily expressions. Indeed, early researchers have argued that CMC is inappropriate for information exchanges when messages are complex and equivocal (Daft/Lengel 1984; 1986; Daft/Lengel/Trevino 1987; Huber/Daft 1987; Lengel/Daft 1988; Brown/Duguid 2000; Cohen/Prusak 2001). Although

Local Virtuality in a High-Tech Networked

Organization

245

laboratory experiments have called this argument into question, there is a need for field data gathered in real organizations. We hypothesized that co-located employees would continue communicating via F T F and the telephone. By contrast, distributed workers for whom F T F is not an option would rely on CMC. Therefore, we expected to find F T F and the telephone used for group-based, local communication and CMC used for boundary-spanning, global communication where FTF is not an option. Thus, we see the role of CMC as an alternative means of communication that leads to boundary-spanning ties and helps bridge organizational and group boundaries. Our findings come from a case study of a high-tech firm that is representative of the workings of networked organizations. Knowledge Management Enterprises (a pseudonym) is a mid-sized organization located at the heart of one of North American's high-tech hubs. KME was founded in 1997 and expanded during the technology boom. The company's main business consists of a software development group, which is responsible for updating and improving KME's software. The company also has a management services group, which is responsible for managing the online services offered by the company. Like many start-ups in the high-tech sector, KME operates in a more horizontal communication structure. As a company founded after widespread diffusion of the Internet and related technologies, it is a good prototype for an investigation of the communication structures of networked organizations.

2. KME: A Case Study for Understanding Networked Organizations We collected data on communication patterns and media use at Knowledge Management Enterprises. We selected KME as the site for the study because it is a software and services company that routinely uses CMC as part of its everyday life. KME started when Internet technologies were widely available in the United States, and it has incorporated CMC into its operations from its founding. Hence, KME is not an organization that first operated before the Internet and then adapted to CMC. Moreover, KME was a useful research site because employees were co-located at a single place, while at the same time conducting business on a global scale as its customers and users were located across North America, Europe, and Asia. This allowed examining boundary-spanning and local communication simultaneously. At KME, we studied the software development and management services groups, whose 28 employees comprise 35 percent of the total workforce. These two groups have existed as functional units for at least one year, and the tasks accomplished within each group are interrelated and fairly homogeneous. Hence, we expected to find stable patterns of communication and use of CMC. Twentyseven employees completed the questionnaire, yielding a 96 percent response rate. Before administering the questionnaire, we conducted a pilot test with 6 respondents. There were 11 (3 women) participants in the software development group and 16 (5 women) in the management services group. Respondents

246

Anabel Quan-Haase/Barry

Wellman

had worked for K M E for an average of 28 months (range: 5-48 months). Six respondents had a high school diploma or less, 12 respondents had completed an undergraduate degree, and eight respondents had a graduate degree. T h e data on education is missing for one respondent. The sample included 3 upper managers, 5 middle managers, and 19 group members. T h e survey included questions about employees' communication frequency at each of the three distances of interest: group-based, with other colleagues in the organization, and with colleagues outside the organization. 1 Within the work group, employees had a total of 11 and 16 potential communication partners in the software development and management services groups, respectively. Members of the software development group and the management services group had ties elsewhere in the organization with 69 and 64 potential communication partners, respectively. 2 For each distance, participants were asked to indicate - on a scale ranging from 1 = never to 7 = daily - how frequently they used the following three media: (1) F T F and the telephone; (2) email; and (3) instant messaging (IM). For each distance and medium, the mean rate of communication was computed across both groups. The 7-point scale was recoded into a scale reflecting 'days per year' to provide a numeric representation of the frequency of use that was easier to comprehend. For example, the category '1/week' was recoded into '52 days per year.' 3 While the 'days per year' scale was used to depict the data in graphs and tables (i.e., for the descriptive statistics), the original categorical scale was used for all the statistical tests (i.e., inferential statistics). To supplement the survey data, we did in-depth interviews and observations with a sub-sample of 10 employees. Interviews lasted about 45 minutes and the observations lasted for an entire workday (approximately 9 AM to 5 P M ) . To aid anonymity, we refer to respondents by pseudonym, and we omit hierarchical position and other identifying characteristics.

1 In the questionnaire, participants also reported on their use of information sources, and their social and communication networks. The analysis in the present paper focuses only on the communication questions. See Quan-Haase/Cothrel 2003 for analysis on the use of information sources data. 2 This study focused on frequencies of communication only. Therefore, no data are available about the overall number of communication partners outside KME. 3 The original 7-point scale was transformed into days per year: 'never' = 0; 'a few times a year' = 5; '1/month' = 12; '1/week' = 52; 'several times a week' = 130; '1/day and several times a day' = 365.

Local Virtuality in a High-Tech Networked Organization

247

Table 1: Items Measuring Demographics and Communication Frequencies Education What is the highest level of education you have attained? Scale: l=High school diploma; 2=Undergraduate degree; 3=Graduate degree Time at KME How long have you worked with this organization? Scale: Years and months Face-to-Face and Telephone Calls For each of the following personal information sources, indicate how frequently you use offline (e.g., meetings, phone calls, lunches, etc.) means of communication for work-related purposes. Colleagues within the work group. Colleagues outside the work group but within KME. Colleagues outside organization. Scale: l=never; 2=a few times a year; 3=l/month; 4=l/week; 5=several times a week; 6=1/day; 7=several times a day. Email How often do you send/receive emails from colleagues within the work group? How often do you send/receive emails from colleagues within your organization, but outside the work group? How often do you send/receive emails from colleagues outside of your organization? Scale: l=never; 2=a few times a year; 3=l/month; 4=l/week; 5=several times a week; 6=l/day; 7=several times a day. Instant Messaging How often do you use instant messaging to communicate with colleagues within the work group? How often do you use instant messaging to communicate with colleagues within your organization, but outside the work group? How often do you use instant messaging to communicate with colleagues outside of your organization? Scale: l=never; 2=a few times a year; 3=l/month; 4=l/week; 5=several times a week; 6=l/day; 7=several times a day.

248

Anabel Quan-Haase/Barry

Wellman

3. Over what Boundaries are High-Tech Employees Communicating? Is communication at KME spanning group boundaries? CMC facilitates communication at a distance overcoming barriers of space and time. CMC allows for easy and quick information exchange across group boundaries connecting colleagues anywhere in the organization. However, what remains unclear is the extent to which CMC facilitates boundary-spanning communication within the organization. Is boundary-spanning communication occurring to such an extent that it replaces group-based communication? Although a large proportion of communication is with colleagues elsewhere in KME, with a mean of 178 communication days per year, most communication continues to be within the work group, with a mean of 285 days per year (see Figure 1). Even in this high-tech organization, where employees have diverse CMC tools available to them for boundary-spanning communication, they continue to exchange information primarily with other group members. The ratio of communication with other colleagues in the organization to group-based communication is 0.62, 4 suggesting that communication with other colleagues in the organization has not replaced group-based communication.

Figure 1: Communication by Distance

If communication has bridged the group boundary and includes other organizational members, is it also spanning the organization's boundary? Considering 4 These ratios were obtained by calculating the proportion of communication between distances. For example, the ratio 'Colleagues Inside Organization/Work Group' is 178/285=0.62. In this example, the mean days per year communication with colleagues elsewhere in the organization is divided by the mean days per year communication within the work group.

Local Virtuality in a High-Tech Networked

Organization

249

that KME is a global organization with clients and partners worldwide, we expect that frequent communication would occur at a distance. However, we do not expect that communication outside the organization would be larger than group-based communication or communication with other colleagues in the organization. We expect that the fact that people are co-located would facilitate local communication. Work communication with people outside of KME occurs at a mean of 65 days per year. This is less communication than with group members (285 days per year) and colleagues elsewhere in KME (178 days per year). The discrepancy between group-based communication and outside communication is stronger than is the case for organizational communication. Much more communication takes place at the group level than with outside colleagues. Thus, even at KME, where business is conducted globally, the group continues to be the key unit for communication. The ratio of communication with members outside the organization to group-based communication is 0.23, showing that most communication at KME occurs within the work group. Similarly, the ratio of communication with members outside the organization to communication with members of the organization is 0.37, suggesting that the amount of communication decreases considerably from internal to external. To compare communication inside and outside the work group, the scores for communication with other colleagues in the organization are pooled with those for communication outside the organization. This yields a mean score for overall outside the work group communication. Even when comparing communication inside the group versus outside the group, group-based communication is more frequent, with a ratio of 0.85. The ubiquity of CMC and its boundary-spanning capabilities do not seem to replace group-based communication with boundaryspanning communication. By contrast, the data show that even when all outside communication are lumped together, group-based interactions are still higher. Despite this high level of within-group interaction, the data show some support for viewing KME as a networked organization. A large proportion of all communication takes place with people outside of the group: both with other colleagues in the organization as well as outside the organization. This suggests that CMC helps overcome group and organizational boundaries without making group and organizational boundaries obsolete. Although propinquity may constrain organizations less today, the group in high-tech KME continues to be the primary source for information and communication.

4. What Media are Used for Different Organizational Involvements? Although most communication continues to take place within the boundaries of the group, a sizeable amount of communication takes place elsewhere in the organization and even outside of the organization. What media are supporting boundary spanning? Theories of the networked organization argue that CMC encourages communication across boundaries because it provides a meeting space

250

Anabel Quan-Haase/Barry

Wellman

for people with shared interests that overcomes the limitations of space and time (Baym 1995; 1997; Sproull/Kiesler 1991; Wellman/Gulia 1999). In turn, boundary-spanning communication with multiple, distant others fosters sparsely knit, networked organizations. While we agree with analysts who claim that CMC has seriously affected the organization of many forms of work—especially where employees work with bits rather than atoms—we wonder about the nature of these transformations. We wonder if: 1. Does the preference for CMC only apply to geographically-distant communication partners? 2. Do co-located employees communicate primarily via F T F and the telephone because these media are richer and better support complex and equivocal messages (Daft/Lengel 1984; 1986; Daft et al. 1987; Huber/Daft 1987; Lengel/Daft 1988)? 3. If F T F communication were not an option because of constraints of distance and time, then would CMC be the most frequently used medium? By examining these three questions, our research provides a better understanding of how high-tech employees use CMC and traditional media for communicating at different distances. The K M E data show that communication within the local group relied more on CMC (email and IM) than on traditional media ( F T F or the telephone). Repeated-measures analysis of variance (3x3 MANOVA) confirms these findings, with distance and type of media as the within-subjects factors. The analysis yields a significant effect of distance, F ( l , 2 6 ) = 66.82, p < .001, and of type of media, ,F(1,26) = 15.60, p < .001. There is no significant interaction of distance and type of media, (1,26) < 1) refers to the depreciation of the value of peer approval that is due to online interaction. Equation (4) expresses the benefits from social rewards An that worker i derives from the outcome of iteration t of the game. N

Ait = Y^ i=i

fajitihlj

+ (1 - hi])w).

(4)

To include effects of task uncertainty, we assume that local workers always have perfect information on the true effort investments of all other local team members in the past. By contrast, a virtual team member only knows the outputs (1 —rrij)wj that his colleagues j generated in all previous iterations, but he does not know their true effort investments and whether a mishap occurred. Correspondingly, the colleagues of a virtual team member i also only know the output generated by i but they are not aware of i's true effort. Decision making on basis of reinforcement

learning

To model workers' decision making in the repeated game, we employ a backwardlooking learning model that assumes that actors follow a simple decision heuristic (cf. Flache/Macy 1996; Flache 1996). In this model, actors optimize by learning and adaptation rather than by calculating the marginal return on individual investment. In other words, actors adjust both their effort level and their attitudes toward other members in response to social cues that signal whether the investment was worthwhile. The actors thus influence one another in response to the influence they receive, creating a complex adaptive system. Such systems

How May Virtual Communication Shape Cooperation in a Work Team?

265

lend themselves more readily to computational rather than analytical models (Axelrod 1997; for 'backward-looking' computer simulations of collective action and social exchange, see Macy 1989; 1990). The computational model consists of three basic components: a decision rule, a reward function by which outcomes are evaluated as satisfactory or unsatisfactory, and a learning rule by which these evaluations modify choice propensities. We describe these components in turn. Decision rule The decision process is stochastic rather than deterministic. Learning actors follow choice propensities that are altered after they experience the consequences of their behavior. The stochastic decision rule assumes that each actor i has some propensity pit representing the probability that i will work at time t(wlt = 1). With probability 1 — pu, i will shirk (wlt = 0). Similarly, plJt represents the probability that i will approve of j at time t(aijt = 1). With probability 1 —p'ijt, i will not approve of j{a,ijt = 0). Reward function: the evaluation of work and approval Actors economize in our model on cognitive effort with three shortcuts: reliance on propinquity as a lowcost proxy for causality, 'satisficing' as a lowcost proxy for the identification of global optima, and separate evaluation of decisions as a proxy for analysis of the joint effects of simultaneous actions. Reliance on propinquity is modeled by the 'law of effect': when an action is associated with a satisfactory outcome then increase your probability to repeat the action. Conversely, when a dissatisfactory result obtains, decrease the probability to repeat the associated behavior. 'Satisficing' (Simon 1982) implies that the better the outcome, the more likely the actor will deem it to be 'good enough' rather than risk an inferior result by searching for something better. The poorer the outcome, the more likely the actor will be to take the risk. For simplicity, the model formalizes this by evaluating outcomes relative to the midpoint of the payoff distribution. Finally, separate evaluation of actions implies that actors adapt their propensities to work or to approve separately per decision, based on their satisfaction only with some components of the outcome of the preceding iteration. Satisfaction with the work decision increases with the actor's share of the total group wage and decreases with the costliness of the decision. In addition, local workers always evaluate their local colleagues' true effort investment rather than their output, because even under task uncertainty it is seen as success when a colleague tried to cooperate. Moreover, effort is deemed worthwhile to the extent that it is associated with high levels of approval. More formally, N

Sit = Wit+li-Yl

hm3wi 3=1

+ Ait -

CWit -

Se,i.

(5)

Su is Vs satisfaction with current work effort, such that the sign of Stt indicates positive or negative evaluation. Wu and An are defined in (2) and (4)

266

Andreas Flache

above. It is important to note that the effect that others' work has in comparison with others' approval on an actors' satisfaction with his work decision is scaled by the parameters a and ¡3. These parameters enter equation 5 indirectly via the terms for wage payment, Wu, and approval from peers, Ait, respectively. The N

term U f j ^ IjirijUij in equation (5) is needed to correct for the difference in the j=i situation of local workers compared to virtual team members. This term assures for local workers that not only the outputs but also the true cooperation of their local colleagues is taken into account. Se,i is the reference point that determines whether actor i evaluates the outcome of iteration t as success or as failure. For simplicity, we position for each actor the reference point at the midpoint of the range of possible payoffs that this actor can obtain. This implies that local team members have a higher expectation level than virtual team members, because local team members can obtain a higher maximal social reward level than their virtual colleagues. Moreover, in their expectation level team members take the number of local and virtual colleagues into account. Equation 6 specifies the ensuing reference points for satisfaction with the work decision. i(a + ( N - N V - l + NVw)ß-c),

if/» = l (6)

l(a + { N - I p p - c ) ,

if Zi = 0

The decision by i to approve of j, aijt, is evaluated in the same way except that the collective action problem is now disaggregated into a matrix of dyadic games, one for each of the possible dyads. Rather than taking into account overall group effort and overall approval received from the group, i considers only i's benefit from j's effort and the approval received from j. Since i's benefit when all N members work is a , i's benefit when one individual j works is a/N. We use the evaluation of j's approval in our model to incorporate the distinction between normative and affective dependence of workers. Workers can obtain behavioral confirmation from their peers mainly for their investments in work effort, because with respect to the decision to work or shirk all team members have a clearly defined common interest in high effort of their colleagues. Technically, normative dependence is expressed with the parameter p in equation (6) that scales the extent to which peer approval reinforces an actors' work decision. Affective dependence refers to the extent to which team members wish to be embedded in ties of mutual social approval ('friendship ties') with their colleagues, independently from their work behavior. A worker's desire for affective rewards can not be satisfied by his colleague's obedience with collective work norms, but it can be satisfied when the worker receives approval from a colleague in response to his own relational investment into the tie with the colleague. Accordingly, we use the symbol P' to distinguish affective dependence from normative dependence p. Affective dependence shapes the extent to which a worker Ego deems it as a success of his approval decision with respect to a particular colleague Alter, when Alter approved of Ego. Like in the evaluation of the work decision, approval that is expressed in online interaction is less valuable

How May Virtual Communication Shape Cooperation in a Work Team?

267

to the recipient than approval received in face-to-face interaction, scaled with the online-depreciation parameter ui. In addition, we assume that the working situation always affects at least to some extent workers' social interactions. That is, Ego always deems it as success of his approval decision when Alter invested in work effort. Finally, the evaluation of approval decisions also takes into account that local workers can see the true effort levels of their local colleagues and take these true efforts as basis for success or failure of their own approval decisions. In this, the aggregate benefits Ego obtains in his relationship with Alter are evaluated against the relationship-specific expectation or reference point S'e i j :

S'ijt = jj(hh

+ (1 - M;)(l - mj))wJt + {hi, + (1 - hlj)u;)f}'ajit

- S'eAj.

(7)

According to (5) and (7), the parameters ¡3 and ¡3' represent the weight i places on others' approval relative to their work effort. With (3 = ¡3' = a/N, i values others' face-to-face approval equally with work effort both in evaluating his work decision and in evaluating his approval decisions. With /? = / ? ' > a/N, others' approval of i has a stronger effect on i's satisfaction than work effort. Conversely, with /? = / ? ' < a/N others' work effort has a stronger effect on both i's work behavior and his investment in social relations than others' approval of i. The difference between normative dependence ¡3 and affective dependence (3' expresses whether peer approval mainly affects an actors' evaluation of his work behavior, or whether approval decisions are also evaluated in terms of approval received. With 0 > ¡3' > 0, workers deem approval more as normative confirmation of their work decision than as affective reward for their social behavior. With P > ¡3' = 0, they even give approval exclusively in response to the effort decisions of their peers. Finally, with /? = / ? ' > 0, actors make their approval decisions contingent on both work behavior and relational investments of their peers, a situation in which social interactions in the group are related to both work and social ties. We exclude the possibility that relational dependence exceeds normative dependence (/? < /?'), because this seems implausible for the situation of a professional team where wage payments are based on collective work output. S'e i j , finally, is the relationship-specific reference point that corresponds to the midpoint of the reward distribution in the evaluation of approval. Again, the maximum satisfaction that can be obtained in a relationship between two local workers is higher than in a relationship with at least one virtual team member. Equation (8) specifies the corresponding expectation levels: i ( % + n s,« H

,

a 1^ = 1 otherwise

w

The evaluation of each decision is transformed into a positive or negative reinforcer constrained to the interval [1,+1]. Let Rxt be the reinforcer corresponding to Sit- Then

268

Andreas Flache

Rit =

(9)

where r is a learning parameter that scales the magnitude of reinforcement and 5 m a x , j is the highest possible absolute value that the worksatisfaction Su can take for worker i. Analogously the reinforcer for the approval decision is obtained as follows R'at =

•

(io)

Learning function The final component of the model is the learning function by which propensities to work and approve are modified by satisfaction or dissatisfaction with the outcomes associated with those behaviors. The learning function is adapted from a conventional Bush-Mosteller stochastic learning model (Bush/Mosteller 1955). Actor ¿'s propensity to work, pu, is reinforced when effort seems to pay (•wu — 1 and Su > 0) or when shirking is costly (Wu = 0 and Su < 0):

Pi,t+1 = Pit + Rit{ 1 - Pit)wit - Rit{ 1 - Pit) (1 - wit).

(11)

The benefits of hard work and the costs of free-riding are indicated in the equation by two adjustments to pu, one positive (when wu = 1) and the other negative (when wu = 0). Hence, the reward to workers is added to the propensity when wu = 1, while the penalty for shirkers is subtracted when w l t = 0, causing the propensity to increase in either case. Conversely, if feckless behavior pays off or hard work is suckered, then the propensity to shirk (1 — pu) is reinforced, i.e., 1 — pu is substituted for pit on both sides of the equation and 1 — wlt is substituted for Wu, giving

Pi,t+1 = Pit + RitPitWu ~ RitPiti 1 - wu)The propensity for approval p y t is modified in the same way. If S'l]t > 0, or aijt and S'l]t < 0, then:

(12) alJt = 1 and

Pij,t+1 = Pijt + R'ijti 1 - Pijt) a'jt ~ R'ijti 1 - Pijt) (1 - aijt)-

(13)

Conversely, if a,j t = 0 and S[jt > 0, or alJt = 1 and 5,' Jt < 0, then:

Pij,t+1 = Pijt + RijtPijtaHt ~ R'ijtP'ijt (1 - aijt)-

(14)

The terms that moderate the effect of the reinforcement, (1 — p) in equations (11) and (13) and p in equations (12) and (14), assure that propensities remain within the valid range [0..1]. This implies in particular that learning curves tend to flatten when the same action is repeatedly rewarded (S > 0) or repeatedly punished (S < 0) and propensities approach their limit. However, with the high learning rates that we choose for our simulation experiments (see below), this does not imply inertia in the learning process. Particularly in the region

How May Virtual Communication Shape Cooperation in a Work Team?

269

close to the boundaries, learning is highly sensitive to adverse experience. The higher the propensity for a certain action, the stronger is the effect that negative reinforcement (S1 < 0) has on the corresponding propensity. Conversely, the lower the propensity for a certain action, the stronger is the effect that positive reinforcement (5 > 0) has.

4. Results We use simulation experiments to explore the three mechanisms through which virtual communication may shape the performance of work teams, task uncertainty, normative tie strength and affective tie strength. In all simulation experiments, we vary systematically the level of average team virtuality V of teams of 10 members (7V=10) from purely local teams (V=0) to purely virtual teams (V=l), in steps of 0.1. More precisely, the simulations use 100 replications per level of virtuality, where V indicates the probability that a particular team member is a virtual member. Per level of team virtuality, we measure the level of cooperation as the average propensity in iteration 1000 of a team member to work hard (u>=l). Furthermore, we use in all simulations the starting conditions that the maximum wage is fixed at unity (a = 1), workers are initially indifferent between working and shirking (pto = 0.5) as well as between approval and shunning in all dyadic relationships (pJ -0 = 0.5, i ^ j). This indifference may seem a rather pessimistic assumption, as there is rarely a company where on average only half of the employees work. However, we believe initial indifference is an appropriate starting assumption for our analysis, because we want to explain how workers learn to cooperate rather than assume that cooperation has been established from the outset. Moreover, as Macy (1989) showed, the equilibria to which the stochastic learning model converges in the long run are robust against variation of initial propensities within a large range. Finally, we assume that all workers have a learning rate of r=0.75, a condition that makes behavioral responses sufficiently flexible to allow the emergence of stable cooperation through random coordination (cf. Flache/Macy 2002). Simulation experiment 1: the interaction of task uncertainty and team virtuality Task uncertainty may undermine trust and thus cooperation between team members, and virtual communication may even exacerbate the trust problems generated by task uncertainty. Particularly in teams where task uncertainty may result in unintended failures by members to deliver their contributions to the work task, colleagues may more readily interpret this as a deliberate free riding and, as a consequence, withhold their own efforts for the group task. To test whether our model can reproduce this expectation, we simulated a team situation in which cooperation based on mutual reinforcement is possible but not trivial in a purely local team. For this, we assumed that costs of effort are more than twice as large as the marginal individual gains of a work contribution (c = 0.25 > 1/N — 0.1). To focus on effects of virtual communication on pure

270

Andreas Flache

task reciprocity, we set both normative and affective dependence to zero and thus excluded social pressure (/? = /?' = 0). Finally, to explore the interaction of team virtuality with task uncertainty, we varied task uncertainty between low (e = 0.01), medium (e = 0.05) and high (e = 0.10). Figure 1 shows the results.

Figure 1: Interaction of team virtuality with task uncertainty e on level of cooperation. Average cooperation rate measured as average individual propensity to work in iteration 1000, based on 100 replications (N = 10, a = 1, c = 0.25, ¡3 = (3' = 0,r = 0.75). Figure 1 confirms2 the expected interaction between task uncertainty and team virtuality. The simulation parameters are tuned such that in a purely local team (V=0), task reciprocity can sustain an average cooperation rate in iteration 1000 of about 70%. From this starting condition, increasing team virtuality reduces cooperation rates downto about 40% in a fully virtual team under high task uncertainty. The negative effect of team virtuality is somewhat weaker under medium task uncertainty, with a decline from 70% cooperation to about 50%, and it is hardly discernible in teams with low task uncertainty. The underlying reason is the distortion of cooperative equilibria when task uncertainty makes intended cooperations fail. To explain, previous simulation studies with the same learning model and similar starting conditions showed that without task uncertainty, cooperation will sooner or later emerge and stabilize through random coordination of a critical mass of contributors. With the parameters underlying Figure 1 (N = 10, a = 1, and c = 0.25), the only stable equilibria of the individual learning processes are situations in which seven or more team members work. Only then all team members, including shirkers, can be satisfied with their work decisions and their learning curves will stabilize on the corresponding outcome after a sufficient number of repetitions (cf. Macy/Flache 2002). 2

We are aware that this is a purely theoretical statement. To avoid complicated prose, we use in the following terms as 'confirm' and 'test' in the sense of 'show that the formal model is consistent with the informal reasoning'.

How May Virtual Communication Shape Cooperation in a Work Team?

271

Task uncertainty can distort equilibria of the learning process. Moreover, the likelihood and frequency of such distortions increases with team virtuality. When for some worker a mishap occurs, this reduces for all virtual colleagues the satisfaction with their work decision, because they interpret the lack of contribution as a defection. Moreover, when it was a virtual worker who failed, then also all his local colleagues interpret the failure as a defection. Thus, even when everybody is satisfied because there are seven or more contributors, the expected share of team members who may become dissatisfied with their work decision after some unintended defection is V+(1 — V)V, a number that increases with team virtuality V. The learning mechanism implies increased likelihood for behavioral change after dissatisfaction. After a mishap, the majority of team members who work hard may at least temporarily reduce their propensity to work and explore defection. After a while the critical mass of contributors may be restored and stabilize, until the next distortion occurs. This explains the interaction effect in figure 1. The larger the probability of unintended defections (task uncertainty) and the larger the share of team members who become dissatisfied after such a distortion (team virtuality), the more likely it is that emergent cooperation may collapse temporarily, or even fail to stabilize at all when task uncertainty becomes too large. Simulation experiment 2: the interaction of online-tie strength and team virtuality without affective dependence. A common argument in the study of virtual teams is that trust may be fostered by deliberate offline meetings that increase the strength of ties between team members who otherwise interact only online. In terms of our model, such policies reduce the extent of online depreciation UJ of the value of peer approval. To test whether our model can replicate the expected interaction between offlineembeddedness of virtual communication and team virtuality on cooperation, we adapted the starting conditions of simulation experiment 1 such that cooperation is much harder to attain without social control. For this, we increased the costs of effort to one third of maximum wage, i.e. c = 0.33, as compared to one fourth in experiment 1 (c = 0.25). To test whether peer pressure is needed in this new scenario, we kept all other conditions equal to experiment 1 and simulated the cooperation rate in the fully local team (V = 0). As expected, the cooperation rate dropped to 34.2% from the level of about 75% shown in figure 1. We created a situation in which informal social control can considerably improve cooperation between team members in offline interaction. Based on previous work we assumed that the normative reward value of a unit of approval is about twice as large as the marginal benefits of a work contribution (cf. Flache/Macy 1996), i.e. (3 = 0.2. To focus furthermore in this simulation experiment on the motivational aspect of interpersonal relationships alone and let aside information exchange, we assumed that task uncertainty plays no role in the simulated team (e = 0). Our main interest in this experiment is the potential of social relationships to mediate informal social control. Accordingly, we neglect the possibility that ties may be used for the mere production of pos-

272

Andreas Flache

itive affections and assume zero affective dependence, i.e. ¡3' = 0.0. All other assumptions are equal to those used in simulation experiment 1. To test the interaction of offline-embeddedness with team virtuality, we varied the online-depreciation of tie strength w between low, medium and high depreciation, where a unit of approval received in online communication has 75%, or 50%, or 25% of its rewarding value in face-to-face interaction, respectively (LJ = 0.75, LJ = 0.50, LJ = 0.25). Figure 2 below shows the results. Cooperation

Figure 2: Interaction of team virtuality with online depreciation of tie-strength (w) on level of cooperation with high normative dependence (¡3 = 0.2) and no affective dependence (/?' = 0). Cooperation rate measured as average individual propensity to work in iteration 1000, based on 100 replications (N = 10, a = 1, c = 0.33, /3 = 0.2,0' = O,e = 0.0, r = 0.75). Figure 2 shows a cooperation rate of about 80% in a fully local team (V = 0), where all social ties have their full strength (w = 1) by definition. Social control more than doubled the cooperation rate as compared to the benchmark simulation that we conducted without normative dependence {¡3 = 0.0) and otherwise equal conditions. But figure 2 also shows that team virtuality undermines social control. At all three levels of the depreciation of tie strength, the cooperation rate considerably declines with the degree of team virtuality. When online ties have 75% of the strength of offline ties (w = 0.75), cooperation drops from the rate of 80% in fully local teams (V = 0) to about 60% in fully virtual teams (V = 1). As expected, the decline is much sharper when online ties are weaker relative to offline relationships. With u> = 0.50, the cooperation rate in fully local teams exceeds that of fully virtual teams by about 45% and with ui = 0.25 this difference is even somewhat larger. To understand how weaker ties erode social control in the learning model, consider Flache/Macy's (1996) explanation of why social control effectively sustains cooperation in the absence of virtual communication. Like in experiment 1 in the fully local team, cooperation is achieved through a self-reinforcing learning dynamic where actors' random search generates by chance an outcome that satisfies all group members with all their decisions. There is only one way how

How May Virtual Communication Shape Cooperation in a Work Team?

273

all group members can be simultaneously satisfied with both types of decisions in a fully local team under the parameters used in experiment 2. This is when everyone works hard and all group members approve of their peers. If some group member takes it easy in his work, then all others will remain dissatisfied with their approval decision with respect to the shirker and keep changing their approval behavior. As a consequence, the shirker will sooner or later experience low levels of approval and be dissatisfied with his corresponding work decision. This, in turn, leads shirkers to change their work behavior so that at some point group members coordinate on the equilibrium of hard work combined with mutual approval. The level of 80% cooperation in fully local teams (V = 0) in figure 2 indicates that in iteration 1000 this point has on average not yet been reached for 20% of the group members. However, further simulation tests confirmed that the cooperation rate always increases up to 100% after a sufficient number of iterations (typically about 5000). The larger the proportion of weak online ties, and the weaker these ties are as compared to face-to-face contacts, the more this may change the dynamics of backward-looking social control. For all group members, higher levels of team virtuality reduce the benefits from social rewards that actors can obtain in the evaluation of their work decisions (equation 4). For local team members, the approval of their virtual colleagues becomes less valuable, and for virtual team members it is the approval of every other team member that is less highly regarded. This also reduces actors' corresponding expectation level. As a consequence, the higher the degree of team virtuality, the smaller the impact that changes in the amount of approval for an actor can have on the actor's satisfaction with his effort investments. In other words, team virtuality reduces the power of the group to make workers satisfied and shirkers dissatisfied with their effort investments. This effect is particularly strong for virtual group members. To be sure, team virtuality does not change the basic assumption in experiment 2 that approval decisions are considered by every actor as dissatisfactory as long as they are directed towards a shirker. However, team virtuality makes shirkers less vulnerable to social pressures. In particular, the larger the proportion of weak virtual ties, the larger is the possible number of shirkers who may be shunned by most or even all of their colleagues, but who still remain satisfied with their shirking behavior. They remain satisfied, because they put much more weight on the wage benefits generated by their working colleagues than on their loss of social status in the group. This explains why in figure 2 cooperation rates consistently decline with increasing team virtuality and why this effect becomes stronger when online ties become relatively weaker. Simulation experiment 3: the interaction of online-tie strength and team virtuality with affective dependence. In the third simulation experiment we tested how team virtuality may interact with the negative effect of strong ties proposed by Flache/Macy (1996). The possible weakness of strong ties identified by Flache and Macy suggests an interesting new possibility for the effects of virtual communication on cooperation

Andreas Flache

274

on work teams. Greater social distance of online interactions may avoid possible negative effects of social cohesion in those teams where members are highly affectively dependent on their peers. As a consequence, weaker ties in online interaction may promote cooperation in the work task, because they make it harder for affectively dependent team members to be satisfied by social contacts alone. To test this argument, we adapted the conditions of simulation experiment 2 such that without affective dependence cooperation can thrive even in virtual teams. For this, we reduced the costs of work effort to c = 0.15, a level at which the group faces a moderate social dilemma. Furthermore, we kept the assumption of experiment 2 that normative pressure is a powerful motivator (/3 — 0.2), but online communication considerably reduces tie strength both normatively and affectively {uj = 0.25). Finally, to test how affective dependence may change the effects of team virtualization, we varied affective dependence j3' between a low, medium and high level, where the high level corresponds to a situation in which affective rewards are valued equally to normative rewards, while at the low level affective rewards are only 25% as valuable as normative rewards. That is, we used in the simulations /3' = 0.05, ¡3' = 0.1 and (3' = 0.2, respectively. All other conditions were kept equal to simulation experiment 2. Figure 3 shows the results. Cooperation 1.00

Affective dependence:

ß' ~

0.20

ß' = 0 10

/?'= 0.05 Team virtuality V 0 00

000

0.20

040

0.60

080

1 00

Figure 3: Interaction of team virtuality with affective dependence (/?') on level of cooperation with high normative dependence (/3 = 0.2) and high rate of online-depreciation of tie-strength (w = 0.25). Cooperation rate measured as average individual propensity to work in iteration 1000, based on 100 replications {N = 10, a = 1, c = 0.15, £ = 0.0, r = 0.75). Figure 3 confirms the expected interaction of affective dependence with team virtuality on cooperation. More precisely, the figure shows that when actors' affective dependence is low (/3' = 0.05), team virtuality reduces cooperation from a level of almost 100% in a fully local team (V = 0) down to about 70% in a fully virtual team (V = 1). Higher levels of affective dependence invert the effect of team virtuality. With ¡3' = 0.1, team virtuality slightly increases

How May Virtual Communication Shape Cooperation in a Work Team?

275

cooperation levels from a rate of about 58% in a fully local team (V = 0) up to approximately 70% in a fully virtual team. This effect is even more pronounced when affective dependence is high (/?' = 0.2), with a rise in cooperation from 40% in local teams (V = 0) to 70% in virtual teams (V = 1). The negative effect of team virtuality under low affective dependence in figure 3 is generated by the same mechanism that shaped the negative effects of V displayed in figure 2. In a fully local team and with affective dependence much lower than normative dependence ((3 = 0.2 = 4 * /?'), the unique equilibrium in which all players can be simultaneously satisfied with their work decisions and their approval behavior remains the state of universal cooperation and universal mutual approval. Just like without affective dependence, the learning process sooner or later converges on this equilibrium. Furthermore, as team virtuality increases, both normative and affective rewards loose their gratifying value in an increasing proportion of relationships in the team, and social control becomes less effective as a solution to the free rider problem. When affective dependence exceeds the critical level of 0' = 0.1, social control is even in fully local teams no longer effective. Equations 7 and 8 imply that now new equilibria in the learning mechanism arise in which all group members can be satisfied with their approval decisions as long as approval is reciprocated, regardless of the work effort of their colleagues. At the same time, universal mutual approval undermines the need to work, due to the still high normative value of peer approval, f3. With sufficient affective dependence, this peer approval may now in equilibrium also be given to shirkers. On the other side, with ¡3 = 0.2, sufficient approval can make both workers and shirkers be satisfied with their work decision, even when most group members fail to contribute. Figure 3 shows the consequence of this, the level of only 40% cooperation in a fully local team with high affective and normative dependence (/? = /?' = 0.2), as compared to almost 100% cooperation in a fully local team with low affective dependence ( P = 0.2,/?' = 0.05). In this condition, an increasing level of team virtuality fosters rather than undermines cooperation. The reason is that with sufficient online-depreciation of tie strength, mere mutual approval is no longer enough to satisfy group members with their online social relations with shirkers. As a consequence, as time virtuality rises, an increasing proportion of the group is required to work hard before the learning process obtains an equilibrium state in which all members can be satisified with both work and approval decisions. As figure 3 exemplifies, this results in a positive effect of team virtuality on cooperation rates in those conditions where affective dependence is medium or high.

5. Discussion and Conclusion We have proposed in this paper a formal model of the effects of virtual teamwork on trust and cooperation in a team. With this, we move beyond the existing literature in the field that is mainly based on case studies or employs theoretical arguments that lack explicit deductive model building. Our model

276

Andreas Flache

incorporates three mechanisms identified in previous research that may underlie effects of team virtuality on cooperation at the workplace, lack of social context information in virtual communication, weaker ties in online interaction and lower affective dependence of team members on peer approval in weak online relationships. Computer simulations revealed conditions that interact with the effects of increasing team virtuality. Our model implies that task uncertainty greatly exacerbates the negative consequences of lack of social context information on trust and thus cooperation between team members. Moreover, the simulations suggest that negative effects of the weakness of ties in online communication on informal control may be reduced when companies adopt policies that reduce the extent to which online interaction depreciates the gratifying value of peer rewards. This formal result replicates the emphasis that the literature puts on the need to flank virtual teamwork with regular offline meetings between team members. Finally, our analysis suggests a new hypothesis not recognized by previous research. Particularly in primarily local teams, social control may badly fail to sustain cooperation when team members are overly dependent on their peers for affective rewards, such as friendship relations. In such teams, higher visualization of teamwork may facilitate cooperation, because with weaker ties workers are less prone to tolerate free riders for the sake of maintaining their social relations with them. We believe that testing implications of our theoretical work may be a fruitful direction for experimental research on effects of virtual team work. Flache (1996) used an experimental exchange game with virtual communication between participants to test the hypothesized negative effects of bilateral exchange relations on collective good production. The experiments showed that such negative effects mainly occur when subjects have relatively low interest in high team performance. Future experiments may test whether this effect interacts with the extent to which communication between players is virtual. Negative effects of virtual communication may mainly show up in teams where incentives to cooperate are high, but virtual communication may help to reduce free rider effects when incentives are low. In a similar way, the conjecture derived from our simulations may be tested that team virtuality can have positive effects on cooperation when affective dependence of team members is high. To do this, experiments could directly manipulate affective dependence through variation in prior private exposure of subjects to each other. Another possibility is to use established measurement instruments from personality psychology to measure prior to an exchange experiment personality traits of subjects that affect their need for peer approval. Our study employs a number of simplifying assumptions that prevent us from modeling the full richness of possible mechanisms through which virtual communication may shape teamwork. We assume in particular homogeneity of team members, while the literature points out that diversity in skills and backgrounds is particularly high in geographically dispersed virtual teamwork (Griffith/Neale 2001). Moreover, a virtual team often consists of employees in different organizational locations or departments, with different skills and backgrounds, who

How May Virtual Communication Shape Cooperation in a Work Team?

277

work together for a specific task (see e.g. Jarvenpaa/Leidner 1999). In its present form, our model does not represent this situation well. Instead, it rather matches the situation of one local team with a number of isolated external members who can only communicate via electronic means. However, our approach can be readily extended to include heterogeneity of team members, for example in terms of their dependence on peer approval and their capability to contribute to the team task. This can be done in particular by modeling a production function in which each team member needs to make his specific contribution to the common effort and each team member has different opportunity costs of effort that reflect specific demands from the team members' own 'home' department. Another weakness of our study is that we have only insufficiently explored the parameter space of the model and the possible interactions effects between the three mechanisms we addressed. In particular, in order to complete our simulation study, it would be necessary to test whether the positive effects of team virtuality that we found for affective dependence would remain when the team also faces task uncertainty. We expect that task uncertainty combined with affective dependence may generate a tipping point in the effects of team virtuality on cooperation. It seems plausible that this tipping point arises from a combination of the negative effects of team virtuality under task uncertainty, shown in simulation experiment 1, and the positive effects shown in experiment 3 in the conditions with high affective dependence. Such a non-linear interaction effect would clearly be a new and innovative hypothesis in the field. Future research may show whether the model proposed in this paper is consistent with this intuition. In any case, we believe to have demonstrated that a formalized model of semi-virtual teamwork can both replicate findings from previous empirical work and generate new testable hypotheses that generalize beyond a small number of case studies.

Bibliography Alchian, A./H. Demsetz (1972), Production, Information Costs and Economic Organization, in: American Economic Review 62, 777-795 Axelrod, R. (1984), The Evolution of Cooperation. New York — (1997), The Complexity of Cooperation, Princeton Bendor, J./D. Mookherjee (1987), Institutional Structure and the Logic of Ongoing Collective Action, in: American Political Science Review 81, 129-154 Bush, R. R./F. Mosteller (1955), Stochastic Models for Learning, New York Caproni, P. J. (2001), The Practical Coach: Management and Skills for Everyday Life, Upper Saddle River Edwards P. K./H. Scullion (1982), The Social Organization of Industrial Conflict, Oxford Flache, A. (1996), The Double Edge of Networks, Amsterdam — / M . W. Macy (1996), The Weakness of Strong Ties: Collective Action Failure in a Highly Cohesive Group, in: Journal of Mathematical Sociology 21, 3-28 — (2002), Stochastic Collusion and the Power Law of Learning: A General Reinforcement Learning Model of Cooperation, in: Journal of Conflict Resolution 46, 629-653

278

Andreas

Flache

Friedman, J. W. (1971), A Non-Co-operative Equilibrium for Supergames, in: Review of Economic Studies 38, 1-12 — (1986), Game Theory with Applications to Economics, New York Griffith, T. L./M. A. Neale (2001), Information Processing in Traditional, Hybrid and Virtual Teams: from Nascent Knowledge to Transactive Memory, in: Research in Organizational Behavior 23, 379-421 Handy, C. (1995), Trust and the Virtual Organization, in: Harvard Business Review 73, 40-50 Homans, G. C. (1951), The Human Group, New York — (1974), Social Behavior. Its Elementary Forms, New York Jarvenpaa, S. L./D. E. Leidner (1999), Communication and Trust in Global Virtual Teams, in: Organization Science 10, 791-815 Kirkman, B. L./B. Rosen/C. B. Gibson/P. E. Tesluk/S. O. McPherson (2002), Five Challenges to Virtual Team Success: Lessons from Sabre, Inc., in: Academy of Management Executive 16, 67-79 Kitts, J./M. W. Macy/A. Flache (1999), Structural Learning: Attraction and Conformity in Task-Oriented Groups, in: Computational and Mathematical Organisation Theory 5, 129-146 Kollock, P. (1993), An Eye for an Eye Leaves Everyone Blind: Cooperation and Accounting Systems, in: American Sociological Review 58, 768-786 Lipnack, J . / J . Stamps (1997), Virtual Teams: Reaching Across Space, Time and Organizations with Technology, New York Macy, M. W. (1989), Walking out of Social Traps. A Stochastic Learning Model for the Prisoner's Dilemma, in: Rationality and Society 2, 197-219 — (1991), Learning to Cooperate: Stochastic and Tacit Collusion in Social Exchange, in: American Journal of Sociology 97, 808-843 — /A. Flache (2002), Learning Dynamics in Social Dilemmas, in: Proceedings of the National Academy of Sciences USA 99, 7229-7236 Olson, M. (1965), The Logic of Collective Action, Cambridge O'Mahony, S./S. R. Barley (1999), Do Digital Telecommunications Affect Work Organization?, in: Research in Organizational Behavior 21, 125-161 Petersen, T. (1992), Individual, Collective and Systems Rationality in Work Groups: Dilemmas and Nonmarket Solutions, in: Rationality and Society 3, 332-355 Raub, W. (1988), Problematic Social Situations and the Large Number Dilemma, in: Journal of Mathematical Sociology 13, 311-357 Seashore, S. E. (1954), Group Cohesiveness in the Industrial Work Group, Ann Arbor: University of Michigan Institute for Social Research Simon, H. A. (1982) (first published 1955), Models of Bounded Rationality, in: Behavioral Economics and Business Organisation 2, 239-258 Taylor, M. (1987), The Possibility of Cooperation, Cambridge Wong, S. S./R. M. Burton (2001), Virtual Teams: What are their Characteristics and Impact on Team Performance?, in: Computational and Mathematical Organization Theory 6, 339-360 Wu, J./R. Axelrod (1995), How to Cope with Noise in the Iterated Prisoner's Dilemma, in: Journal of Conflict Resolution 39, 183-189 Yamagishi, T./K. S. Cook (1993), Generalized Exchange and Social Dilemmas, in: Social Psychology Quarterly 56, 235-248

Analyse & Kritik 26/2004 (© Lucius & Lucius, Stuttgart) p. 279-301

Margit Osterloh/Sandra

Rota

Trust and Community in Open Source Software Production*

Abstract: Open source software production is a successful new innovation model which disproves that only private ownership of intellectual property rights fosters innovations. It is analyzed here under which conditions the open source model may be successful in general. We show that a complex interplay of situational, motivational, and institutional factors have to be taken into account to understand how to manage the 'tragedy of the commons' as well as the 'tragedy of the anticommons'. It is argued that the success of this new innovation model is greatly facilitated by a well balanced portfolio of intrinsic and extrinsic motivation, low costs for contributors and governance mechanisms that do not crowd out intrinsic motivation.

1. Introduction Until recently it was undisputed t h a t economic development is the more successful • the more extensive the privatisation of production; • the more strongly the state is willing to protect private property rights (e.g. North 1981). A powerful justification for the privatisation of common goods and strong private property rights was given by Garret Hardin's (1968) metaphor of the 'tragedy of the commons'. This metaphor highlights the problems of overuse and undersupply of common resources. Under certain conditions it might be necessary to change the metaphor to 'tragedy of the anti-commons' (Michelman 1982; Heller 1998), in which scattered owners have the right to exclude others from a scarce resource while no one has an efficient right to use it. If the transaction costs to bundle the property rights are too high, such a resource is prone to underuse. This problem might even be worse in the case of intellectual property rights, e.g. in biomedical research (Heller/Eisenberg 1998) and the software industry. This was empirically tested in a natural experiment (Bessen/Maskin 2000). Before 1980, patent protection for software was very limited in the United States, as it still is in the EU today. A series of court * The authors thank Bruce Ackerman, Charles Baden Puller, Yochai Benkler, Geoffrey Brennann, Bruno S. Ftey, Stefan Haefliger, Cynthia Home, Janos Kornai, Bernhard Kuster, Margaret Levi, Roger Luethi, Michael Baurmann, Susan Rose-Ackerman, Bo Rothstein, Sebastian Spaeth and Georg von Krogh for valuable contributions.

280

Margit Osterloh/Sandra

Rota

decisions in the early 1980's extended patent protection considerably. Consequently the number of issued patents increased. However, in contrast to what was expected, R&D expenditures relative to sales in relevant samples of the software and software related industries dropped significantly. The question arises under which conditions private ownership of intellectual property rights in effect hinders innovation and when state regulators should be careful in proliferating patent rights. Employing the example of one of the most innovative industries, open source software production, we study these conditions. Open source software production is an innovation model which is characterized by 1. (partly) public ownership of intellectual property, and 2. user driven distributed innovation. Some projects like e.g. Linux and Apache managed to attract huge communities of contributors in which intrinsically and extrinsically motivated members voluntarily work together in a complementary manner. 1 We will argue that since open source projects have no clear group and resource boundaries, nor does there exist a central formal authority, the development and maintenance of such communities depends on their ability to a) develop and enforce rules of cooperation in a self-organized manner, and to b) develop self-enforcing swift trust which is based on generalized reciprocity between group members. Open source software is the best known but by far not the only example of this innovation model. Other examples are the NASA Clickworkers (a project where volunteers mark and classify craters on maps of Mars), Slashdot (a site with 'News for Nerds' where users can post submissions, comment on their content and classify the comments themselves as to their helpfulness) and Project Gutenberg (peer-based distribution of books that includes volunteer scanning of hard copies and proofreading) (Benkler 2002). Studying open source software helps us to understand why and when private ownership of intellectual property rights should be proliferated carefully. Even though the creation of rules in open source communities is largely self-organized, state regulators are heavily involved. Thus the endeavour of this paper is threefold. Firstly, it studies how trust is developed and sustained in such virtual communities of innovation. Secondly, we analyze under what conditions intellectual communities are able to develop and put through their own governance rules. Thirdly, it identifies conditions under which private intellectual property rights can hinder innovations. We show that state-imposed private property rights can dramatically impede this process. State intervention thus may solve one tragedy but cause another. 1 Unfortunately not all projects are thus successful. As an empirical study by Krishnamurthy 2002 shows, many projects indeed dismally fail in this endeavour.

Trust and Community in Open Source Software Production

281

In the second section of this paper, a short overview of the characteristics of open source software is provided. The third section distinguishes various types of actors in open source software production according to their motivation to contribute to this kind of software. The fourth section discusses the role of trust in overcoming a first and second order social dilemma arising in situations of public good production. We argue that only projects that can be trusted to be able to solve both the first and second order social dilemma are attractive for potential new members and can thus hope to attract a large community. We show that on both levels of the social dilemma, trust based on encapsulated interests (Hardin 2002) is not sufficient, but that the emergence of swift trust based on the existence of a sufficient number of intrinsically motivated contributors is an important condition for the success of such projects. In section five, we analyse under what conditions this is possible. We show that low cost situations and appropriate institutional arrangements that do not destroy intrinsic motivation are essential for building trust that makes "virtual communities of innovation" work without central authorities and privatisation of intellectual property rights, even when no clear group and resource boundaries exist. We then go on to show how state regulation might adversely affect the open source innovation model by turning low cost into high cost situations. It is concluded that considering the complex interplay of motivational, situational, institutional and regulatory factors, trust and motivation issues should be given more weight in designing property rights. Managers and policy makers should be aware that the best policy not always is to blindly apply orthodox economics. Rather, they should consider the variety and interplay of existing extrinsic and intrinsic motivations and establish conditions under which self-governed 'communities of innovation' based on trust can emerge and be sustained.

2. What is Open Source Software? Open source is a collective term for software licences that give the user the right to read the source code of the software. Users are also allowed to change the source code and to publish these amendments with the original or the changed source code. Furthermore, one is not allowed to raise any licence fees or other fees for the source code. The open source software code thus constitutes a public good in the classical sense. Linux, Apache and Sendmail are three of the most famous examples of this very successful innovation model. Linux as a server operating environment already holds 13.7% of the $50.9 billion market for server computers. This share is expected to rise even further during the next years (Businessweek 2003). In January 2003 the open source web server Apache was used by over 65% of active servers across all domains. It received many industry awards for excellence.2 Sendmail routes at least 42% of mails in the Internet. In comparison, Microsoft as the closest competitor only holds a market

2

http : / /www. netcraft. co. uk/S urvey /

Margit Osterloh/Sandra

282

Rota

share of 18%.3 SourceForge.net, a repository of open source projects, lists more than 50.000 projects and more than 550.000 registered users.4 To explain the success of the new innovation model, one must take into consideration three interlinked characteristics that ensure an efficient concurrence of design and testing in open source software production. These characteristics, which make open source programs more innovative and robust than proprietary programs (Kogut/Metiu 2001), will be discussed in turn: a) Open source software is produced under licences that assure (partly) public ownership by allowing: • to read and have access to the software's source code as a necessary first step before one can change it, • to make copies and to distribute those copies, • to modify the program and distribute modified versions. In the special case of the GNU General Public License, the modified versions have to be published under the same terms as the original software (Stallman 1999). The various open source licences differ to the extent to which they allow public property to be mixed with private property rights. One of the most far reaching is the GNU General Public Licence (GPL). It forces every program that contains a free software component to be released in its entirety as free software. In contrast to the conventional copyright, this licence is called "copyleft". It 'infects' the open source software with a 'virus' to enforce compliance to the copyleft. Thus, it is ensured that any derived software will remain a public good. Other licenses, like Apache's, allow programmers to make their modifications private and distribute them as proprietary products. This blending of open and proprietary source, however, is sometimes condemned as a threat to the ideals of the open source community (Stallman 1999). b) User driven distributed knowledge production in rapid feedback cycles implements concurrence in design and testing of software modules and thus enables a very efficient new product development process. In traditional software production, the software is usually sold in a form giving no access to the source code. Customers therefore have only limited possibilities to detect mistakes ('debugging') and to improve the program. They can only give feedback to the seller about any malfunctions. In contrast, in open source software production, program innovations are disclosed to the users. A large audience tests the program, debugs it during use and gives immediate feedback. This is the reason why open source software is considered to have a lower defect density than proprietary software: "given enough eyeballs, all bugs are shallow" (Raymond 2001). The user driven rapid feedback cycles work not only with debugging but also with the production of whole modules. These are contributions to the source code which are published and reviewed by peers before they become part of the next release of the software. 3 4

http://www.dwheeler.com/oss-fs.why.html http://sourceforge.net

Trust and Community in Open Source Software

Production

283

c) Successful open source projects form voluntary 'virtual communities of innovation'.5. Following Tonnies (1920) communities are defined as groups of people whose actions are geared towards a collective goal. This differentiates communities from societies, which are abstract entities and in which interactions are characterized by the rational pursuit of individualized goals. In contrast, membership in a community is based on a feeling of belonging and shared values. This firstly ensures intensive network embeddedness based on a strong common culture. Secondly the members of these communities have a common know how on an expert level without having face-to-face interaction. In contrast to proprietary software, the users of open source software are often more sophisticated. Thirdly, Benkler (2002) argues that in virtual communities of innovation, transaction costs of matching talents to tasks can be reduced dramatically compared to market or hierarchical modes of organization. Individuals can judge for themselves in which tasks their talents might be put to most efficient use. This voluntary matching not only avoids information losses which are characteristic for market transactions and within firms, due to incomplete contracts. It also strengthens intrinsic motivation as a result of the autonomy of individuals. 6

3. Multiple Types of Contributors to Open Source Software Why should thousands of programmers contribute freely to the provision of a public good? Two alternative explanations are discussed. • Is it the result of the collaboration of self-interested individuals who invest in their reputation (e.g. Lerner/Tirole 2002a) or who calculate their investment in the open source community lower than their personal benefits (e.g. von Hippel 2001; von Hippel/von Krogh forthcoming)? • Is it fun, altruism or prosocial preferences which guide the contributors, as some of the leaders of the open source community claim (e.g. Kollock 1999; Raymond 2001; Stallman 1999; Torvalds 1998)? The two alternative explanations refer to a distinction between two kinds of motivation (Deci/Ryan 2000; Prey 1997; Osterloh/Frey 2000): Extrinsic motivation works through indirect satisfaction of needs, most importantly through monetary compensation. Intrinsic motivation works through immediate need satisfaction. An activity is valued for its own sake and appears to be selfsustained. The ideal incentive system for intrinsic motivation consists in the work contents itself. Intrinsic motivation has two dimensions. Following Lindenberg (2001), one can differentiate between enjoyment-based and obligation-based intrinsic motivation. 5 Communities of innovation are comparable to communities of practice. These are groups of people informally bound together by shared expertise and interest (Brown/Duguid 1991; 1998). Virtual communities of practice are described by Faraj/Wasko 2001 and Tuomi 2000. 6 See section 3 on intrinsic and extrinsic motivation.

284

Margit Osterloh/Sandra

Rota

• Enjoyment-based intrinsic motivation is the incentive focused on by Deci and his group (Deci et al. 1999). It refers to a satisfying flow of activity (e.g. Csikszentmihalyi 1975) such as playing a game or reading a novel for pleasure. • Obligation-based intrinsic motivation was introduced by Prey (1997) as a further important form of incentives. Empirical field evidence for the relevance of obligation-based rules are tax morale and environmental ethics, or organizational citizenship behavior (e.g. Organ 1995). We will argue that in the open source community there exist a variety of ideal types of contributors with different extrinsic and intrinsic motives.7 In this section, we will distinguish five different types of contributors according to their motives. In reality, these types are overlapping (Hars/Ou 2002; Lakhani et al. 2002). 3.1 Commercial Service Providers Commercial service providers make money with open source software in spite of the fact that open source software is a public good. The most prominent example is Red Hat. This company does not actually sell the source code (which anybody can download from the Internet for free). Instead, it sells support and services. In addition, it adds value by integrating autonomous open source components into a working and reliable operating system that can easily be installed by inexperienced users. Other commercial firms like Hewlett Packard sell hardware, like printers, and contribute add-ons, like printer drivers, to make their products work with open source software. Finally, companies like IBM contribute to open source software by making their hardware compatible with it. These firms are absolutely vital for the widespread adoption of this kind of software, because the inexperienced consumer gets reliable services and add-ons, thus helping to drive these programs into the mainstream (Kogut/Metiu 2000). 3.2 Software Customizers Contributors to open source can gain non-monetary benefits by tailoring the software to their own needs (von Hippel 1988; von Hippel/von Krogh forthcoming). They follow the saying "if you want something done right, do it yourself" (Lakhani/von Hippel forthcoming). They have sufficient incentives to contribute to an innovation when they expect the personal benefits to exceed their costs. Why should the benefits of publishing one's improvements on the Internet exceed the costs of revealing information? It is argued that, firstly, publication opens up the possibility that other users might work with the amendments of the code, maintain and develop them. That includes the elimination of possible errors (e.g. von Hippel 2001; Lerner/Tirole 2002a). Secondly, the Internet makes it possible for a software developer to access a wide audience with very low costs. 7 An ideal type, according to Weber 1949, is a construct which has been arrived at by the thinking accentuation of certain elements of reality. In its conceptual purity, this ideal type cannot be found empirically anywhere in reality.

Trust and Community in Open Source Software

Production

285

Because publication costs are small, publication on the Internet can pay off even if the expectations for helpful comments from other users are relatively low. Besides, the gains the developer reaps from the newly developed functionalities are not diminished by additional users. 3.3 R e p u t a t i o n Investors Contributors can make money indirectly by signalling their ability in the open source community which can then be turned into money through employment by a commercial software company or through easier access to venture capital. Employers or venture capitalists can take the reputation of a programmer as a signal for his/her abilities which would otherwise be hard to identify. It is argued that in open source projects reputation can be more easily made visible than in proprietary projects due to the system of files that list people who made contributions, and due to the public nature of mailing list archives (Lerner/Tirole 2002a; Moon/Sproull 2000). This system makes open source production comparable to the production of research in an academic community where reputation is made visible through citations. As in the academic community, strong norms exist regarding the public validation of innovative results. 3.4. H o m o Ludens While commercial service providers, software customizers and reputation investors are extrinsically motivated, much evidence exists that for many programmers the work itself is intrinsically rewarding. This idea corresponds to Huizinga's (1986) homo ludens, the playful human that receives some form of benefit simply from carrying out the programming or from dealing with a software problem. In that case, contributions to the open source code are not a cost but a benefit, not investment but consumption. Important contributors to open source software report that they are doing the programming "just for fun" and the public display of one's abilities (Torvalds/Diamond 2001). 8 As Raymond (2001) puts it: "We're proving not only that we can do better software, but that joy is an asset" (see also Brooks 1995). Writing or debugging software is perceived as a "flow experience" (Csikszentmihalyi 1975). More than 70% of open source developers report that they lose track of time while programming (Lakhani et al. 2002). As Ullman (1997) shows, programmers often experience a strong personal satisfaction from creating 'something that works'. This kind of motivation is fostered by voluntary work without time pressure (Deci et al. 1999). Creativity and motivation to volunteer in unpaid helping activities are higher when the external pressure is low (Stukas et al. 1999; Amabile et al. 2002). Not being subjected to delivery deadlines is an important characteristic of open source projects (Raymond 2001).

8 Displaying ones abilities is intrinsically motivated in so far as it is not aimed at building monetarizable reputation but rather at forming a consistent self-image or, as Akerlof and Kranton 2000 call it, a feeling of identity.

286

Margit Osterloh/Sandra

Rota

3.5 M e m b e r s of t h e Tribe The open source community is often described as a gift-culture instead of an exchange culture (e.g. Raymond 2001). A gift is characterized by receiving no tangible rewards but psychological benefits such as the 'warm glow' of sympathy or the satisfaction of living up to a moral commitment (Rose-Ackerman 1998). Gift-giving reveals the motivations of altruism or generalized reciprocity. Open source contributors report that they like the sense of 'helping others' or 'giving something back' to like-minded others (Faraj/Wasko 2001). Norms of generalized reciprocity sustain kindness as a social institution and lead people to provide help (Constant et al. 1996). These motivations are apart from transactional exchange relationships, because the receiver is often unknown to the giver. Participants report that "the person I help may never be in the position to help me, but someone else might be" (Rheingold 1993). People seem to reply to the entire group when answering an individual question (Wellman/Gulia 1999). The good of the community enters into the preferences of the individual contributor. Thus a reciprocal trust based on shared values and on warm feelings towards the group is existing (Rose-Ackerman 2001), rather than trust based on encapsulated interests (Hardin 2002). The belief that it is the right thing to give software away as a common good leads to the corollary that private ownership of intellectual property can be damaging. 9 The open source movement seems to be fuelled to some extent by the aim to destroy Microsoft's monopoly (e.g. Markus et al. 2000; Raymond 2001). Members of the tribe thus produce a public good of two different orders. Firstly, they contribute to the functionality and quality of the programs (first order public good). Secondly, they are engaged in monitoring and sanctioning activities to ensure that the source code stays open (second order public good 10 ). This includes a heated discussion between various fragments of the open source community on what kind of licence best supports these moral concerns. While some believe that only the GNU General Public Licence guarantees that source code remains open, others feel that the viral effect of this licence actually reduces freedom.

9 Empirical evidence about the importance of this motive within the open source community is ambivalent. According to Ghosh et al. 38% of open source developers report that their motivation to contribute to the community is their believing in that software should not be a proprietary good (Ghosh et al. 2002). In a different survey, Lakhani et al. 2002 find that 11% of open source developers are driven by the motivation to beat proprietary software. The difference might be explained by different samples and methodology. As always one has to be careful to take data based on self-reports at face-value due to possible social desirability biases. However, as heated discussions on the Internet about license terms and the examples discussed in chapter 5.1 show, a significant part of the community demonstrates their concerns about keeping the source code open very actively. 10 For first and second order public goods see section 4.

Trust and Community in Open Source Software Production

287

3.6 Complementarity of the Different Types We showed that in the open source community different types of motivation among the contributors exist. In the following we argue that these different types do not only coexist but are complementary to each other. Intrinsically motivated 'fun seekers' and 'members of the tribe' play an especially important role during the starting phase of open source projects. In this phase a usable product does not yet exist and the risk of project failure is very high. The enthusiasm of intrinsically motivated contributors helps these projects to gain enough momentum. In that way, they reduce the costs for extrinsically motivated contributors so that contribution becomes more attractive for them (Bessen 2002; Pranck/Jungwirth 2003). Without intrinsically motivated contributors • commercial service providers would lack the basis of their business. They make money on support only if the open source software is successful. • software customizers would have to make higher set up investments so that the costs are more likely to exceed the benefits. • reputation investors would not be able to produce marketable signals. Employers and venture capitalists are only attracted by successful projects that have already produced a critical mass of source code.11 On the other hand the success of open source software is dependent on extrinsically motivated contributors. Commercial players, software customizers or reputation investors trigger a leverage effect: • If the open source movement were solely based on intrinsic motivation, the products would not be linked the way they are to the needs of the users. A disadvantage over commercial development might result. • Inexperienced consumers could not use this kind of software which was originally designed by and for experts. So far, it is not known which proportion of intrinsically and extrinsically motivated people exist. But there are some preliminary empirical findings about what presently drives open source programmers and participants of newsgroups. In an empirical study with participants of a user-to-user Apache field support system Lakhani and von Hippel (forthcoming) report generalized reciprocity as the most agreed-with statement ("I have been helped before, so I reciprocate", "I help now so I will be helped in the future"), followed by identification with 11 The decision of reputation investors to join a project in its starting phase depends on their attitude towards risk taking. While risk adverse reputation investors will wait and see how a project develops before deciding to join, reputation investors with risk seeking preferences might be willing to join before it is clear whether the project will be a success or a failure. In both cases, however, the risk that the reputation investor has to take is lower the higher the number of intrinsically motivated contributors. We thank an anonymous referee for help in clarifying this point.

288

Margit Osterloh/Sandra

Rota

the community ("I answer to promote open source software"). These self-reports might emphasize 'socially correct' answers. Observable behaviour however supports these claims. The same empirical study found that in one of the Usenet newsgroups 57% asked questions only (and can thus be classified as free riders), 21% both asked questions and gave answers (they can be classified as reciprocators), and 22% provided answers only. Whether they do that to invest in their reputation or for altruistic reasons cannot be established by the data. Taking into account other empirical work one may conclude that extrinsic and intrinsic motivation exert about the same influence (Ghosh et al. 2002; Hars/Ou 2002).

4. The Role of Trust in Open Source Communities In open source extrinsically and intrinsically motivated contributors work together in a complementary manner to produce a public good. In the absence of a central authority that has the power to enforce contribution, the presence of extrinsically motivated utility-maximizers usually leads to an undersupply and overuse of public goods due to free-riding. This problem is known in the literature as the social dilemma: Social dilemmas arise if the actions of selfinterested individuals do not lead to socially desirable outcomes (Dawes 1980; Ostrom 1998). The consequence might be a 'tragedy of the commons' as Hardin (1968) described it, where the public good is not produced at all. But why do we not observe this problem in open source? In this chapter we turn to the role of trust and intrinsic motivation in overcoming the social dilemma in open source. In open source, the social dilemma is located on different levels. On the first level free riding can take place with respect to the production of software itself, because open source software constitutes a public good. Since nobody can be excluded from open source software, there is a problem of undersupply. 12 On the second level, the rules of the game that prevent first order free riding have to be observed and sanctioned. The worst kinds of first order free riding are not honouring the terms of the licence, using open source components in proprietary commercial products without giving anything back to the community or not citing or removing the credits of a contributor. As Raymond (2001) points out "surreptitiously filing someone's name off a project is, in cultural context, one of the ultimate crimes". Second order free riding is related to the monitoring of the compliance to the open source software norms and to the application of sanctions that prevent first order free riding. Reprimanding rule breakers in order to enforce the code of ethics is itself a public good and thus constitutes a social dilemma of a higher order: "Punishment almost invariably is costly to 12 In newsgroups this kind of free riding is known as lurking. This means reading ongoing discussions without contributing. Lurking is usually not really a problem as long as enough individuals are willing to contribute, because there is no rivalry in consumption. But there can exist a rivalry in attention, due to excessive crossposting and trolling. Given the huge amount of information that is transferred, it is critical that contributors respect the focus of the problem that is dealt with and therefore avoid crossposting (Kollock/Smith 1996). Trolling refers to deliberately posting messages with no other aim than to provoke other users.

Trust and Community in Open Source Software Production

289

the punisher, while the benefits from punishment are diffusely distributed over all members. It is, in fact, a public good" (Elster 1989, 41). As we will show trust is necessary on both levels to enhance cooperation. But what form does trust in open source communities take? It is important to note that trust in open source communities is rather institutional than personal. The number of participants to a given open source project is often very large and the communities are open to exits and new entries. Thus the development of trust cannot be based on repeated interactions of the same individuals who get to know each other over time and learn to trust each other. Rather, trust development in open source communities takes the form of what Meyerson et al. (1996) termed swift trust. 'Swift trust' describes a form of trust that is found in teams that only work together for a limited period of time and do not have the opportunity to develop trust based on personal relationships and mutual control. Members of such temporary teams decide on how much they think they can trust the others even before actually joining the team. This decision is based on stereotypical social categories and on a subjective appraisal of the intrinsically motivated adherence to mutual norms of reciprocity within a community. We differentiate between two different types of swift trust: trust based on encapsulated interests (Hardin 2002) and cognitive trust which is based on knowledge about the characteristics of the trustees, e. g. their dominating (intrinsic or extrinsic) motivation (Lewicki/Bunker 1995). Trust based on encapsulated interests means that I trust somebody because I believe it is in the trustee's (i.e. the trusted person's) best personal interest not to deceive my trust. This kind of trust is based on an estimation of the situation in which an interaction takes place. Given this specific situation, do I believe that the other person has enough incentives to behave trustworthy? Cognitive trust, on the other hand, is based on an estimation of the characteristics of the person I interact with. Do I believe that a person will honour my trust even if it would be in their best personal interest not to do so? In the open source context cognitive trust means trust that there are a sufficient number of intrinsically motivated contributors in a given project. The trustor himself might well be extrinsically motivated, i.e. he might contribute to the open source community in an instrumental way. The development of trust is especially important for potential new members, because they will only be willing to join a project which they believe is able to solve the first and second order social dilemmas in a sustainable way. We will now discuss the social dilemmas in open source in turn. As it turns out, the solution of the first order social dilemma is not a big problem in open source. As we showed in chapter 3, contribution to open source projects is not a pure public good. The different types of contributors all have individual incentives to participate which make the contribution option more worthwhile for them than merely free-riding. This holds for extrinsically as well as intrinsically motivated contributors. If enough people with sufficient individual incentives exist, the social dilemma is transformed into a coordination game where more than one equilibrium exists (Sen 1974). In a coordination game, the production of a public good mainly depends on the estimation of potential con-

290

Margit

Osterloh/Sandra

Rota

tributors, how many others will contribute as well. This is equivalent to saying that the contribution decision depends on the amount of trust based on encapsulated interests, i.e. the belief that it is in the personal interest of a sufficient number of other potential contributors to participate. But even on this level trust based on encapsulated interests alone is not enough. As we showed in chapter 3.6., especially in the beginning phase of a project, success is very much facilitated by a high number of intrinsically motivated participants. This means that also on the level of the first order social dilemma, cognitive trust in the intrinsically motivated trustworthiness of a relevant part of the community is a very important condition. Potential new members can estimate the trustworthiness of a community simply by observing its behaviour. Due to the publicity of the Internet they are able to judge whether norms of generalized reciprocity are lived up to within a community before deciding to join. For example they can observe whether members offer mutual support, provide helpful remarks to each other and answer questions in newsgroups. Thus the first order social dilemma in open source can be solved even in the presence of purely extrinsically motivated trustors. Unfortunately, this kind of cooperation is rather unstable. As soon as golden opportunities arise, selfinterest maximizing egoists cannot be counted on their cooperation anymore. This is the reason why the enforcement of cooperation rules is so important. They protect the community from exploitation by opportunists who face a golden opportunity. Second order social dilemmas of rule enforcement can be solved without a central authority if a sufficient number of obligation-based intrinsically motivated people exist who are prepared to punish rule-breakers even if such punishment is costly to them. Laboratory empirical evidence for the existence of such people can be found in one-shot public good games (Camerer/Fehr forthcoming; Ledyard 1995). In the open source community, these sanctions take place by violently blaming individuals on the Internet, called 'flaming'. Flaming is not simply a way of punishing rule-breakers, but also has an expressive function in assuring users that others are doing their part in using the public good wisely (Kollock/Smith 1996).13 Other sanctions are the public announcement of 'killfiling' (stating that one doesn't want to receive mails from a specific person) or shunning (deliberately refusing to respond). Monitoring the behavior of participants is often easy in the open source community because the Internet gives full transparency. 14 Sanctioning, however, is more of a challenge. Firstly, many sanctions (like flaming) are informal in nature. Secondly, the community members are often anonymous and no clear group and resource boundaries exist. Insofar the conditions in open source communities are different from the communities Ostrom (1990) has analysed. She argues that only if clearly defined group and resource boundaries exist, self governance of 13 This is especially important for potential new contributors who, before deciding to contribute, need to estimate whether they can trust in the stability of a community. 14 With the exception of illegally including open source code in proprietary programs.

Trust and Community in Open Source Software

Production

291

the commons can be successful. Nevertheless in the open source community self governance works. It is reported that sanctions have a significant effect on behaviour (Kollock/Smith 1996) though these sanctions often do no actual harm. They can influence behaviour indirectly however by damaging the sanctioned member's reputation or by inducing shame. In these cases, one has to assume that not only the sanctioner, but also the sanctioned person must be intrinsically committed to obligation-based rules. Purely extrinsically motivated egoists would not feel any shame (Elster 1999; Orr 2001). It can be concluded that the solution of the second order social dilemma in open source software production is greatly facilitated by the presence of intrinsically motivated trustees. Again, potential new members judge the trustworthiness of a community by observing its behaviour. If the existence of intrinsic motivation facilitates the development of swift trust and the complementary interaction of the different types of contributors to open source software, the question arises, under which conditions the required amount of intrinsic motivation and trust can exist and be maintained.

5. The Antecedents of Swift Trust in Open Source Projects We showed that different kinds of trust are needed on the trustor's and the trustee's side. On the trustor's side, extrinsically motivated trust based on encapsulated interests is sufficient. In contrast, on the trustee's side, there must exist a sufficient number of intrinsically motivated contributors as a precondition for the development of swift trust on the trustor's side. The problem therefore boils down to the question, under what conditions the trustor will judge the probability that a sufficient amount of intrinsic motivation exists and can be maintained as high. We identify two antecedents: Firstly, institutions must be created that ensure that the existing intrinsic motivation of the trustees is not destroyed. Secondly, even intrinsically motivated members will not contribute to a public good if the costs of doing so are high. Thus keeping the contribution costs low is the second antecedent of swift trust. We then go on to show that state intervention can disturb the equilibrium between intrinsically and extrinsically motivated contributors by turning low cost into high cost situations. 5.1 T h e Institutional Level: Initial Intrinsic Motivation M u s t N o t be Crowded Out It is hard to analyze the reasons why people develop an initial sense of fun for or a commitment to certain projects. But we know the institutional conditions under which initial intrinsic motivation is crowded out (undermined) or crowded in (strengthened) by external interventions (Frev/Osterloh 2002). Two conditions are relevant for the required institutional governance mechanisms to foster the new innovation model: self-determination (1) and conditional cooperation (2). 1) Self-determination: External interventions crowd out intrinsic motivation if the individuals affected perceive them to be controlling. In that case selfdetermination and self-esteem suffer and the individuals react by shifting their

292

Margit Osterloh/Sandra

Rota

'locus of causality' from inside to outside. In contrast, intrinsic motivation is crowded in if a person's feelings of self-determination are enhanced. (Deci/Ryan 2000; for a comprehensive overview over the empirical evidence see Frey/Jegen 2001). In open source projects self determination is enhanced for two reasons. Firstly, contributors choose for themselves where and what they wish to contribute (Benkler 2002; for empirical evidence see von Krogh et al. 2002). Secondly, a variety of self governance mechanisms give contributors large possibilities to participate in collective decision making in a transparent way. Extensive experimental and field research show that civic virtues are strengthened by procedural utility (Benz et al. forthcoming; Frey/Stutzer 2002; Osterloh et al. 2002) and that 'organizational citizenship behavior' is strengthened by participation and procedural fairness (e.g. Organ/Ryan 1995). Though governance rules in open source projects differ to a great extent (for an overview see Markus et al. 2000), open source contributors submit themselves voluntarily to these rules without any contract. 2) Conditional cooperation: Empirical evidence shows that many individuals contribute voluntarily to public goods in social dilemmas as long as some other individuals contribute also. They are conditional cooperators (Fischbacher et al. 2001; Levi 1988; Ostrom 2000). KDE (K Desktop Environment) gives an impressive example of how conditional cooperation can be undermined. KDE is a Windows-like desktop for Linux and other open source operating systems developed by the open source community. It is based on a graphical interface toolkit called Qt. Qt was developed by TYolltech, a commercial software firm. It did not comply with the requirements of open source (Stallman 1999). Even though the Linux community agreed that KDE was a technically excellent product, many members refused to endorse it because they didn't agree with the terms of the Qt licence. These members started a parallel project called GNOME that is distributed under copyleft. Finally Trolltech reluctantly relicensed their product. By now Qt is available under a copyleft license. Two consequences to maintain conditional cooperation follow. Firstly, intrinsic motivation is crowded out by free riders. Therefore institutional governance mechanisms must be set in place which hinder exploitation of voluntary donors. Open source licences, in particular copyleft, are such institutional mechanisms (Franck/Jungwirth 2003). They impose to contributors as well as to free riders what Hansmann (1980) calls the nondistribution constraint which is characteristic for non profit organizations (Rose-Ackerman 1996): Voluntary contributions cannot be redistributed among those who have a main impact on the organization. The nondistribution constraint is a major institutional precondition for voluntary donations to organizations. It is the reason why institutions like the Red Cross or most universities are governed as non profit organizations. Also for commercial providers who are dependent on the goodwill of the developers, it is crucial to commit credibly to the nondistribution constraint and not to appropriate the joint project in an unfair manner (Franck/Jungwirth 2003). Otherwise conditional cooperation breaks down and their business model will fail (Benkler 2002).

Trust and Community in Open Source Software Production

293

Red Hat has submitted itself voluntarily to constraints beyond the obligations of the open source license to strengthen conditional cooperation: As mentioned, Red Hat does not sell the Linux code. Instead, it sells support, services and value added by assembling and testing a running operating system that is compatible with other products carrying the same brand. After a short while, other CD-ROM distributors were advertising the same CD-ROM for a considerably lower price than Red Hat charged for its product. Even if Red Hat does not own intellectual property rights on the entire source code distributed on their CD-ROM, they do have the copyright on parts of the CD. How did Red Hat react? The somewhat astonishing answer is: not at all. The managers of Red Hat argued that the norms of the open source community precluded any claim on private property rights on their product. Since Red Hat is dependent on the goodwill of the open source community, it adheres to rules which foster conditional cooperation. Secondly, as experiments about conditional cooperation show, intrinsic motivation is crowded out if the existing rules of cooperation in a public good situation, e.g. the rules of nondistribution, are disregarded (Fischbacher et al. 2001). Therefore rule breaking must be hindered. Rule breaking can be made more difficult if costs for monitoring and sanctioning are low. As mentioned, monitoring in open source software production is easy due to the publicity of the Internet. Also, sanctioning by flaming or kill-filing is a low cost activity. But it seems reasonable to assume that in particular anonymous defectors are not vulnerable by sanctions. In these cases, sanctioning is only insofar effective as defectors in the open source community still feel a minimum of intrinsically motivated shame. Because informal graduated sanctions have a strong expressive function, they are very suitable not to crowd out shame by alienating people from the community (Kollock/Smith 1996; Orr 2001; Ostrom 1990). 5.2 Low Cost Situations: Economy of Virtue Why does production flourish in open source projects based on the intrinsic motivation of many people? Why do other fields like the pharmaceutical or bio-medical industry not apply this model? The simple answer is that this production model only works in situations in which the benefits exceed the costs. This is also true for intrinsic benefits. Even among intrinsically motivated donators, martyrs and saints are in short supply. Donators are more willing to contribute if the private opportunity costs are not too high (Rose-Ackerman 2001, 553), thus 'economizing on virtue' (Ackerman 1993, 198). According to North (1990, 43) there is a downward sloping demand curve for moral concerns. The more costly it gets, the less people contribute. On the other hand, if there exists a low cost situation, many people contribute small bits to the public good so that the total amount of contributions rises considerably (Kirchgassner 1992; Kliemt 1982). We distinguish two different aspects: The costs and benefits of actually producing the code and the costs and benefits of revealing it to the community. Even though these aspects are intermingled, we will now consider them in turn.

294

Margit Osterloh/Sandra

Rota

1) For the actual production of the source code, two conditions which are often found in system product industries are beneficial for the cost/benefit ratio, namely sequential and complementary innovation processes (Bessen/Maskin 2000; Somaya 2003). • Sequential means that each successive invention builds on the preceding one, in particular that there are incremental, not radical steps of innovation. This allows users and contributors to amortize their initial investments in project specific human capital over several rounds of innovation, thus keeping costs low. 15 • Innovation is complementary if several inventors, by following somewhat different research lines, enhance the overall probability that a certain problem is solved or, more generally, an innovation arises. This can be explained by internal dynamic economies of scales alongside trajectories, which help to concentrate successful search to a narrow field. Thus, if the efforts of an actor further the chances of discovery sufficiently, potential benefits are enhanced and this actor has an incentive to contribute to the process of discovery even if, in principle, she could wait until someone else makes the invention. The impact of such trajectories is of greater weight during the period of exploration than during exploitation (March 1991), because in that period uncertainty is more important. 2) Software is an area where the monetary costs of revealing innovations compared to the benefits are often quite low. There are two kinds of monetary costs to be considered. Firstly, costs of diffusion are low. Participants simply post their contributions on the appropriate Internet site. Secondly, the losses stemming from sharing intellectual property rights by using an open source license are often low compared to the gains from the expected feedback by other participants. In low cost situations, chances that many people are sufficiently intrinsically motivated to contribute their bit to the first and second order public goods are higher. Trustors have more reason to believe that the social dilemmas on both levels will be solved within a community. In the next section we show that state intervention may well turn low into high cost situations, thus undermining the necessary antecedents of both trust based on encapsulated interests and cognitive trust in the intrinsically motivated trustworthiness of a sufficient number of contributors. 5.3 T h e Impact of S t a t e Regulation: Turning Low Cost into H i g h C o s t Situations As we showed, open source software production challenges conventional economists' wisdom that innovations are better supported the more they are protected by private intellectual property rights (e.g. North 1981). We showed that in 15 Von Krogh et al. 2002 studied the process of joining an open source project empirically, using the example of Freenet. They show that this joining process takes quite some time.

Trust and Community in Open Source Software

Production

295

many open source projects the problem of underprovision seems to have been overcome by a complex interplay of extrinsically and intrinsically motivated contributors. 16 This interplay depends on the one hand on institutions that do not crowd out intrinsic motivation and on the other hand on low cost situations which keep the costs of intrinsically motivated moral behaviour within reasonable limits. However, state intervention could very well damage the future success of this new innovation model. Software production is an example of collective production in which private property rights might even cause a 'tragedy of the anti-commons' (Heller 1998; Heller/Eisenberg 1998). If property rights on a resource are scattered among many parties and the transaction costs to bundle the property rights are too high, this resource is prone to underuse. 17 In the case of intellectual property rights exclusion is made possible by patents. While patents are intended to further innovation by enabling innovators to collect the rents on their investment, in some cases they are used solely to block competitors from using an innovation. New software is usually based on existing programs and develops them further. Some authors even go as far as to state that software development today is impossible without infringement of intellectual property held by some other party (Bessen 2002). For open source software production, the potential effects of patent protection are even worse than in proprietary software production for several reasons. Firstly, since open source software developers cannot make money with their programs as such, they simply cannot afford to stand up in legal fights about patent infringements (Bessen 2002). 18 Secondly, especially in systems product industries like the computer and software industry, patents can be used to strengthen one's bargaining power in the case that some other party wants to block access to its own patents. Systems product industries are characterized by the fact that their products incorporate numerous inventions made by other parties. Since access to others' patents is essential, firms can build up patent portfolios to have something valuable to offer in exchange. Somaya (2003) tested this empirically and found evidence for this behaviour in the computer industry. In open source projects by definition it is impossible to build up such bargaining power. If a patent holder chooses to sue for infringement, open source projects have nothing to offer in exchange for an out of court settlement. The resulting 'patent thickets' threaten the ability of open source developers to improve software (Bessen 2002). Thirdly, the option to patent software rather than simply 16 In open source software production, the problem of overuse does not occur since there is no rivalry in consumption. Additional users can even generate positive external network effects. 17 An example of how privatisation in postsocialist economies can trigger a 'tragedy of the anti-commons' was given by Heller 1998. He started by asking why several years after the transition 'from Marx to markets' storefronts often remain empty while small kiosks full of goods mushroom on the streets. He argues that the problem is not primarily a lack of clearly defined property rights, corruption, or disobedience of the law, but the way government scatters property rights rather than creating coherent bundles of rights. 18 So far only a few open source developers have been sued for patent infringement (Bessen 2002). Still the potential threat must not be neglected.

296

Margit

Osterloh/Sandra

Rota

having the copyright on the source may simply increases opportunity costs, thus turning low cost into high cost situations. When faced with situations where such a tragedy of the anti-commons could arise, governments should not blindly apply orthodox market economics by increasing the scope and sophistication of regulations for private appropriation of intellectual property rights. Regulators should spend their efforts providing tools which help to avoid the 'tragedy of the anti-commons' rather than supplanting copyright protection, on which the viability of the new innovation model depends, with patent protection (Benkler 2002).

6. Concluding Remarks Open source software production is a highly successful innovation model. But it is, by far, not a singular case but rather one example of 'virtual communities of innovation'. The purpose of this paper is to inquire under which conditions the new innovation model might work in general. We argued that a stable solution of the social dilemmas of public good production is greatly facilitated by the existence of different kinds of motivations and trust. The first order public good consists of direct contributions to the development of open source software. The second order public good consists of monitoring and sanctioning deviations from the open source software norms. It is necessary to maintain conditional cooperation and generalized trust of the benevolent contributors. Extrinsically motivated participants only contribute to the first order public good in an instrumental way. Their aim is to tailor the products to their own needs, to invest in their reputation for monetary purposes or to enlarge the user base for their complementary products. They can only be trusted to contribute as long as it is in their best personal interest to do so. Intrinsically motivated members on the other hand contribute to the first and the second order public good. Intrinsic motivations are twofold, enjoyment-based (fun, public display of ones abilities) and obligation based (following norms of generalized reciprocity). Without these intrinsically motivated contributors cooperation in open source software is not sustainable since in many cases only their efforts safeguard the community against exploitation by purely self-interest maximizing opportunists in the presence of a golden opportunity. Thus, potential new members will only join a community if they can build cognitive trust that there exist a sufficient number of intrinsically motivated members who are willing to contribute to the public goods even if the costs should exceed their personal benefits. Thus the trustworthiness of a community to a large degree depends on the amount of intrinsic motivation of a relevant part of its members. We identified two conditions that make the presence of this kind of motivation more likely: Institutional arrangements that do not crowd out intrinsic motivation and low cost situations. On the institutional level, licences like copyleft seem to be a good solution to foster the conditional cooperation of intrinsically motivated contributors on the

Trust and Community in Open Source Software Production

297

one hand and serve the interests of extrinsically motivated investors on the other hand (Franck/Jungwirth 2003). Nevertheless, it seems to be important that the commercial providers commit themselves to the norms of the open source community beyond the obligations of the open source licence. As a consequence, companies like Red Hat submit themselves voluntarily to constraints beyond copyleft to maintain conditional cooperation in the community. Low cost situations foster trust by making intrinsically motivated contributions to the public goods more likely, thus strengthening the trustworthiness of the community. Thus, situational and motivational factors are highly interlinked. Low cost situations also explain why private ownership of intellectual property rights is sometimes inefficient. Private ownership of intellectual property among independent suppliers is less efficient in low-cost situations, (a) characterized by incremental and complementary innovations or (b) whenever concurrence of design and testing is crucial. In these cases, owners have a right to exclude others from a scarce resource while no one has an efficient right to use it. This condition not only holds in software production but also in other peer productions of intellectual goods (Benkler 2002). It does not hold in situations (e.g. in the pharmaceutical industry) where testing and market launching demand high investments. Further empirical research is needed to shed light on the quantitative dimension of this situational factor. Even though this new innovation model is based on self-regulation in the absence of a central authority, state intervention in the form of a strengthening of intellectual property rights could very well damage its success. Patent protection can not only reduce the room in which these self-governed institutions can flourish. They also turn low cost into high cost situations and therefore diminish individual willingness to contribute to the public good. To conclude, though we are far from fully understanding the interplay between motivational, situational, and institutional factors which make this new innovation model work, open source software production shows that even in virtual communities with no clear cut boundaries, under certain conditions trust and trustworthiness can flourish.

Bibliography Ackerman, B. (1993), We the People, Cambridge/MA Akerlof, G. A. (1982), Labor Contracts as Partial Gift Exchange, in: Quarterly Journal of Economics 97, 543-569 Akerlof, G. A./R. E. Kranton (2000), Economics and Identity, in: Quarterly Journal of Economics 105, 715-753 Amabile, T. M/C. N. Hadley/S. J. Kramer (2002), Creativity under the Gun, in: Harvard Business Review 80, 52-61 Barbera, F. R. (1999), Berkman Center rebuffed, in: Harvard Law Record 109, 1 Benabou, R./J. Tirole (2002), Intrinsic and Extrinsic Motivation, Princeton University Working paper, Princeton Benkler, Y. (2002), Coase's Penguin, or, Linux and the Nature of the Firm, in: The Yale Law Journal 112, 369-446

298

Margit Osterloh/Sandra

Rota

Benz, M./B. S. Prey/A. Stutzer (forthcoming), Introducing Procedural Utility: Not only What, but also How Matters, in: Journal of Institutional and Theoretical Economics Berlecon Research (2002), Floss Final Report - Part 3: Basics of Open Source Software Markets and Business Models, http://www.berlecon.de/studien/floss/FLOSS_Grundlagen.pdf Bessen, J. (2002), What Good is Free Software?, in: R. W. Hahn (ed.), Government Policy Toward Open Source Software, Washington, 12-33 — /E. Maskin (2000), Sequential Innovation, Patents and Imitation, MIT Economics Department Working Paper, Cambridge/MA Brooks, F. P. (1995), The Mythical Man-Month: Essays on Software Engineering, Reading Brown, J. S./P. Duguid (1991), Organizational Learning and Communities-of-Practice: Toward a Unified View of Working, Learning, and Innovation, in: Organization Science 2, 40-57 — /P. Duguid (1998), Organizing Knowledge, in: California Management Review 40, 90-111 — / — (2001), Knowledge and Organization: A Social-Practice Perspective, in: Organization Science 12, 198-213 Businessweek (2003), The Linux Uprising, 3.3.2003 Camerer, C. F./E. Fehr (forthcoming), Measuring Social Norms and Preferences Using Experimental Games: A Guide for Social Scientists, in: J. Henrich/R. Boyd/S. Bowles/C. F. Camerer/E. Fehr/H. Gintis (eds.), Cooperation, Self Interest and Punishment: Experimental and Ethnographic Evidence from Small-Scale Societies, Oxford Constant, D./L. Sproull/S. Kiesler (1996), The Kindness of Strangers: The Usefulness of Electronic Weak Ties for Technical Advice, in: Organization Science 7, 119-135 Csikszentmihalyi, M. (1975), Beyond Boredom and Anxiety, San Francisco Dawes, R. M. (1980), Social Dilemmas, in: Annual Review of Psychology 31, 169-193 Deci, E. L./R. M. Ryan (2000), The 'What' and 'Why' of Goal Pursuits: Human Needs and the Self-Determination of Behavior, in: Psychological Inquiry 11, 227268 — / R . Koestner/R. M. Ryan (1999), Meta-Analytic Review of Experiments: Examining the Effects of Extrinsic Rewards on Intrinsic Motivation, in: Psychological Bulletin 125, 627-668 Dempsey, B. J./D. Weiss/P. Jones/J. Greenberg (2002), Who is an Open Source Software Developer? in: Communications of the ACM 45, 67-72 Deutsche Bank Research (2002), FYee Software, Big Business? Open Source Software Tightening their Grip on Industry and the Public Sector, in: E-conomics 32 Elster, J. (1989), The Cement of Society: A Study of Social Order, New York — (1999), Alchemies of the Mind: Rationality and the Emotions, New York Faraj, S./M. M. Wasko (2001), The Web of Knowledge: An Investigation of Knowledge Exchange in Networks of Practice, Florida State University Working Paper, Tallahassee Fischbacher, U./S. Gächter/E. Fehr (2001), Are People Conditionally Cooperative? Evidence from Public Good Experiments, in: Economic Letters 71, 397-404 Franck, E./C. Jungwirth (2003), Reconciling Investors and Donators - The Governance Structure of Open Source, in: Journal of Management and Governance 7, 401-421

Trust and Community

in Open Source Software

Production

299

Prey, B. S. (1997), Not Just for the Money: An Economic Theory of Personal Motivation, Cheltenham — /R. Jegen (2001), Motivation Crowding Theory: A Survey of Empirical Evidence, in: Journal of Economic Surveys 15, 5, 589-611 — /M. Osterloh (2002), Successful Management by Motivation. Balancing Intrinsic and Extrinsic Incentives, Berlin — /A. Stutzer (2002), Beyond Outcomes: Measuring Procedural Utility, Berkeley Olin Program in Law & Economics, Working Paper Series, Working Paper 63 Ghosh, R. A./R. Glott/B. Krieger/B. Robles (2002), Floss Final Report. Part 4: Survey of Developers, http://www.infonomics.nl/FLOSS/report/FLOSS.Final4.pdf Hansmann, H. B. (1980), The Role of Nonprofit Enterprise, in: Yale Law Journal 89, 835-901 Hardin, G. (1968), The Tragedy of the Commons, in: Science 162, 1243-1248 Hardin, R. (2002), Trust and Trustworthiness, New York Hars, A./S. Ou (2002), Working for Free? Motivations for Participating in Open Source Projects, in: International Journal of Electronic Commerce 6, 25-39 Heller, M. A. (1998), The Tragedy of the Anticommons: Property in the Transition from Marx to Markets, in: Harvard Law Review 111, 622-688 — / R . S. Eisenberg (1998), Can Patents Deter Innovation? The Anticommons in Biomedical Research, in: Science 280, 698-701 Huizinga, J. (1986), Homo Ludens, Beacon Press Kirchgässner, G. (1992), Toward a theory of low-cost-decision, in: European Journal of Political Economy 8, 2, 305-320 Kliemt, H. (1982), The Veil of Insignificance, in: European Journal of Political Economy 2, 333-344 Kogut, B./A. Metiu (2000), The Emergence of E-Innovation: Insights from Open Source Software Development, Reginald H. Jones Center Working Paper, Philadelphia — / — (2001), Open Source Software Development and Distributed Innovation, in: Oxford Review of Economic Policy 17, 248-264 Kollock, P. (1999), The Economics of Online Cooperation: Gifts and Public Goods in Cyberspace, in: M. A. Smith/P. Kollock (eds.), Communities in Cyberspace, London, 220-242 — /M. Smith (1996), Managing the Virtual Commons: Cooperation and Conflict in Computer Communities, in; S. Herring (ed.), Computer-Mediated Communication: Linguistic, Social, and Cross-Cultural Perspectives, Amsterdam, 109-128 Kreps, D. M. (1997), Intrinsic Motivation and Extrinsic Incentives, in: American Economic Review 87, 359-364 Krishnamurthy, S. (2002), Cave or Community? An Empirical Examination of 100 Mature Open Source Projects, http://opensource.mit.edu/papers/krishnamurthy.pdf Lakhani, K./E. von Hippel (forthcoming), How open source software works: 'Free' user-to-user assistance, in: Research Policy — /B. Wolf/J. Bates/C. DiBona (2002), The Boston Consulting Group Hacker Survey, http://www.osdn.com/bcg/bcghiickersurvey.pdf and http://www.osdn.com/bcg/bcghackersurvey-0.73.pdf Ledyard, J. (1995), Public Goods: A Survey of Experimental Research, in: A. Roth/J. Kagel (eds.), Handbook of Experimental Economics, Princeton, 111-194

300

Margit Osterloh/Sandra

Rota

Lerner, J . / J . Tirole (2002a), Some Simple Economics of Open Source, in: Journal of Industrial Economics 50, 197-234 — / — (2002b), The Scope of Open Source Licensing, Harvard Business School Working Paper, Boston Levi, M. (1988), Of Rule and Revenue, Berkeley — (1991), Are There Limits to Rationality?, in: Achives Européennes de Sociologie 32, 130-141 Lewicki R. J./Bunker, B. B. (1995), Trust in Relationships: A Model of Trust Development and Decline, in: B. B. Bunker/J. Z. Rubin (eds.), Conflict, Cooperation, and Justice, San Francisco, 133-174 Lindenberg, S. (2001), Intrinsic Motivation in a New Light, in: Kyklos 54, 317-343 March, J. G. (1991), Exploration and Exploitation in Organizational Learning, in: Organization Science 2, 71-87 Markus, M. L./B. Manville / C. E. Agres (2000), What Makes a Virtual Organization Work? in: Sloan Management Review 42, 13-26 Merton, R. K. (1993), On the Shoulders of Giants, Chicago Meyerson, D./K. E. Weick/R. M. Kramer (1996), Swift Trust and Temporary Groups, in: R. M. Kramer/T. R. Tyler (eds.), Trust in Organizations: Frontiers of Theory and Research, Thousand Oaks, 166-195 Michelman, F. E. (1982), Ethics, Economics and the Law of Property, in: Nomos 24, 9 Milgrom, P. R./J. Roberts (1992), Economics, Organization and Management, Englewood-Cliffs Moon, J. Y./L. Sproull (2000), Essence of Distributed Work: The Case of the Linux Kernel, in: Firstmonday 5, 11 North, D. C. (1981), Structure and Change in Economic History, New York — (1990), Institutions, Institutional Change, and Economic Performance, New York Organ, D. W./K. Ryan (1995), A Meta-Analytic Review of Attitudinal and Dispositional Predictors of Organizational Citizenship Behavior, in: Personnel Psychology 48, 776-801 Orr, S. W. (2001), The Economics of Shame in Work Groups: How Mutual Monitoring can Decrease Cooperation in Teams, in: Kyklos 54, 49-66 Osterloh, M./B. S. Frey (2000), Motivation, Knowledge Transfer, and Organizational Firms, in: Organization Science 11, 538-550 — / J . Frost/A. Weibel (2002), Solving Social Dilemmas: The Dynamics of Motivation in the Theory of the Firm, University of Zürich Working paper, Zürich Ostrom, E. (1990), Governing the Commons: The Evolution of Institutions for Collective Action, New York — (1998), A Behavioral Approach to the Rational-Choice Theory of Collective Action, in: American Political Science Review 92, 1-22 — (2000), Crowding out Citizenship, in: Scandinavian Political Studies 23, 3-16 Raymond, E. S. (2001), The Cathedral and the Bazaar, Sebastopol Rheingold, H. (1993), The Virtual Community: Homesteading on the Electronic Frontier, New York Rose-Ackerman, S. (1996), Altruism, Nonprofits, and Economic Theory, in: Journal of Economic Literature 34, 701-728 — (1998), Bribes and Gifts, in: A. Ben-Ner/L. Putterman (eds.), Economics, Values, and Organization, New York, 296-328 — (2001), Trust, Honesty and Corruption: Reflection on the State-Building Process, in: European Journal of Sociology 42, 27-71

Trust and Community

in Open Source Software

Production

301

Sen, A. K. (1974), Choice, Orderings and Morality, in S. Körner (ed.), Practical Reason: Papers and Discussions, Oxford, 54-67 Somaya, D. (2003), Strategic Determinants of Decisions not to Settle Patent Litigation, in: Strategic Management Journal 24, 17-38 Stallman, R. (1999), The GNU Operating System and the Free Software Movement, in: C. Dibona/S. Ockman/M. Stone (eds.), Open Sources: Voices from the Open Source Revolution, Sebastopol, 53-70 Stukas, A. A./M. Snyder/E. G. Clary (1999), The Effects of'Mandatory Volunteerism' on Intentions to Volunteer, in: Psychological Science 10, 59-64 Tönnies, F. (1920), Gemeinschaft und Gesellschaft - Grundbegriffe einer reinen Soziologie, Berlin Torvalds, L. (1998), FM Interview with Linus Torvalds: What Motivates Free Software Developers, in: Firstmonday 3, 3 — /D. Diamond, (2001), Just for Pun: The Story of an Accidental Revolutionary, New York Tuomi, I. (2000), Internet, Innovation, and Open Source: Actors in the Network, in: Firstmonday 6, 1 Ullman, E. (1997), Close to the Machine: Technophilia and its Discontents, New York Varian, H. R./C. Shapiro (1999), Information Rules - a Strategic Guide to the Network Economy, Boston Von Hippel, E. (1988), The Sources of Innovation, New York — (2001), Innovation by User Communities: Learning from Open Source Software, in: Sloan Management Review 42, 82-86 — /G. von Krogh (forthcoming), Open Source Software Development and the PrivateCollective Innovation Model: Issues for Organization Science, Organization Science Von Krogh, G./S. Spaeth/K. Lakhani (2003), Community, Joining, and Specialization in Open Source Software Innovation: a Case Study, in: Research Policy 32, 1217-1241 Weber, M. (1949), Objectivity in Social Science and Social Policy, in: H. A. Finch/E. A. Shils (eds.), The Methodology of the Social Science, New York, 49-112 Wellman, B./M. Gulia (1999), Virtual Communities as Communities, in: M. A. Smith/ P. Kollock (eds.), Communities in Cyberspace, New York, 167-194

Analyse & Kritik 26/2004 (© Lucius & Lucius, Stuttgart) p. 302-312

Eric

Hilgendorf

Crime, Law and the Internet*

Abstract: After some introductory remarks on the German legal system and German legal politics, the main forms of datanet crime on the Internet are sketched. After that, one of the most important Internet-cases of the last decade, the CompuServe case, is discussed in some detail. One of the main problems of datanet crime is its global reach. The world-spanning nature of the cyberspace significantly enlarges the ability of offenders to commit crimes that will affect people in a variety of other countries. On the other hand, the jurisdiction of national criminal law cannot be expanded at will by any single nation. A transnational criminal law for the Internet is possible but should be restricted to the defence of universally (or nearly universally) accepted interests and values. In effect, it seems that the problems of computer-related crime on the Internet cannot be solved by criminal law alone.

1. Introduction The problem of Internet criminality has recently been the focus of great public attention. Some years ago, a computer virus romantically entitled "I love you", was spread via the Microsoft Outlook programme and managed to paralyse millions of computers world-wide, causing an estimated damage of several billion Euro. Indeed, very unique sweet nothings! On this occasion, however, they did not come from Moscow but from a frustrated IT-student who sought to impress his professor. A few weeks earlier, major companies and authorities had been impaired or paralysed by immense amounts of electronic messages. Here too the computer crashes caused enormous damages. Since then, attacks by means of computer viruses and Internet-worms have become more or less a normal part of day to day life. Obviously, Internet-criminality has finally left the area of science fiction and scientific speculation and has materialised into a severe practical problem. This issue also touches upon aspects of criminal law, since it is under this title that questions regarding the conditions for the punishment of Internet criminals arise. Firstly, I shall make a brief introduction into the laws and legal politics of the Federal Republic of Germany. This basis is helpful for further regarding the indictability of Internet criminals. Particularly in Germany, the idea that new technologies, such as the Internet, should be globally regulated by law is quite common. Criminal law is seen as part of legal social control (Krey 2002, 3). As far as this matter is concerned, the German legal system is relatively advanced in comparison with most other *I thank Miss Nicol Gates for looking over my English

Crime, Law and the Internet

303

legal systems. The law makers endeavour to protect everybody against any risk and to provide them with a fair legal decision in any single case. The numerous European regulations which are in place seek to pad out the various domestic legal systems within European states. The German welfare state provides the German people with a high standard of security and social justice. Yet, when considered in comparison to its advantages, a large number of disadvantages of the welfare state become apparent: The surge of new rules over decades has created the problem, that even cases which appear to be simple, pose complex questions of law. In areas such as taxation, social security and telecommunications, even specialists get lost in the jungle of laws and rules. This over-regulation leads to confusion and a lack of precision. Critics argue that complete regulation of life leads to losses of freedom. This loss of freedom can be sensed increasingly. Hand in hand, constitutional problems arise in the sense that, since apparently nobody knows all of these regulations, nobody can follow them. In order to try and regulate this chaos, suggestions have been put forward that new laws be initially passed for a limited period only. Their efficiency should be reviewed following expiration of this period. Yet a lasting solution can only be found when society moves away from the idea of a law maker who is responsible for everything and allocates the individual more power in finding solutions to individual problems. As far as Internet law is concerned, this means in practical terms, that the law maker should not be called upon for every small problem. Instead we must attempt to solve problems by applying existing laws. In addition, there is a growing need for private efforts at crime prevention on the Internet. This means preventive efforts on the part of prospective victims, as well as the implementation of adequate technology on the part of Internet service providers and the private monitoring and surveillance of cyberspace. Computer-law differs from most other branches of law, as it does not fall exclusively under public law, civil law or criminal law but crosses all three branches. It is therefore an overlapping subject, and as lawyers normally only work in one of the three main branches of law, there are very few who deal with IT-law as a whole. The computerisation of our life has produced questions in all areas of law. There are numerous special regulations, ranging from computer-criminal law to the signature-statute. In the following I will limit myself to a small section: new forms of crime in the Internet. The fast world-wide expansion of the Internet, and consequently the bordercrossing of data-streams leave the domestic legal systems of individual nations facing severe problems. This is most apparent in criminal law, since, unlike other areas of law, its roots are to be found in the respective national culture. This is typified by the fact that an Internet publication can be received world-wide; it cannot be limited to the territory of any one state. It follows that domestic criminal law is often overcharged in trying to counter datanet crime. Presently, international criminal law exists only in the basic form of Public International Law. It is traditionally accepted that the power to penalise remains sensitive to

304

Eric

Hilgendorf

national sovereignty and individual states only very reluctantly surrender this power to a superior jurisdiction. For these reasons it is often said that the sanctioning of datanet crimes is not a national but a global problem, which can only be solved by international agreements. It would be arrogant for a single nation to call itself the 'master of nets'. One should however take this position in its context: The problem of mastering datanet crime constitutes a small section of a much broader range of problems created by global markets. The bearers of such problems are generally national legal systems and national politics. The argument often arises that global processes are simply out of control. Therefore individual states should limit themselves to securing economical frameworks and to correcting excessive severity in single cases. We are here talking about the importance of economics and the end of politics—and therefore about the end of law. To my mind, this pessimistic attitude is wrong and even dangerous: In a democratic state it is not the economic market but politics which is responsible for finding solutions to questions of social order. Criminal law provides a good means to secure highly regarded legal values. The spreading of datanets does not affect the distribution of legal duties. The opinion, which is often expressed that domestic criminal law cannot tackle new datanet crimes and that success is only promised by international means, would bring with it hand in hand a severe weakening of legal protection in favour of the interests of the market. For the same reason, the idealistic idea of an Internet without (criminal) law is naive. The discussion concerning penalties for datanet criminality encounters further hurdles in other areas, such as those concerning the highly specialised technology involved, combined with the modern, specified language, containing a lot of English terms. As technology is rapidly developing, the terminology is also constantly changing (advertising copywriters have a large influence upon this). Criminal law should not have to adapt itself to this ever-changing language. To render the solutions to important dogmatic questions dependent upon mere technical details, which will soon disappear, also fails to provide us with a convincing argument. In the near future, the usage of datanets will become so common that technical aspects will fade into the background, just as the case is today with telephone or broadcasting. The criminal countering of datanet criminality should not result in the creation of a new legal discipline with its own principles and rules, but means the consequent application of traditional yet well proven legal structures. It goes without saying that such structures should correspond with the constitution. In the following I shall analyse how criminal law reacts to the challenge of datanet crime. Firstly, I shall present some of the new forms of datanet crime (2). Secondly I will outline the controversial CompuServe-case (3). I shall then move on to look at International Criminal Law in combination with the German penal code and how this determines the severity of the sanction given (4). I aim to demonstrate that in applying such law, a satisfactory solution to the question of applicability can be found. To conclude (5) the possibility of a transnational or international criminal Internet law shall be discussed. In the following therefore, not only penal law but also penal politics shall be discussed.

Crime, Law and the Internet

305

2. Appearance Forms of Datanet Crime All criminal offences can basically be committed via Internet or other datanets, even bodily injury or murder. However, the expression 'datanet criminality' should be limited to those forms of crime which are facilitated by the Internet. The following groups are presently of particular significance (Hilgendorf 2001): • Monitoring of data traffic. • 'Hacking'. • Blocking and sabotage of e-mail-addresses, especially those of big companies and authorities. • Access to or sending of forbidden material (in data form), such as rightist propaganda or pornography. • Prohibited changing or displacing of data. • Prohibited copying of data (this includes the infringement of copyrights e.g. by file sharing). • The simulation of a false identity. This may happen where a sender uses a false password, or as an addressee of a message, e.g. by the simulation of a connection to a bank-computer. • The spreading of false information intentionally, e.g. in order to manipulate stock-market courses. These eight groups are not the only ones relevant to datanet criminality, yet should give an overview of its most important forms at present. In the future, the simulation of a false identity and the providing of access to or the sending of false information will become all the more important due to the increasing usage of e-commerce and electronic banking. In my opinion, the blocking and the sabotage of e-mail-addresses which has drawn immense public attention over the last few years will gradually loose its importance as countermeasures are developed. In the future, the main features of datanet criminality will not be hacking or computer sabotage, but will be fraud, respectively computer fraud. Faked news on the Internet is often identified very late. When Bill Clinton gave his first press conference on the Internet, he stated that he energetically advocated a greater amount of Internet pornography. At first, nobody seemed to regard this statement as astonishing (which might indicate an image problem of the former US President!). Only later did it become clear that indeed a skilled hacker had produced this statement. Alongside this there are further cases concerning the support of criminal acts committed by others, e.g.: • Provider liability and • Liability for placing links to pages with illegal contents.

306

Eric

Hilgendorf

The discussion about the countering of datanet criminality is still in its early stages, whereas the problem concerning the liability of providers has already attracted much attention. The verdict of the County Court of Munich against the manager of CompuServe Germany is a milestone in this discussion.

3. The CompuServe Case The CompuServe case is ranked amongst the most sensational German criminal cases of recent years. On the 28th of May 1998, the former manager of CompuServe Germany was sentenced to two years imprisonment for, amongst other things, spreading paedophilic pornography. The sentence was set on probation of three years. This led to global protests; 'incompetence' and 'presumption' were the mildest of the reproaches made against the presiding Judge. In the USA German beer was poured on the streets, and the question was asked, who had actually won the war. The County Court's judgement was annulled on the 17th of November 1999 by the District Court of Munich. What happened? The accused was the manager of CompuServe Germany— the latter belonging 100% to CompuServe USA. The company operated and administered connected computers, which gave German members access to the data banks of CompuServe USA. In November 1995 the managers of CompuServe Germany were told by the Bavarian criminal prosecution authorities that paedophilic pornography could be delivered from the US-American server to Germany. Special interest was drawn to five sites with such contents. The manager immediately forwarded this information to the mother-company in the USA, which then closed down the identified newsgroups. In the course of events, welldesigned net-control systems were procured for the customers of CompuServe. Hereafter the newsgroups were opened again. Nevertheless the Bavarian criminal prosecution authorities could download several news articles containing paedophilic, sodomitical and sadomasochistic pornography from CompuServe USA. This was possible because the filter-software which CompuServe had applied did not work perfectly. Paedophilic pornography is not permitted in Germany; those who own it or make it accessible to third parties must face severe penalties. The prosecution accused CompuServe Germany's manager of not having prevented access to the illegal contents for German users by not applying better filter software. However, after specialists from the federal bureau for security in information technology stated in court that a complete control of the border-crossing data traffic was not possible due to technical reasons, both the prosecution and the defence sought that the charges be dropped. It came as quite a surprise when the county court sentenced the accused manager to two years imprisonment. The sentence does, in opposition to the opinion of many computer magazines, have good argumentation. It is based on the idea that CompuServe Germany and CompuServe USA form an economic union and therefore must also be treated as such in terms of German criminal law. This means that omissions by the US mother company can be ascribed to the German daughter company. The Bavar-

Crime, Law and the Internet

307

ian judge regarded it as proven that the US Company could have permanently banned the contents of paedophilic pornography from its servers. By not doing this they committed an offence, for which they were liable. Indeed, an offence which can be ascribed to the manager of CompuServe Germany. The most important points of the judgement state that: "By closing down the newsgroups, which evidently direct to violent, paedophilic and sodomitical pornography, CompuServe USA could have stopped the usage of this material. If CompuServe USA had taken those newsgroups in question out of its data store, the customers of CompuServe Germany would not have had access to them. ... The fact that other news servers provide public access to hard pornography as well does not influence the penal judgement on CompuServe USA. ... CompuServe USA's interest in providing their customers in Germany with a forum for hard pornography is not worthy of being secured. On the contrary, there is a great social interest in not letting the technical development in the branch of telecommunication become an unlawful area, in which respected social values such as juvenile security and the prevention of sexually motivated violence are subordinate to commercial interests and thus sacrificed." (Neue Juristische Wochenschrift 1998, 2840; translated by the author) In my opinion this statement deserves much attention. In principle, it is convincing to rank the social values outlined above higher than commercial interests. Nevertheless many legal details remain unanswered, such as to which extent the manager is to be held responsible for the actions of others who neither closed down nor withdrew the illegal contents. The manager of CompuServe Germany had very little influence on the American mother company. Due to his subordinate position it was not appropriate to ascribe the mother company's omission to the daughter company's manager. The district court Munich I (Neue Juristische Wochenschrift 2000, 1051), which found the accused not guilty, explained that the actions of the German manager could only qualify as assistance. The accused had lacked the intention to spread hard pornography in the Internet, on the contrary—he had tried hard to make the US mother company close down and erase the objected contents. To render him guilty of providing access to hard pornography would therefore not be correct. The judgement of not guilty was generally met with approval among law specialists. It is quite evident however, that many important questions on the indictability of Internet criminality in general and the providers' liability remain unanswered. The CompuServe case does pose special problems since the criminal act was not committed in Germany but on foreign territory—the USA.

308

Eric

Hilgendorf

4. The Applicability of German Criminal Law to Illegal Publications in the Internet The common thread shared by all types of Internet criminality is their internationality. This leads to the question of whether those offences, which are committed abroad but whose effects are felt in Germany fall under German Criminal Law. On the Internet the crossing of national borders is nothing new. This applies to both information which is sent to Germany (e.g. via e-mail) and information, which is drawn from a foreign Internet page onto German territory. Problematic is to which extent German criminal law can be applied to foreign information, which is received in Germany. Such questions are regulated by 'International Criminal Law'. A cross border application of domestic criminal law is basically an attack upon the sovereignty of a foreign country, according to principles of Public International Law. The principle of non-interference, according to which the intervention in the affairs of a foreign nation is generally not allowed, is of most relevance. The current discussion concerning the applicability of German criminal law to the Internet has shown that there exist a large number of quite different possible solutions. There is, however, as of yet, no commonly accepted opinion and a uniform practice amongst prosecuting authorities has not yet come about. Nevertheless, the German Federal Prosecutor has made his intention to apply German criminal law to the Internet world-wide quite clear. The Federal Court of Germany accepted this view in the Toeben case (Körber 2003). This would mean that an American, who places rightist material in the Internet on USA territory, renders himself liable under German criminal law, although his action is not actually prohibited by US American criminal law. The starting-point for allocating such criminal liability is the principle of territorial jurisdiction: German criminal law is valid for crimes carried out within its territorial borders. The principle regarding the location of the crime is laid down in § 9 of the German Criminal Code. This section states that German criminal law is applicable when the perpetrator acted in Germany or when the effects of the act are felt upon German territory. 4.1 A n Unlimited Applicability of German Criminal Law? German criminal law seems to be applicable to Internet publications without question when illegal contents can be received in Germany. This would mean that German criminal law can be applied to any publication on the Internet, which falls within the scope of the German Criminal Code. However, viewed in the light of Public International Law, this position is rather problematic as it effects the sovereignty of individual nations and thereby breaks the rule of noninterference. Furthermore, from a political point of view, the global application of German criminal law does not seem to be justifiable. In order to make such an application justifiable, international agreement would be required.

Crime, Law and the Internet

309

A single sided global application of German criminal law in the Internet may well appear to foreigners as a new version of the old German slogan: "Am deutschen Wesen soll die Welt genesen" ("At the hand of my German brother, the world shall recover"). It must be taken into consideration that other nations, for example, Iran or China, could also try to globally apply their domestic criminal law to the Internet. Prom this point of view, it appears necessary to limit the applicability of German law (Hilgendorf 1997). 4.2 Limitation of the Rule of General Application? Some argue that German criminal law is only applicable when the perpetrator, by his act, actually sought effects upon German territory. Hence, this means that the operator of an Internet page would have to intentionally open the illegal data to Germany if he is to become subject to German criminal law. This limitation, however, does not seem quite practicable: The mere intention to produce effects within Germany is very difficult to be proven when the perpetrator acted outside German territory. Other opinions state that German criminal law should not be applied to acts which are not illegal in the country of their performance. If rightist propaganda is allowed in the USA, it should not be penalised when it reaches Germany via the Internet. This, however would lead to the creation of a legal oasis, where Internet-offences can be carried out without the threat of legal sanctions. A third opinion held therefore looks to the location of the server itself before deciding upon the applicability of German criminal law. The main problem with this theory is that many contents are not only kept on one, but on numerous servers. In all of the theories so far put forward, there is the necessity for a 'special point of contact' (or 'genuine link') to Germany in order to apply German law. This means that German criminal law should only be applicable to offences which are committed outside of Germany, if the effects of the offence are felt in Germany. In addition, a genuine link to Germany must be shown. Regarding datanet offences, such genuine links can be found in, for example, an arrest on German territory, German citizenship, a domicile in Germany or the simple intention to have effects in Germany. The criteria requiring a domestic point of contact is based upon well proven regulations in international civil and business law. It is also supported by the principles of Public International Law. In the Toeben case, the existence of a special point of contact or 'genuine link' to German affairs is a problem. Neither was the German language used, nor did Toeben aim at a German audience. It WEIS only because of the globality of the Internet that Toeben's homepage could be seen in Germany as well as in all other countries of the world. Therefore, a large part of Germany's criminal law experts are sceptical about the Federal Courts decision to apply German criminal law to the Toeben case.

310

Eric

Hilgendorf

5. On the Way to an 'Intercultural Criminal Law'? 5.1 D o e s an 'Intercultural Criminal Law' Exist? Due to the world-wide growth of the Internet it may be expected that in the future, courts will increasingly have to deal with perpetrators who come from a foreign country or even from another culture. The Internet law, much more than other areas in which law crosses borders, requires the internationalisation of criminal law (Hilgendorf 2002). This poses the question as to how far cultural diversity should be taken into consideration when determining criminal liability. To be subject to a criminal charge, the perpetrator must as a rule know that he has done wrong. Cases such as those concerning Americans providing Germany with access to rightist propaganda on a US server (§ 130 German Criminal Code), or of the Scandinavian who intentionally sends data with pornographic contents to Germany (§ 184,1, 4 German Criminal Code) are unproblematic because these perpetrators generally belong to the same culture. Problematic however, is the question of which rules are to be applied when an Internet page, operated in Saudi Arabia but in the German language, calls upon Moslems in Germany to contravene the regulations of the German authorities (§§ 113, 26). Similarly problematic is an Internet page which is operated in Africa requesting that Africans in Germany perform female genital mutilation, and giving detailed instructions thereof (§§ 223, 224, 26). In cases such as these, it can be argued that the offender was not able to realise that his doings were wrong. German criminal law is not completely unprepared for these challenges. Migration streams of recent decades have produced large multi-cultural groups within Germany. The requirement for the consideration of cultural diversity is therefore nothing new to German criminal law. As far as actual knowledge of the unlawfulness of the act is concerned, which is the most important factor in determining the perpetrator's guilt, cultural diversity can be easily taken into account. The perpetrator should not be punished if he did not and could not have had such knowledge. Yet, regarding foreigners living in Germany, one may assume that they are fully aware of the illegality of their actions, at least concerning the basics of criminal law. One practical example is incest, which is forbidden in Germany, however not in many Southern European countries; there incest generally seems to be socially accepted. For perpetrators acting via Internet however, it can be assumed that they do not know exactly what constitutes a crime in the state with which they are dealing. This is particularly so when the person in question has never left his or her home cultural area. Someone who, for example, lives in a country in which several marriages are common and advises his nephew in Germany to take a second woman beside his wife will not render himself liable under German law, although bigamy is a criminal offence under § 172 of the German Criminal Code. The case of female genital mutilation might be viewed differently. Such actions infringe the fundamental human right of personal integrity. In this case, knowledge of the illegality of the act can be assumed, even if the perpetrator is a

Crime, Law and the Internet

311

foreigner. The same goes for an attempt of the offence as well. The operator of such a homepage in an African country could therefore be liable under German criminal law, if such an offence is committed in Germany. Questions such as these have, however, not yet appeared before the German courts. Therefore, one can say that criminal law is indeed able to take cultural diversity into consideration. In this respect 'Intercultural Criminal Law' already exists.

5.2 A Transnational Criminal Law For the Internet? As regards those acts which exert a negative effect within a state's territory, but which escape national criminal liability owing to the absence of the earlier mentioned 'link', a different problem arises. If such acts are not criminal under the domestic law of the country where they were originally performed, it seems to be appropriate to aim towards an intercultural and transnational criminal law. This function cannot be assumed by the Internet's international code of behaviour, so-called 'Netiquette'. The present development of Public International Law and the emergence of the International Criminal Court Statute, show that the idea of an effective transnational Internet criminal law is not unthinkable. It is, of course, not to be expected that all states would accept such an International Criminal Law for the Internet, nor regard it necessary to bring such law into existence. Moreover, the contents of a criminal law for the Internet are still to be defined. Before a solution can be found, political considerations are necessary and the ideas of all involved must be examined. It would not be right to base such an Internet law solely on the cultural values held by Western Europe and North America. This is of particular importance when the basis is not clear, but hidden by such vague terms as "human rights" or "justice". This does not mean that the importance of human rights should be forgotten; it merely means that culturally varying points of view should be considered. The common reproach against 'western cultural imperialism' rings true with regard to the Internet. The following argument is possibly the one pointing us in the right direction. Generally speaking, criminal law has the task of securing those values which the respective society thinks are the most important. These values become legal. When talking of the construction of an International Criminal Law, only those values can be taken into account, which are of universal interest. In practical terms, these are values which are based directly upon human nature, such as the interest in life, the interest to be without injury, personal freedom and the security of personal honour. Accordingly, a criminal law for the Internet which is to be established worldwide would have to be limited to those cases, where liability is based solely upon such nearly universally accepted values. Yet these values, with the exemption of personal honour, are seldom affected by Internet publications. Rightist propaganda, for example, does not necessarily affect the personal honour of any man or woman. For this reason, most parts of Internet-criminality have to be left to

312

Eric Hilgendorf

domestic criminal law regulations. States must, however apply such domestic regulations carefully, with regard for the sovereignty of other nations. There is, as well as true International Criminal Law and purely national criminal law for the Internet, a third way to cope legally with the criminality on the Internet. So called 'universal jurisdiction' allows states to apply their own domestic criminal law worldwide, when certain criteria (which are defined by public international law) are met. Universal jurisdiction concerning Internet publications is possible only in limited cases. As far as German (or any other national) law is concerned, this would mean that only severe Internet crimes fall within the scope of national criminal law, irrespective of where in the world they were committed. This is already the practice as regards paedophilic and violent pornography (compare § 6 German criminal code). Any attempt, however, to enlarge the number of offences which fall under a single state's criminal law must be met with caution. If countries with deep rooted democratic traditions, such as the USA, regard the right to freedom of speech so highly, that they fail to punish rightist propaganda, that is their domestic business. It is not any single other nation's job (and certainly not Germany's) to try and counter this by expanding its domestic jurisdiction to cover the entire Internet. If national criminal law is to be expanded worldwide, this should be based on international agreements. In effect, this means that the problem of computer-related crime on the Internet cannot, on a world-wide scale, be solved by criminal law alone.

Bibliography Hilgendorf, E. (1997), Überlegungen zur strafrechtlichen Interpretation des Ubiquitätsprinzips im Zeitalter der Internet, in: Neue Juristische Wochenschrift 29, 1873-1878 — (2001), Die Neuen Medien und das Strafrecht, in: Zeitschrift für die gesamte Strafrechtswissenschaft 113, 650-680 — (2002), Nationales oder transnationales Strafrecht? Europäisches Strafrecht, Völkerstrafrecht und Weltrechtsgrundsatz im Zeitalter der Globalisierung, in: H. Dreier/H. Forkel/K. Laubenthal (Hg.), Raum und Recht. Festschrift 600 Jahre Würzburger Juristenfakultät, 333-356 Körber, F. (2003), Rechtsradikale Propaganda im Internet - der Fall Toben, in: E. Hilgendorf (Hg.), Das Strafrecht vor neuen Herausforderungen. Band 1, Berlin Krey, V. (2002), German Criminal Law. General Part. Textbook in German and English. Volume I: Basics, Stuttgart

Lyrics © 2003 by Geoffrey Brennan. Music © 1959 by Rodgers and Hammerstein

Internet-Hymn iLli

In-

ter-

~a Ev-

'ry

a Ev-

er

net,

In-

morn- ing

ter

net,

greet me.

you

you,

love

É you're

always

hap-

py

to

meet

me.

É

£

Thus, what's

the

fuss,

you

have

won

my

trust.

È You

your

and

E-trade

we

millions

must,

of

we can't

browsers.

go

bust,

È We'll never

lose

our

tr-

m ous-

ers.

Internet, you're my bet. I know that you'll never bug me, You'll replace face to face — good for me, 'cos I'm ugly. Ordinary life seems unreal to me — dull and grey and tiring, But on-line, everything's fine — Internet you're inspiring. Internet, Internet. I've got Internet fever! Let us get more virtual yet! Internet for e-ver.

Authors Hans W. Bierhoff

Gary E. Bolton

Geoffrey

Brennan

Universität Bochum Fakultät für Psychologie Universitätsstr. 150, 44780 Bochum [email protected] Pennsylvania State University Smeal College of Business Administration 310k BAB, University Park, PA 16802 [email protected] The Australian National University Research School of Social Sciences Canberra A C T 0200 [email protected]

Coye Cheshire

Stanford University Department of Sociology 450 Serra Mall, Stanford, CA 94305-2047 [email protected]

Karen S. Cook

Stanford University Department of Sociology 450 Serra Mall, Stanford, CA 94305-2047 [email protected]

Andreas Flache

University of Groningen Interuniversity Center for Social Science Theory and Methodology Grote Rozenstraat 31, 9712 T G Groningen [email protected]

Werner

Giith

Max Planck Institute for Research into Economic Systems Strategie Interaction Group Kahlaische Strae 10, 07745 J e n a gueth@mpiew-jena. mpg. de

Russell

Hardin

New York University Department for Politics 726 Broadway, New York, NY 10003-9580 [email protected]

Eric Hilgendorf

Universität Würzburg Juristische Fakultät Domerschulstr. 16, 97070 Wrzburg hilgendorf@j ura. uni-wuerzburg. de

Authors Elena Katok

Pennsylvania State University Smeal College of Business Administration 509H BAB, University Park, PA 16802 [email protected]

Hartmut

Universität Duisburg-Essen Fach Philosophie Lotharstr.65, 47048 Duisburg [email protected]

Kliemt

Bernd Lahno

Universität Duisburg-Essen Fach Philosophie Lotharstr.65, 47048 Duisburg b. lahno@t-online. de

Uwe Matzat

Eindhoven University of Technology Department of Technology Management Sociology Section P.O. Box 513, 5600 MB Eindhoven [email protected]

Victoria McGeer

Princeton University Department of Philosophy 1897 Hall, Princeton, NJ 08544-1006 [email protected]

Axel Ockenfels

Universität zu Köln Staatswissenschaftliches Seminar Albertus Magnus Platz, 50923 Köln [email protected]

Margit Osterloh

University of Zürich Institute for Organization and Administrative Science Plattenstrasse 14 8032 Zürich [email protected]

Philip Pettit

Princeton University Department of Politics Corwin Hall, Princeton, NJ 08544-1012 [email protected]

Anabel Quan-Haase

University of Toronto Department of Sociology 455 Spadina Avenue, Toronto M5S 2G8 aquan@chass. utoronto. ca

Sandra Rota

University of Zürich Institute for Organization and Administrative Science Plattenstrasse 14 8032 Zürich [email protected]

Authors

316 Chris Snijders

Eindhoven University of Technology Department of Technology Management Sociology Section P.O. Box 513, 5600 MB Eindhoven [email protected]

Henk de Vos

University of Groningen Department of Sociology Grote Rozenstraat 31, 9712 TG Groningen [email protected]

Bernd Vornefeld

Universität Bochum MA 6/47 Universitätsstr. 150, 44801 Bochum [email protected]

Barry

University of Toronto Centre for Urban & Community Studies 455 Spadina Avenue, Toronto M5S 2G8 wellman@chass. utoronto. ca

Richard

Barry

Wellman

Zijdeman

Wellman

University of Utrecht ICS/Department of Sociology Faculty of Social Sciences Heidelberglaan 2, 3584 CS Utrecht [email protected] University of Toronto Centre for Urban & Community Studies 455 Spadina Avenue, Toronto M5S 2G8 wellman@chass. utoronto. ca