Gossip and Reputation in Business Networks

Transcription

1 European Sociological Review VOLUME 27 NUMBER DOI: /esr/jcp062, available online at Online publication 21 January 2010 Gossip and Reputation in Business Networks Gerrit Rooks, Frits Tazelaar and Chris Snijders In this article we examine the determinants of gossip in purchasing and supply management, where gossip is defined as talking about your business partner (in a negative way) to a third party. Although gossip is often conceived as mere small talk, we argue and show that gossip can be analysed and is being used as a rational response in the management of business transactions. The hypotheses are tested using a dataset consisting of about 400 problematic purchase transactions of over 350 buying firms, collected in Germany. Consistent with our hypotheses, gossip is in general more likely when the problem is larger, when there are many common third parties available to buyer and supplier, and when the business network is dense. Also, gossip is less likely when buyer and supplier have been in business together for a longer time, and more likely when buyer and supplier expect to do future business. We discuss the implications of our findings in light of the literature on informal management mechanisms in business. Introduction In the social sciences gossip is seen as a potentially important, but under researched area of inquiry (Noon and Delbridge, 1993). Gossip typically gets the role of being a coincidental by-product of research into other organizational matters [for an overview, see Waddington and Michelson (2007)]. It is then mostly seen as problematic, taking away attention from the real work (e.g. Baker and Jones, 1996; Therrien, 2004). There are, however, arguments that allow for the economic analysis of gossip as one mechanism to safeguard business relations: gossip as the vehicle of reputations. In his classic article Non-contractual relations in business: A preliminary study, Stewart Macaulay (1963) observed that businessmen often prefer to rely on a man s word in a brief letter, a handshake, or on common honesty and decency instead of contracts. Business was supported by a norm that was widely accepted: commitments were to be honoured in almost all situations, and one should produce a good product and stand behind it. Macaulay suggested that such norms of good conduct are enforced by the wish to continue successfully in business. Reneging on deals and delivering low-quality products tarnishes one s general business reputation, and a good reputation is the magical recipe for surviving in inscrutable markets. All firms enjoying a good name are spared from repeating the burden of proof at every transaction and are relatively sheltered from competitive threats, which makes few things more valuable that a good reputation (cf. Gambetta, 1994, p. 227; Burt, 2008, p. 27). Reputation also facilitates collaborations by creating a cost for misbehaviour, and defines social obligation and identity (Burt, 2005, pp ). Burt posits that it is the positive and negative stories ( gossip ) exchanged about you that defines your reputation (cf. Burt and Knez, 1996; Burt, 2008, p. 27). Poor performances may become the subject of discussion in the gossip exchanged by managers. In this way, gossip becomes the vehicle of reputations. ß The Author Published by Oxford University Press. All rights reserved. For permissions, please journals.permissions@oup.com

2 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 91 Many scholars agree that reputations help stabilize relations and safeguard transactions (Lorenz, 1988; Coleman, 1990; Kreps, 1990; Raub and Weesie, 1990; Ellickson, 1991; Jones, Hesterly and Borgatti, 1997; Richman, 2004; Bolton, Katok and Ockenfels, 2005). However, there is mixed empirical evidence and a debate concerning the precise merits of reputations in business relationships. Some authors find that in networks of buyer supplier relationships firms are more trusting than in isolated buyer supplier transactions (e.g. Lorenz, 1988; Rooks et al., 2000; Gierl and Baumbacher, 2002). Other authors however find no relation between embeddedness within a business networks and trust (e.g. Blumberg, 2001; Buskens, 2002). As a possible explanation for the absence of network effects on trust, Blumberg (2001) remarks that network-embedded firms are not only able to damage a partner s reputation, but that in the process of spreading information, its own reputation may be harmed as well: Once gossip is spread, people will not necessarily apply the principle in dubio pro reo to judge the truth of the gossip (Blumberg, 2001, p. 845). Hence, gossip should be used with caution because it may be costly, and whether gossip works well is dependent on contextual factors. A similar point was made by Williamson (1996, pp ): third parties may not be willing to contribute their own resources to help the spread the information or they may even use such information strategically to their own advantage. Although gossip is a fundamental process in the creation of reputations (Emler, 1994), it has been neglected as a research topic in inter-firm relations and business networks. The aim of this study is to fill this gap and contribute to the literature by studying the conditions under which business gossip emerges. We first review the literature on gossip and then derive hypotheses about conditions under which gossip is likely to occur. To test our hypotheses empirically, we analyse a comprehensive dataset on purchases of information technology in Germany. The article concludes with a discussion and recommendations for further research. Gossip in the Scientific Literature Gossip traditionally has a negative connotation, and is often merely seen as idle talk, chitchat, slander, or scandal mongering (Fine and Rosnow, 1978). In social scientific definitions of gossip it is seen as small talk with a (social) purpose. Gossip is accordingly defined in more neutral terms, such as evaluative talk about a person who is not present (Eder and Enke, 1991, p. 494), or...the process of informally communicating value-laden information about members of a social setting (Noon and Delbridge, 1993, p. 25). Some definitions are related to a specific field of study. For instance, Kurland and Pelled (2000, p. 429) define gossip as...informal and evaluative talk in an organization, usually among no more than a few individuals, about another member of that organization who is not present. Gossip is everywhere. Gluckman (1963, p. 308) for example notes: I imagine that if we were to keep a record of how we use our waking-time, gossiping would come only after work for some of us in the score. Likewise, Wilson et al., (2000, p. 347) think that... people in all cultures gossip with an appetite that rivals their interest in food and sex. Gambetta (1994) claims that gossip is universal; people indulge in it regardless of culture. Moreover, he states that evidence suggests that its content also may be universally shared: most topics which, say, Zinacantecos (a native Mexican people) gossip about coincide with the topics we gossip about (Gambetta, 1994: 214). According to Michelson and Mouly (2004, p. 189) extant research on gossip is primarily divided among anthropology, sociology, social psychology, literature, and communication. Especially in a business context, gossip is widely practiced but surprisingly little understood (Michelson and Mouly, 2004, pp ). As an activity in management and organizational studies, scholars have considered reputation but largely ignored gossip (Noon and Delbridge, 1993, p. 31; Kurland and Pelled, 2000, p. 428). What little organization-based research was conducted, tends to be narrowly focused or lacks analytical rigor (Michelson and Mouly, 2004: 190). We are clearly in need of theory equipped with solid micro-foundations that can enable us to make non-trivial predictions about gossip s social dynamics (Gambetta, 1994, p. 204). 1 Although plenty of rich qualitatively accounts of gossip in the literature can be found, the occurrence of gossip has hardly been studied quantitatively. In organization sciences, from an empirical point of view gossip has been studied mainly as a phenomenon that takes place within organizations (to the best of our knowledge no systematic empirical studies of inter-organizational gossip have been carried out). Wittek and Wielers (1998) find that gossip flourishes in social networks that have a large number of so-called coalition triads. That is, the gossipers have good relationships among themselves, but both gossipers have bad relationships with the object of

3 92 ROOKS, TAZELAAR AND SNIJDERS gossip. Gossip in the workplace has repeatedly been characterized as a tool of abuse in the context of workplaces (e.g. Davenport, Schwartz and Elliot, 1999). Kurland and Pelled (2000) argue that in the workplace gossip enhances the power of a gossiper over a gossip recipient. Although at first sight gossip seems to have only negative side-effects or at best has no purpose other than social entertainment, it certainly has positive side-effects as well, as has been noted by many social scientists (Michelson and Mouly, 2004: 199). Most researchers point to gossip s ability to deter opportunistic behaviour: if one knows that opportunistic behaviour will become known, opportunistic behaviour itself is deterred. For this mechanism to work, one needs a network that is dense enough so that the threat of gossip is strong enough to deter opportunistic behaviour. One can look at this as a form of informal social capital: the network is such that it facilitates the flow of gossip, which itself precludes opportunism (Coleman, 1990, 1991). So instead of considering gossip as a nasty by-product of the cohabitation of gossip-loving humans, it can be considered a useful governance mechanism that can work sideby-side with more formal governing mechanisms (Merry, 1984; Waddington and Michelson, 2007). Interpreting gossip in that sense, we now derive hypotheses on the conditions under which gossip is likely to occur. Hypotheses In this study we define gossip as talk between business people from distinct firms about a problematic transaction with an absent third party (Merry, 1984, p. 275; Burt, 2005, p. 105). 2 Permeating all derivations of the hypotheses is the notion that problems in business are a violation of the norm that one should live up to the promise of providing adequate goods and services. This gives rise to three reasons why one would use gossip in a business context (cf. Merry, 1984). The first reason is that people tend to spread negative information about others when they feel they are treated badly, as an emotional response. The second reason is that spreading gossip allows network members to mutually agree and confirm which kinds of behaviour are considered acceptable, and which not. Gossip is used as a way of testing and reconfirming the industry s norms with peers. The third reason is a more instrumental one: gossiping about a business partner to a third party is a way to maintain the sanctioning system and to harm the reputation of the cheating partner. Opportunism and the Size of Problems One factor that might affect gossip is simply the size of the problems encountered. The more matters went wrong, the more likely it is that business people will experience that the norm you should deliver as promised is violated irrespective of whether the partner is to blame for all problems. As a consequence, the stronger the incentive to talk about the problem to others will be. All reasons to be willing to gossip are in place: a transaction that has lead to a huge amount of trouble creates hostile feelings (see below) and makes for an attractive war story during social talk. Moreover, the more and bigger the problems are, the higher the need is to effectuate the implicit gossip threat: for the gossip mechanism to work it is necessary that especially the bigger problems get sanctioned through active gossiping. In addition, it is likely that there exists an effect of the perceived cause of the ex post transaction problem. It makes a difference whether a problem was caused by circumstances beyond the control of the transaction partner, such as misunderstandings or force majeure, or whether the problem was caused by active opportunistic behaviour of the transaction partner. When deceit-oriented violation of implicit or explicit promises (Williamson, 1985, p. 47) occurs it is known to have detrimental effects. Besides the direct damage caused by the opportunistic behaviour, it decreases trust in both the competence and the general trustworthiness of the other party that itself has a negative effect on relationship commitment (Morgan and Hunt, 1994). It is also well documented that deceitful violation of promises often leads to feelings of being treated unfairly and the urge to retaliate (Kaufman and Stern, 1988; Stirling, 1956). Recent research in experimental economics also supports the idea that unfair behaviour is likely to be informally sanctioned by gossip, even at the expense of costs to ego (Falk, Fehr and Fischbacher, 2005). In a prisoners dilemma with punishment possibilities, Falk et al. find that cooperator s punishment is mainly motivated by the wish to harm those who acted unfairly. We see no reason why the same arguments should not apply to managers. This leads to the following hypotheses: Hypothesis 1a: The more problems arise during or after a transaction, the higher the probability that gossip will occur.

4 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 93 Hypothesis 1b: If a transaction problem is (perceived to be) caused by opportunistic behaviour of a transaction partner, the probability of gossip is higher. Size and Kinds of Business Networks In theories of social control, gossip is often seen as a strategy to enforce norms in close-knit groups of individuals (Merry, 1984; Ellickson, 1991; Wittek, 1999). When a group norm is violated, gossip is used as a first mild form of control in a sequence of steps, gradually increasing in force (Ellickson, 1991, p. 214). 3 Gossip as such is a form of punishment, since it can damage a valuable reputation. This implies that the likelihood that a manager will talk with a third party about a problematic transaction will be greater if this third party itself is related to the object of gossip: only then will the gossip have its strongest effect. If a norm violation occurs, then gossiping to a complete outsider has no group serving function, and will therefore be less likely (Ellickson, 1991, p. 177). This argument also follows from a more evolutionary perspective. Gossip can be seen as fitness enhancing behaviour, which can be group serving, as well as self-serving. Since in settings without common third parties there is no coherent group in which to use gossip instrumentally, one reason for gossip vanishes, so gossip will be less likely in settings where a mutual network tie does not exist. Moreover, we can also relate gossip to the size of the network of common third parties. As Nee and Ingram (1998) note, network size affects a group s ability to establish, coordinate, and enforce incentives. In a triad, the costs of sanctioning are large, but the costs will be smaller if the group is larger. In our case, the risk that gossip will be detected, and the gossiper will be identified by the target of the gossip, is greater in the triad than in a set of four firms. Hence, the more third parties will be present, the more likely gossip will be, because both the effectiveness of and the opportunities to gossip are higher. In all likelihood this effect will not be linear. However, the marginal effect of a 10th mutual tie in the network will be smaller than the marginal effect of a fourth member in a network. Hypothesis 2a: The more third parties a buyer and seller have in common, the more likely that gossip will occur. Hypothesis 2b: The relation between the numbers of third parties on the probability that gossip occurs, shows a decreasing slope. The marginal effect of an additional third party will diminish to zero. The above two hypotheses concern effects of the number of third parties on the likelihood of gossip. Another characteristic of a business network that may affect gossip is the degree to which firms interact and have close connections with each other in a given sector. In accordance with the social network literature, we refer to such connectivity as network density. The closure s reputation mechanism is the reason why dense networks increase the potential costs to uncooperative behaviour (Coleman, 1988a,b; Granovetter, 1985; Putnam, 1993, pp ). The more closed a network, the more likely that misbehaviour will be detected through gossip and punished (Merry, 1984, p. 297; Burt, 2005, p. 110; Burt, 2008, p. 162). Gossip about an absent person or organization often occurs as a by-product of social interactions. The more frequent interactions between members of a business network are, the more opportunities there are for gossip about problematic transactions and the more effective and efficient gossip can be. Hypothesis 3: The denser the network of common third parties, the more likely it is that gossip will occur. The occurrence of opportunism is a serious norm violation. As Ellickson (1991, p. 191) notes, most scholars agree that there is a universalistic norm, so also in business, against lying and deceit. Hence, if opportunism occurs within a network there will be a strong motive to gossip, besides the earlier discussed emotional incentive. Furthermore, there may be a norm operative to spread information about opportunistic behaviour of actors within a network (Ellickson, 1991, p. 232). Hence, opportunistic behaviour should lead to gossip especially if the transaction is embedded in a business network. In isolated transactions opportunism will not (or less likely) lead to gossip, since in that setting there is no group, and gossip serves no control function. This leads to the following hypothesis. Hypothesis 4: The effect of opportunism on the likelihood that gossip occurs is higher within a network, than outside a network. Dyadic Relationships and Gossip Another factor that can influence the extent to which gossip will be used is the relation between the business partners. When business partners have done business

5 94 ROOKS, TAZELAAR AND SNIJDERS Table 1 Responses to the telephone interview Refusal Willing to participate in face-to-face interview Willing to participate in postal interview Does not meet requirements Total Halle/Leipzig 187 (11.8 per cent) 428 (27.1 per cent) 72 (4.6 per cent) 894 (56.5 per cent) 1,581 (100.0 per cent) Munich 421 (14.5 per cent) 479 (16.5 per cent) 115 (4.0 per cent) 1,883 (65.0 per cent) 2,898 (100.0 per cent) Total 608 (13.6 per cent) 907 (20.3 per cent) 187 (4.2 per cent) 2,777 (62.0 per cent) 4,479 (100.0 per cent) before, a certain degree of common understanding is usually present, and spreading negative information can seriously backfire if the business partner finds out about the gossiping. Under such conditions, gossip puts a long-standing business relationship on the line. Indeed, experimental evidence has shown that gossipers are more likely to withhold negative information about their long-term business partner (McAndrew and Milenkovic, 2002). The same kind of argument holds in case business partners can foresee that future business together is likely. The benefits of all possible sanctions, including gossip, should be weighed against the risk of a break-up in the relation. For single-shot interactions where no business has been done together before and future business encounters are unlikely, this argument does not hold and gossip will be less likely to backfire and hence more easily used. Hypothesis 5: If a common past between buyer and supplier exists, gossip is less likely to occur. Hypothesis 6: The higher the expected relationship continuity, the less likely it is that gossip occurs. Monitoring Capacity Buyers face monitoring problems if they cannot easily assess the quality of the product or service to be purchased. Note that such monitoring problems can be clearly distinguished from ex post problems during contract execution such as delivery delays, compatibility problems, or inadequate service. A lack of monitoring capacity serves as an indicator for the buyer s behavioural uncertainty and makes opportunism feasible and attractive for the supplier (Batenburg, Raub and Snijders, 2003, p. 148). For instance, delivering a product of inferior quality might go unnoticed when the business partner has no clue what to expect. Transaction cost theory argues that behavioural uncertainty will affect contractual ex ante governance (Williamson, 1985). Using transaction cost theory and theory on embeddedness effects in economic exchange Batenburg, Raub and Snijders (2003) investigated the extent to which effort invested by the buyer in writing and negotiating a contract can be explained by a lack of the buyer s monitoring capacity. Based on a comprehensive survey on the management of IT transactions in the Netherlands (N ¼ 964) they find that the more difficult it is for the buyer to monitor the transaction, the larger the effort invested in negotiating and contracting is (Batenburg, Raub and Snijders, 2003, p. 164). Starting from a similar framework Rooks, Raub and Tazelaar (2006) and Raub, Rooks and Tazelaar (2007) focused on the direct effect of monitoring capacity on the occurrence of ex post problems. Based on similar data (1205 IT-transactions of 775 buyers) they find that a lack of expertise and limited monitoring capacity of the buyer has a significant and substantial positive effect on the likelihood and size of ex post problems, even after controlling for the effort invested in ex ante contracting (Rooks, Raub and Tazelaar, 2006, p. 262; Raub, Rooks and Tazelaar, 2007, pp ). Here we continue this line of research, and focus on the effects of monitoring capacity on ex post problem management through gossip and informal sanctions. Although there are cases where the (cause of the) transaction problem is obvious, in practice matters are not always that clear cut. We assume that, especially when the buyer (or more general: receiver) of a good or service has limited expertise and relatively few ways to monitor the extent to which the good or service lives up to its promise, ex post problem management becomes more difficult. In order for sanctions to be credible and effective the buyer must be able to present his argument in a clear and unambiguous way. If not, sanctions in general and gossip and informal sanctions in particular are less likely. This argument holds in particular in the IT sector, where the amount of knowledge and expertise on the side of the supplier can contrast strongly with the lack of knowledge and expertise of the buyer.

6 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 95 Hypothesis 7: The higher the buyer s monitoring capacity, the more likely that gossip occurs Data and Method To test the hypotheses, we make use of a database resulting from a comprehensive survey of IT purchases by German small and medium-sized business (Berger, Kropp and Voss, 2002). This multi-purpose survey was based on two earlier surveys that were conducted in a related research project in the Netherlands (Batenburg and Raub, 1995; Batenburg, 1997; Batenburg and Van de Rijt, 1998). The questionnaire contained six major parts, concerning (i) the product and/or service, (ii) product and supplier selection, (iii) the relationship between the buyer and the supplier, (iv) the agreement and contracting, (v) the performance of the supplier, problems, and problem management, and (vi) the buyer and his or her relation to the supplier. In total, per transaction more than 300 items were scored. In the survey buyers of information technology in two regions, Halle/Leipzig and Munich, in (then East and West) Germany were sampled. At the time of the survey both of the regions (Halle/Leipzig and Munich) were economically prosperous regions (Berger, Kropp and Voss, 2001/2). The sample consisted of small- and medium-sized firms with employees. In order for a transaction from a given firm to be included in the survey, the (respondent of the) firm had to meet the following requirements: (i) the firm had managed the purchase itself, that is, the decisions about the product and supplier had been taken by the firm instead of a mother company; (ii) an employee of the firm who could give detailed information about the transaction had to be present; (iii) the transaction was completed not too long ago if possible not longer than 3 years; (iv) the transaction should involve only one supplier. To compile the sampling frame the yellow pages were used ( Gelben Seiten für Deutschland. Frühjahr 1999 ). The data collection was conducted in two phases. First, a member of the research team contacted the firm in the sample by phone to determine whether the firm met the requirements to be part of the survey, and if this was the case, whether the firm was willing to cooperate in the survey. If a firm refused to cooperate in the survey (608 times), some questions were asked about reasons not to cooperate. A large fraction of these firms indicated that they had too little time to cooperate (43 per cent), another part of the firms indicated that they saw no personal interest to cooperate (26 per cent), and a third reason that was often given was that the firm never participated in surveys (17 per cent). If a firm agreed to cooperate with the survey, and met the requirements, then a knowledgeable contact person who had been responsible for purchases of information technology was selected and an appointment was made for a face-to-face interview (if firms refused a face-to-face interview a questionnaire was mailed to them). The telephone interviews started in March 1999 and were concluded in August 1999 (Berger, Kropp and Voss, 2001/2; Berger, Kropp and Voss, 2002). All firms who agreed to cooperate were sent a letter of confirmation immediately after the phone interview. In this letter the selected transaction and the date and time of the appointment were named again. Shortly before the agreed face-to-face interview a member of the research team phoned the firm once more to confirm the appointment. If possible the respondent was asked after the interview whether he or she was willing to fill out a second (written) questionnaire. As a result of the high-care intensity and the personal assistance in filling out the questionnaires, the response rates to the face-to-face interview were high. The survey team realized 84.2 per cent of all the promised interviews. Additionally, 24.5 per cent of the respondents filled out a second questionnaire. The response to the mail questionnaire was clearly smaller (36.4 per cent). The response rates are presented in Table 2. The overall response rate was 49 per cent which is Table 2 Response rates Contacted firms Gross response rate (per cent) Firms that met the requirements Net response rate (per cent) Number of firms with interview(s) Number of transactions Halle/Leipzig 1, Munich 2, , Total 4, ,

7 96 ROOKS, TAZELAAR AND SNIJDERS Table 3 Percentages of respondents reporting at least some problems on various issues, and rank order (N ¼ 1,019) Issue/type of problem: Per cent Rank order Slow or late service Improper accompaniment Improper documentation Slow or late adjustment Late delivery Improper installation Incomplete product Deviation from specification Incompatibility Product too slow Costs exceeding expectation high in comparison with the standard response rates in organization research (see Kalleberg et al., 1996). The response rates in the region of Halle/Leipzig were higher than in the region of Munich. Our dataset contains complete and detailed data on N ¼ 1,019 IT transactions from 832 buying firms. From these 832 buying firms, 645 (¼78 per cent) provided data on a single IT transaction and 187 (¼22 per cent) provided data on a second IT transaction as well. Compared to firms in the region of Halle/Leipzig, firms in the region of Munich provided more second questionnaires (29 vs. 15 per cent). After completing the first four sections of the questionnaire, the respondents were presented with a list of 11 types of possible problems ( issues ) that might occur after the contract was signed and the goods were delivered. For each and every issue the question was asked whether, and if so to what extent, these problems actually were experienced during the supply of the product or service. In Table 3 an overview is presented of the various issues and the incidence of problems reported (percentages and rank order). In 435 out of the N ¼ 1,019 transactions (¼43 per cent), at least one problem occurred at least to a certain degree (see Measurements section, below). These N ¼ 435 problematic transactions form the core database on which our hypotheses will be tested. Measurements In this section the construction of the empirical variables is discussed. First we elaborate on the dependent variable, gossip, and then on the construction of the explanatory variables. We end this section with a brief discussion of the control variables that are included in the analyses. Gossip The occurrence of gossip is measured by taking answers from two sets of questions. First, if a transaction was problematic, the respondent was asked how those problems had been managed. A comprehensive list of different problem management possibilities was presented to the respondents. This list was successfully used in two related research projects on purchase management in the Netherlands (Rooks and Snijders, 2001; Snijders and Tazelaar, 2009, ch. 5). It included questions such as Has the supplier been contacted about this problem?, Has there been any form of discussion about the problem with the supplier?, Were any kind of sanctions executed, and if yes, which ones?, Were these measures limited to delaying or withholding payment, or were more severe measures (also) taken, such as involving lawyers, mediators, arbitrators, etc. or ultimately the court?. In this list of issues the question was also put forward whether the respondent had notified other customers of the supplier. In almost 5 per cent of the cases the buyer had notified one or more other customers of the supplier. Second, a question was asked as to whether the respondent had talked with colleagues in other firms about the problems. The variable GOSSIP is constructed on the basis of the answers to both questions: gossip is defined to have occurred if during problem management other customers of the supplier were notified or if there had been informal communication with colleagues in other firms. GOSSIP is a binary variable (1 ¼ yes, it occurred, 0 ¼ no, it did not). Measured as such, gossip took place in 26 per cent of all 435 problematic cases. Ex post Problems The occurrence of ex post problems is measured using detailed information on problems that occurred during and after the focal transaction. In the survey, questions were asked about 11 typical problematic issues that are often associated with IT transactions (Riesewijk and Warmerdam, 1988; Rooks, 2002, ch. 4). Respondents could indicate for each issue to what extent it had occurred and how serious the problem had been. The eleven issues are measured on a five-point scale (ranging from not whatsoever to very severe ). The variable PROBLEM SIZE is derived as a scale score on these eleven issues (Cronbach s

8 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 97 alpha ¼ 0.92; higher represent more and/or more serious problems). To examine validity, we used additional data from the survey. Buyers were asked to indicate their satisfaction with both the product and the supplier by providing a grade between 1 and 6 the scale used for grades in German schools for the product and the supplier. Both satisfaction variables correlate strongly and significantly with PROBLEM SIZE (average correlation coefficient ¼ 0.52; P50.001). In 43 per cent of the transactions (435 out of N ¼ 1,019) at least one problem occurred at least to a certain degree [variable: PROBLEM DUMMY]. Opportunism The degree of opportunism was measured by taking answers from two sets of questions. The first set consists of four questions, which were based on items used in Provan and Skinner (1989). The respondents were asked whether, during the focal transaction, the supplier always gave a reliable image of business, obscured facts, did not keep promises, and was honest and frank. The respondents could answer on a five-point scale ranging from not at all to very much so. The second set consists of three questions, which were based on items used in Rooks and Snijders (2001) and Rooks (2002). The respondents were asked what they thought would be the cause(s) of the problems that they had encountered: supplier s indolence, supplier s incompetence, and supplier s self-interest and opportunism. Here too, the respondents could answer on a five-point scale (ranging from certainly not to certainly true ). Factor analyses showed that the seven items measure a single dimension. Cronbach s alpha scale reliability coefficient of these items is 0.84, which is an indication of a reliable scale (Nunnaly, 1978): OPPORTUNISM. Number of Common Third Parties A number of questions were asked about the business network in which the transaction was embedded. The purchase manager was asked to take in mind the other firms (s)he knew who customers of the supplier were at that time. We used the resulting number of common third parties as an estimate of the size of the business network. An alternative and more elaborate method to estimate the network size is to use a name-generator, however in a large-scale quantitative study Fu (2003) showed that a single-item measure of the kind we also use here was comparable and consistent with more complicated network measures. The distribution of the number of third parties known by the respondent was highly skewed. To cope with this and possible outlier effects the number of firms was maximized to seven: OUTDEGREE. Density of Network of Third Parties When the respondent indicated that there was a social network surrounding the dyadic relation a number of questions were asked about the density of this network. Respondents were asked to indicate the degree to which they agreed (on a five-point scale) with the following statements about the network: there are frequent contacts, and like knows like. The reliability of this scale is somewhat low; only two items are included (r ¼ 0.44; Cronbach s alpha ¼ 0.61 for DENSITY). Network Acquaintance The respondent was asked whether (s)he was acquainted with any of the business partners of the supplier, and if so, which type of business partner(s) (s)he was acquainted with. Five types were given: (i) (other) customers of the supplier [yes/no], (ii) suppliers of the supplier [yes/no], (iii) bank(s) of the supplier [yes/no], (iv) accountants and/or tax consultants of the supplier [yes/no], and (v) other types of business partners of the supplier [yes/no]. If any of these categories was labelled yes by the respondent, the dummy variable NETWORK ACQUAINTANCE was given a 1, if not a 0. Common Past The questionnaire contained four questions concerning the previous relation and transactions with the focal supplier. The questions concerned the length of the relationship, the frequency of previous transactions, the volume of previous transactions, and the satisfaction with previous transactions. Factor analyses indicated that there was one dimension underlying these items. A consistency analysis indicates that the scale is reliable (Cronbach s alpha ¼ 0.89 for PAST). Expected Relationship Continuity The questionnaire included a question on whether the purchase manager had expected, before the focal transaction was concluded, that future transactions with the supplier would be likely. Respondents could choose between five response categories (ranging from no expectation of future transactions to very regular and/or very sizeable future transactions were expected ). We use the score of the buyer on this

9 98 ROOKS, TAZELAAR AND SNIJDERS question for the construction of the variable FUTURE. Note that the measurement of this variable is problematic (see Batenburg, Raub and Snijders, 2003, p. 168; Buskens, 2002, pp ). The respondent had to recall an expectation; often from a number of years ago. The accuracy of answers to such retrospective questions on attitudes rather than behaviour is doubtful (Bernard et al. 1985), but no better measurement is available in the given data. Monitoring Capacity Of the Buyer We measure the monitoring capacity of the buyer as a factor score, based on five questions. These questions ask for the difficulty to judge the quality of the product(s) (five-point scale), the difficulty to compare the different tenders for the product(s) (five-point scale), the buyer s experience with IT-products and services relative to similar other firms (five-point scale), whether the buyer firm had one or more employees with specific knowledge and experience in IT, and whether the buying firm would be capable to assemble and/or produce the product(s) itself (five-point scale). These five items lead to a single factor MONITOR. Control Variables Based on the extant literature about inter-firm relationships and transactions we include five control variables in the analyses, three of which are derived from transaction cost theory (Williamson, 1985; Williamson, 1991), and two from social theory (Coleman, 1990, ch. 11; Coleman, 1991): Switching costs One of the core concepts of transaction cost theory is that of switching costs. Whenever switching to an alternative partner would be costly, a form of dependency is created that might affect the extent to which informal mechanisms such as gossip will be used. We use four questions on different types of switching costs. A principal component analysis revealed that these form one cluster of variables (factor 1 explains 63 per cent of the variance). Furthermore, the resulting scale is reliable (Cronbach s alpha ¼ 0.80 for SWITCHING COSTS). Financial volume To control for the volume of the transaction, we include the answer to the question How much money did you have to pay the supplier for this product and/ or service in total (all later extensions excluded)? : TRANSACTION SIZE. Product complexity To control for confounding effects of product characteristics we include PRODUCT COMPLEXITY, a variable based on 20 characteristics of the purchased IT-product(s) and/or service(s), classified in five categories, from (i) straightforward and standard to (v) highly complex. Power differences and reputation Coleman asserts that powerful actors in communities are less likely to obey the norms of that community and are also less likely to be sanctioned if they do violate the norm (Coleman, 1990, pp ). Moreover, he also suggests that informal sanctions will be less effective against those who already have a bad reputation (Coleman, 1990, p. 287). We construct the variable SUPPLIER POWER based on the log-difference between the size of the supplier and the size of the buyer in terms of full time equivalents. The perception of the buyer of the supplier s reputation is measured by providing a grade between 1 and 6 the scale used for grading in German schools for the reputation of the supplier. The variable LOW REPUTATION SUPPLIER is constructed as a dummy variable by taking the firms with the three lowest grades. Table 4 shows the descriptive statistics for the variables used in the analyses. Cases with missing values on the gossip variable are excluded from the descriptive and explanatory analyses. Results Since gossip is a binary variable, it either occurred or not, logistic regression analysis was used to test our hypotheses. Table 5 lists the results of five logistic regression models of gossip between business people. The first model concerns all problematic transactions, restricted to those of which we have data for all other variables and controls (N ¼ 387). In the second model we replace the variable OUTDEGREE by a dummy variable, capturing whether the respondent knows at least two other buyers from the supplier (OUTD_BIG). This is because closer inspection of the data, using dummies for the separate OUTDEGREE categories, showed that in fact there is virtually no effect for small values of OUTDEGREE and a fixed-size positive effect for larger values of OUTDEGREE. In the third model

10 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 99 Table 4 Correlations between the variables as used in the logistic regression analysis (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Gossip (1) 1.00 Problem Size (2) Opportunism (3) Outdegree (4) Density (5) Past (6) Future (7) Monitor (8) Product complex (9) Switching costs (10) Transaction size (11) Power (12) Low repu.partner (13) Legend: P50.10; P50.05; P50.01; P Cases with a missing value for gossip are excluded. we include the variable DENSITY. It has a lot of missing values, so it decreases our estimation sample size dramatically, but we wanted to at least show it for the cases where we have measured this variable. In model 4 and 5 we run our analyses on two separate samples: the isolated encounters (model 4) and the embedded encounters (model 5). We see that across all models except 3 and 4, there is a positive and significant effect of the size of the problem on the probability of gossip (P50.05). In model 3, apparently the inclusion of the density of the network moderates the effect of the size of the problem. Furthermore we see that the effect of the size of the problem occurs mainly in embedded transactions, not in isolated ones. This suggests that indeed more and larger problems lead to voice more often. This implies that the effect in models 1 and 2 is largely determined by the embedded cases, rendering partial support of our Hypothesis 1a. Hypothesis 1b is not supported. After controlling for problem size and other variables, there is no additional effect of the fact that the problems were caused by opportunism (P440.10). Some caution is warranted here, because of the large correlation between the size of the problem and the degree of opportunism (r ¼ 0.65, P50.001). Analyses with problem size and opportunism included separately show an effect of each. However, when included together the effect of the problem size clearly wins. Hypothesis 2a is supported by our results: it is indeed the case that a larger network (a higher OUTDEGREE) leads to gossip more often (P50.05). However, as far as we can determine, the effect is stepwise: non-existent for small networks and significant and positive if the network is large enough. We had actually expected a concave shape of this effect (marginal returns decreasing to zero), so in that sense Hypothesis 2b is not supported, but the effect is indeed non-linear. The density of the network has a strong positive effect on the probability of gossip (P50.001), supporting Hypothesis 3. Some caution is in order here because there were some measurement problems with this variable and a large fraction of missing values, but for those cases where we have a measurement, the effect is highly significant (P50.001). Hypothesis 4 concerns the differential impact of the effect of opportunism on gossip. Although a direct test cannot be seen in the table, the differential impact is not statistically significant (P40.1) and even in the wrong direction, rejecting Hypothesis 4. We do see that the gist of the arguments underlying Hypothesis 4 does hold when we consider the size and severity of the problems. Then we do see a significant difference: a

11 100 ROOKS, TAZELAAR AND SNIJDERS Table 5 Logistic regression models. Dependent variable ¼ GOSSIP Variable Model-1 Model-2 Model-3 Model-4 Model-5 Problem size Opportunism Outdegree 0.19 Outd_big Past Future Monitor Product Complexity Switching costs Transaction size Low repu.partner Power Density 0.65 _cons N Legend: P50.10; P50.05; P50.01; P Model 1: all main effects and controls included. Model 2: dummy-variable OUTD_BIG (¼ large outdegree ) captures effect of outdegree. Model 3: density added (decreasing N because of missing variables). Model 4: as model 2, but estimated for isolated transactions [dummy NETWORK ACQUAINTANCE ¼ 0]. Model 5: as model 2, but estimated for embedded transactions only [dummy NETWORK ACQUAINTANCE ¼ 1]. strong positive effect of the size of problems in embedded settings versus a weak positive effect in isolated settings (difference significant at P50.05; not reported in Table). The past and expected future of a relationship also affects the likelihood of gossip. A relation with a longer history leads to less gossip across all models, with the exception of the cases where we consider isolated business transactions. This clearly supports Hypothesis 5. However, Hypothesis 6 is refuted. In all but the isolated business settings there is a positive effect of having high expectations of the likelihood of doing future business with the same supplier on gossip, contrary to our hypothesis. The more the respondent expects to deal with this particular business partner in the future, the more gossip occurs. A higher monitoring capacity indeed goes with an increased likelihood of gossiping. The effect is not that strong. Again we see that effects, if they exist, occur in the embedded settings, giving Hypothesis 7 some partial support. Finally, we find that our control variables do not add much to the explanation of gossip: switching costs, transaction size, differences in power between supplier and buyer, and whether the partner has a low reputation, have no significant effect whatsoever on the likelihood of gossip. We only find a small positive effect of the product s complexity: more complex products go with an increased probability of gossip. Once again we find that the effect occurs mainly in an embedded context. Conclusion and Discussion In this study we examined determinants of the likelihood that managers will gossip, defined as talking in a negative way about an absent business partner with a third party. Contrary to what is commonly thought, we show that gossip is a fundamental process in the creation of reputations and can be used as a viable governance mechanism. We tested hypotheses about determinants of gossip using a dataset with about 400 problematic transactions between German buyers and suppliers. Consistent with the hypotheses, it appears that gossip about a problematic transaction is in general more likely when the problem is bigger and more severe. Gossip is also more likely to occur if the problematic transaction is embedded in a diverse and dense enough network. Gossip is less likely when monitoring problems for the buyer are high and occur in existing business relations. Our findings are stronger precisely in embedded relations, which highlights that gossip can indeed be understood as (also) instrumental and managerial in nature, as opposed to only an emotional response based on feelings of being double-crossed. This is somewhat surprising because

12 GOSSIP AND REPUTATION IN BUSINESS NETWORKS 101 especially when gossip is the topic under study one might think that personal and relational rather than instrumental or socioeconomic arguments play a major role. When we measure the closeness of the personal relationship between the individuals involved in the deal, using questions such as how well they know each other, whether they know each other s birthday, what kinds of things they do together outside work, and so on, we find a strong scale (seven items, Mokken s H ¼ 0.70, alpha ¼ 0.82). However, the closeness of the personal relation has no effect whatsoever, linear or non-linear, on the likelihood of gossip. We feel this strengthens our results on the economic correlates of gossip. Apparently, it is not just the case that socioeconomic variables might have an extra effect in addition to a main relational effect (you gossip with the ones you are close to). In fact the relational effect is simply not there and the socioeconomic variables are the only ones correlating with gossip. An unexpected finding was that gossip is more likely in those cases where the expected future with the same business partner is long. With hindsight, there are two reasons we can think of that might be causing this result. First, a possible reason is that in such cases the transaction partners expect their relation to continue because of the fact that they are likely to meet and do business again. Given this, the relationship can withstand some noise. In that sense, strongly interdependent actors can benefit more from actively controlling and sanctioning each other because the risk of a breach is negligible. Another possible reason is based on another instrumental advantage of gossip. Thus far, we emphasized that the instrumental advantage of gossip is based on maintaining the sanctioning system and deliberately harming the reputation of the other party. This argument loosely follows the typical logic of conditional cooperation and game-theoretic rationality (although sanctioning itself creates a second-order free rider problem that we neglected). However, there is another kind of instrumental purpose of gossip. That is that one might want to use gossip to test or calibrate one s own opinions about the other party (Gambetta, 1994, p. 213; Gulati and Westphal, 1999, pp.: ;). The aim then is to gain or renew information about the capacities and credibility of the business partner, information that is otherwise difficult to obtain (cf. Levin and Arluke, 1987; Michelson and Mouly, 2004, p. 195). In such a way gossip may help to become more certain about the partner (Labianca, Brass and Gray, 1998). The more likely it is that future business with the same partner can occur, the more necessary it is to be certain about the business partner, and the more likely it is that gossip will occur. Although we are aware that the measurement of such intentions is notoriously difficult in a survey (cf. Jaeger et al., 1994), in particular when the topic itself, gossip, might have a dubious connotation (Fine and Rosnow, 1978; Michelson and Mouly, 2004, pp ), we have some evidence that is consistent with this claim. When asked why there had been talk with a third party, one out of every four gossipers gave the instrumental reason we wanted to warn third parties for the misconduct of the supplier. Three out of four did not make an explicit reference to such an informal sanctioning intention. However, the vast majority of these other gossipers (79 per cent) had other instrumental reasons to gossip. They stated that they wanted to get information of third parties concerning opinions on such misconduct or on such problems occurring. Only a small minority of these other gossipers (the remaining 21 per cent) did not make such a reference to an intention to calibrate. For these others, gossip was simply small talk. When we redo our analyses and confine ourselves to only the cases where informal sanctioning intentions were mentioned, the positive effect of the shadow of the future largely disappears. And, the positive effect of the shadow of the future on gossip surfaces precisely in those cases where no explicit reference to such an informal sanctioning intention was given. Another intriguing finding in the light of standard theories of business interaction is that we find not much evidence of a separate role of opportunism in predicting gossip. That is, when the business partner has behaved opportunistically there will of course be more problems, and when there are more problems there is a higher likelihood of gossip. But, the fact that the problem was largely caused by opportunistic behaviour has no separate effect after controlling for the number and severity of problems that it caused. We leave it up to the reader whether this particular finding is in line with standard economic theory, transaction cost theory, or other theories of the firm. We feel that it largely corroborates those theories arguing that opportunistic behaviour can lead to problems that should be managed efficiently through transaction governance. When, as we find in our data, there is no separate effect of opportunism remaining, apparently the number and size of problems that occurs mediates the effect of opportunism on gossip. Or, stated differently, we find no effect of a psychological effect of opportunistic behaviour: it is not the case that buyers are more likely to gossip when the cause of the problem is opportunistic behaviour: it is the fact that there is a problem that matters.