Reputation and Trust: A Multi-Dimensional Perspective

Personal reputations are an important factor in trust. Prior work has focused on how favorable reputations lead to higher trust than unfavorable ones. In this paper, I examine the effects of other reputation dimensions on trust, specifically reputation breadth (how widespread the reputation is) and reputation consensus (how well agreed upon the reputation is). In two studies, I present participants with reputation information that varies in favorableness, breadth and / or consensus. I find that, for positive reputations, both breadth and consensus increase trust. For negative reputations, consensus reduces trust, but breadth only has weak effects on trust. Introduction Employees frequently shift work relationships as they engage in project-based teams, rotate through departments, and advance in their careers (Burt, 2006; Kilduff, Tsai & Hanke, 2006). In such dynamic environments, personal reputations are important currency that individuals can use to foster trust with new colleagues (Ferris, Blass, Douglas, Kolodinsky & Treadway, 2003; Kollock, 1994). Employees whose positive reputations precede their entry into a group may find it easier to build trust than if they had entered the group as unknown entities. In contrast, employees entering the group with negative reputations may have more challenges in trust-building than if they had been completely unknown to the group members. Researchers have repeatedly found that positive reputations enhance trust and negative reputations harm it (Buskens, 2003; Buskens & Weesie, 2000; Kollock, 1994). Although these findings establish an important link between reputations and trust, they do not completely characterize the reputation-trust relationship. Reputations vary on other dimensions besides favorableness (Rindova, Williamson, Petkova & Sever 2005; Fischer & Reuber, 2007). For example, people’s reputations differ in how widespread they are (breadth) and how well others agree about them (consensus). Notably little research has examined how actors use these other reputation dimensions in trust decisions. Prior work on social networks has recognized the important role that third parties play in trust (Burt & Knez, 1995; Ferrin, Dirks & Shah, 2005) and has shown that trust is related to the number of third parties linking the trustor and trustee. These studies hint to the importance of reputation breadth, but neither study examines it explicitly. To my knowledge, no prior work has examined how the three reputation dimensions of favorableness, breadth and consensus combine to influence trust. In this paper, I use experimental studies to vary the three dimensions and examine their influence on trust. First, I define reputation and review research characterizing it as a multi-dimensional construct. Second, I draw from research in attribution and advicetaking to introduce a model describing the relationship between trust and reputation favorableness, breadth and consensus. Third, I describe two studies that test the model, using a vignette experiment and a laboratory experiment. Finally, I discuss these findings, limitations of the studies and directions for future research. Going forward, I will refer to the person deciding about trust as the actor and the person whose reputation is of interest (and who the actor may or may not trust) as the target. Reputation as Multiple Dimensions Reputation is typically defined as “an attribute or characteristic ascribed by others” (Raub & Weesie, 1990, p. 630). Raub and Weesie (1990) distinguish two views of reputations as reputations in a narrow sense and reputations in a broad sense. A target’s narrow reputation forms through direct interactions with the actor; in trust decisions, the actor only uses information about how the target has treated him or her in the past (Rousseau et al., 1998). In the narrow sense, if an actor has had no personal experience with a target, the target will not have a reputation with the actor. In contrast, broad reputations include “reputation vis-a-vis third parties with whom [the target] interacts in other situations and who happen to acquire information about [the target’s] behavior in situations in which they are not directly involved” (Raub & Weesie, 1990, p. 631, emphasis in original article). In the broad sense, a target can have a reputation with the actor without directly interacting with him or her. In this research, I am interested in reputations in a broad sense. I examine how third parties’ information about a target influences trust when the actor has no previous experience with the target. Before looking at how actors use reputation information, we must have a clear idea of what reputation information is available for use. A target’s reputation can be characterized by multiple dimensions (Bromley, 1993; Fischer & Reuber, 2007; Rindova et al., 2005). Most research has focused on reputation favorableness, i.e., how positively or negatively the target is viewed, on average. In addition, reputations vary in breadth, or the number of people who have perceptions of the target’s attribute. For example, a new accountant known to only a few department members would have low reputation breadth. In contrast, a new CEO may have high reputation breadth without meeting many of the employees who hold a perception of his or her integrity. In this way, reputation breadth differs from social network measures of centrality. Breadth includes anyone with a perception of the person’s attribute (in this case, integrity), while centrality includes only those with whom a person has direct contact, ranging from frequent and close (“strong ties”) to less frequent and less close (“weak ties”). Prior studies on social networks have found the average employee has about eight ties at work (Lawrence, 2006), but many participants in Lawrence’s study “knew of” more than 50 fellow employees. 1 Other researchers have defined similar dimensions as prominence (Rindova et al., 2005) and scale (Bromley, 1993). Prominence incorporates other concepts beyond breadth, such as how salient the target is in the mind of others. Scale is calculated as a percentage of network members with perceptions of the target instead of a count of these individuals. Using either of these conceptualizations decreases the applicability of opinion integration research. Instead, I use the term breadth to maintain consistency with Raub & Weesie’s terminology. Reputations also vary in consensus (Fischer & Reuber, 2007; Hoyt, 1994) or how well people agree about the target’s attribute. For example, everyone may agree that the accountant has high integrity, giving him or her a high consensus reputation. In contrast, some people perceive the CEO to be a person of high integrity and other people perceive the CEO to have questionable integrity or no integrity. Related to Kelley’s (1967) attribution model, reputation consensus provides information about a target’s consistency, i.e., how similarly the target has behaved in the past. Reputation consensus differs from Kelley’s notion of consensus, i.e., that other people would behave as the target did in the given situation. Taken together, the three dimensions of favorableness, breadth and consensus can be thought of as parameters that characterize the sampling distribution of the person’s reputation. Favorableness shifts the mean of the distribution, consensus changes the spread of the distribution, and breadth changes the sample size. Figure 1 depicts reputations that vary on all three dimensions. The top left distribution represents someone with a well-known and agreed upon positive reputation. The bottom right represents someone with a little known and controversial negative reputation. In the next section, I consider how these dimensions inter-relate to influence trust. Reputation Dimensions and Trust Trust is defined as the willingness to be vulnerable to another person in an uncertain situation (Rousseau, Sitkin, Burt & Camerer, 1998). Trust is based on how positively the actor perceives the target’s attributes (Mayer et al., 1995), and how confidently the actor holds these perceptions (Burt & Knez, 1995; Deutsch, 1960; Doney et al., 1998). Two employees may both estimate a colleague to be “very trustworthy”, but one may claim “I am 100% positive that this colleague is very trustworthy” while another hedges, “this colleague seems very trustworthy to me”. Although perceptions of the colleague’s trustworthiness are high for both employees, the more confident employee will have higher trust in the colleague than the less confident one. Individuals can have reputations regarding multiple attributes, such as reputations for intelligence, friendliness and diplomacy. Perceptions of a target’s trustworthiness are derived from three main attributes: the target’s ability, integrity and benevolence (Mayer, Davis & Schoorman, 1995). In this paper, I focus on reputations for integrity. A person’s integrity is defined as the degree to which he or she acts in accordance with morally justified principles (Becker, 1998). Integrity includes specific characteristics such as honesty and fairness (Becker, 1998; Butler & Cantrell, 1994; Mayer et al., 1995). Researchers have also used the terms morality and character to identify the similar attributes (Morgan Roberts, 2005; Wojciszke, 1994). Reputation information can significantly affect both the valence of perceptions and the confidence with which they are held (Burt & Knez, 1995). Below, I describe how each dimension relates to trust through one or both of these variables. Favorableness and Perceived Integrity Reputation favorableness primarily impacts trust through perceptions of the target’s trustworthiness. In forming his or her own perceptions of the target, the actor will rely on others’ information about the target’s integrity. A favorable integrity reputation indicates that others have found the target to have high integrity in the past. Actors consider this information relevant in forming their own perceptions of the target’s integrity (Buskens & Weesie, 2000; Camerer & Weigelt, 1988). Although few studies directly examine the effect of reputation favorableness on perceptions of integrity, several studies have established a relationship between reputation favorableness and judgments and behaviors indicative of trust. In an experiment in which buyers could bid for various sellers’ goods, Kollock (1994) found that buyers were willing to pay a premium to, and establish a commitment with, sellers who had been truthful about their product quality in earlier exchanges with those buyers. King-Casas and colleagues found that a trustee’s past cooperation in a social dilemma increased the speed by which trusters decided to cooperate with that trustee in future rounds (King-Casas, Tomlin, Camerer, Quartz, & Montague, 2005). In an online negotiation experiment, participants viewed their counterparts less favorably, and shared less information with them, when they were told the counterpart had a distributive reputation than when they were not given reputation information, even though the reputations were randomly assigned (Tinsley, O’Connor & Sullivan, 2002). Based on these studies, I expect reputation favorableness will influence trust by affecting the actors’ perceptions of the target’s integrity. H1: Trust will be higher for targets with favorable reputations than unfavorable ones. H2: Perceived trustworthiness will mediate the relationship between reputation favorableness and trust. Consensus, Breadth and Confidence Trust is affected by both perceptions of trustworthiness and the confidence with which they are held. Kelley’s model of attribution indicates that people will have more confidence in their perceptions of a target when that target has high consistency in behavior, i.e. that the target behaves in a similar way over time (Kelley, 1967). Several studies have supported the positive effect of consistency on confidence (see Fiske & Taylor, 1991 for a review). This research has typically focused on a target’s actual past behavior as opposed to others’ judgments of the target, which may or may not be based on direct observation of the target’s behavior. In measuring agreement among others’ judgments, reputation consensus provides a signal regarding the consistency of a target’s past behavior. If people strongly agree in their perceptions of a target’s trustworthiness, the actor will infer that the target has had a consistent level of integrity in past interactions. This inference of consistency should increase confidence in the actor’s perceptions of the target. Thus, I expect reputation consensus to increase an actor’s confidence in his or her perceptions of the target. Confidence does not have a direct effect on trust, but moderates how strongly perceptions of integrity influence trust. Being confident in one’s positive perceptions of a target will lead to higher trust, but being confident in one’s negative perceptions of a target will lead to lower trust. That is, I expect reputation consensus to increase trust for positive reputations, but decrease trust for negative reputations. H3: Trust will be higher towards targets with positive, high consensus reputations than targets with positive, low consensus reputations. H4: Trust will be lower towards targets with negative, high consensus reputations than targets with negative, low consensus reputations. H5: Confidence in perceptions will mediate the effect of consensus on trust. Reputation breadth can also influence confidence. In the advice-taking literature, psychologists have found that when an actor uses opinions from multiple advisors, the number of advisors increases the actor’s confidence in his or her judgment (Ashton, 1986; Budescu & Rantilla, 2000). An actor considering more opinions will feel that his or her judgment is based on more evidence than an actor who considers few opinions. However, Budescu and Rantilla (2000) found that the number of advisors only increased confidence when the advisors had high consensus. When the advisors had low agreement, having more advisors’ opinions did not help confidence and sometimes decreased confidence. This work can easily apply to reputations where the number of advisors is analogous to reputation breadth. Directly applying the above findings, I expect a target’s reputation breadth to increase the actor’s confidence in perceptions of the target, but only when reputation consensus is high. Hearing consistent opinions about a target from many people will lead the actor to be more confident than hearing consistent opinions from very few people. However, hearing divergent opinions from many people will lead to similar, low confidence as hearing divergent opinions from a few people. H6: When reputation consensus is high, trust will be higher towards targets with positive, high breadth reputations than targets with positive, low breadth reputations. H7: When reputation consensus is high, trust will be lower towards targets with negative, high breadth reputations than targets with negative, low breadth reputations. H8: Confidence in perceptions will mediate the effect of reputation breadth on trust. The Negativity Bias and Reputation-Trust Relationships A large literature in social cognition demonstrates that people generally weigh negative social information more strongly than positive social information (see Baumeister, Bratslavsky, Finkenauer, Vohs, 2001; Skowronski & Carlson, 1989; and Rozin & Royzman, 2001 for reviews). Regarding integrity, negative information is considered more diagnostic of the person’s disposition than positive information (Madon, Jussin & Eccles, 1997; Skowronski & Carlson, 1989; Snyder & Stukas, 1999). Situational explanations for negative behavior are less compelling than situational explanations for positive behavior. Even individuals with low integrity act upstanding at times, due to social conformity, someone monitoring their behavior, or high costs of being caught. Thus, a single positive act does little to help categorize the person as a person with high integrity or low integrity. In contrast, there are few situations that people expect someone with high integrity to lie or steal (Reeder & Brewer, 1979; Skowronski & Carlson, 1989). As a result, a single negative act is often considered enough evidence that the person has low integrity (Kim et al., 2004). Applying this work to reputations, I expect a single favorable opinion to be viewed as less diagnostic than a single unfavorable opinion. An actor will believe that a person who is honest with one individual may have many reasons for behaving that way, such as conformity or impression management concerns. In contrast, a person who was dishonest with one person is more likely to be perceived as having a dishonest disposition. As a result, fewer opinions may be required to be confident in one’s negative assessment of a target compared to a positive assessment. Thus, breadth should affect positive reputations more than negative ones. A target’s negative reputation may not need to have high breadth to decrease trust, but a target’s positive reputation may need to have high breadth to increase trust. H9: The effect of breadth on trust will be stronger for positive reputations than negative ones. Figure 2 depicts a model summarizing these hypotheses. I examine the hypotheses in two studies. In the first study, I used a vignette experiment to examine the effect of favorability and breadth on trust, testing hypotheses 1-2, and 6-9. In the second study, I examine how all three dimensions influence trust in a laboratory study, testing hypotheses 1-9. In both studies, I examine situations where the counterpart lacks relevant experience with the actor, i.e., situations in which counterparts commonly rely on reputational information. Study 1 In the first study, I used a vignette to examine the effects of reputation favorableness and breadth on trust. Participants first read a vignette about working with a colleague they did not know, but about whom others had perceived to be honest or dishonest. Following the vignette, participants rated their intentions to trust the colleague, perceptions of the colleague’s honesty and confidence in the accuracy of these perceptions. Materials Vignette. In the vignette, participants were told that they had moved into a new role in which they would be working closely with an employee they hadn’t met before. I manipulated the reputation favorability and reputation breadth of this employee in a 2 (honest vs. dishonest) by 2 (low vs. high breadth) between-subjects design. In the low breadth condition, participants were told “Curious about the employee, you asked six members of your organization about him. One colleague perceived the employee to be honest and trustworthy (or dishonest and untrustworthy). The other five people you asked had no impression of the employee.” In the high breadth condition, participants were told “Curious about the employee, you asked six members of your organization about him. Five people perceived the employee to be honest and trustworthy (or dishonest and untrustworthy). One person you asked had no impression of the employee.” Trust, trustworthiness and confidence. Following the vignette, I measured participants’ intentions to trust the employee using a ten-item scale from Dunn and Schweitzer (2005) (alpha = 0.86). I measured perceptions of the colleague’s integrity using items from Mayer and Davis (1999) (alpha = 0.81) that were adapted to address a colleague instead of top management. Finally, I measured participants’ confidence in their integrity perceptions using three questions, “how confident are you in the accuracy of your assessment of this person?”, “how likely is it that you have misjudged this person’s integrity?” (reverse-scored) and “how surprised would you be if this person’s integrity was different than your perceptions?” (reverse-scored). The three confidence items were averaged into one measure (alpha = 0.92). All items were measured with 7point Likert scales. Demographic variables. At the end of the survey, participants included their gender, age, and years of work experience. Procedure Survey participants were commuters at a large Northeast train station. Participants completed the study in exchange for a chocolate bar. The surveys were completed anonymously. The survey took approximately five minutes to complete. Results Of the 178 participants who began the study, 167 completed all items in the survey. These 167 participants were used in the analysis. The mean age of participants was 41 (SD = 12) and 57% were female. On average, participants had 16 years of work experience. There were no significant effects of age, gender or work experience on the variables in this study. I examined the effects of reputation favorability and breadth on three variables trust, perceived trustworthiness and confidence using analysis of variance. I depict these results in Table 1. Reputation favorabililty had large significant effects on trust and perceived trustworthiness, but no significant effect on confidence in perceptions. Reputation breadth had a significant effect on trust and confidence. Finally, there was a significant interaction between reputation favorability and reputation breadth for trust. Means across all four conditions are depicted in Table 2. Trust (t (84) = 4.45, p < .01) was greater for positive reputations of high breadth than positive reputations of low breadth. However, trust was only marginally lower for negative reputations of high breadth than negative reputations of low breadth (t (78) = 1.71, p < .10). In summary, when reputations were positive, breadth had a significant effect on trust, perceived trustworthiness and confidence. When reputations were negative, breadth had a weaker effect on these variables. These findings support hypothesis 1, 6, and 9. Next, I tested whether perceived trustworthiness mediated the relationship between reputation favorableness and trust, and whether confidence mediated the relationship between reputation breadth and trust for positive reputations. As predicted, perceived trustworthiness fully mediated the favorableness-trust link (Sobel’s Z = 3.19, p < .01). When both variables were included in the regression, the coefficient for favorableness was no longer significant (β = 0.19, n.s. vs. β = 0.78, p < .01 in original equation) and the coefficient for perceived trustworthiness was significant (β = 0.84, p < .01). Confidence partially mediated the breadth-trust relationship for positive reputations (Z = 2.27, p < .05). When both reputation breadth and confidence were included in the regression, reputation breadth was less significant (β = 0.23, p < .05 vs. β = 0.54, p < .01) and confidence was significant (β = 0.33, p < .05). Discussion. Study 1 provides evidence that reputation breadth affects trust, and that it’s impact differs for positive and negative reputations. An actor who heard positive reports from many sources had higher trust in the colleague than an actor who heard a single positive reference. Breadth had only a marginal effect on trust for negative reputations. A truster hearing one negative view had considerably low levels of trust, and additional negative views resulted in only a small decline. In this study, actors only receive reputation information that has high consensus. When five people had impressions of the colleague, they were either all positive or all negative. When only one source is known, there’s no opportunity for differing opinions. In the next study, I examine the effect of reputation breadth on trust when consensus is high and when it is low, for both favorable and unfavorable reputations. Study 2 In the second study, participants made trust decisions based upon reputation information presented as numerical ratings from third parties. A similar paradigm has been used by Meyer (1981) and West and Broniarczyk (1998) in examining the effect of critic ratings on product preferences. Regarding social judgments, people rarely exchange numerical ratings about other people in casual conversation. However, there are many popular websites in which individuals rate the behavior of others on Likert scales. In some cases, such as eBay, individuals engage in social exchange with others but never meet face-to-face. The trust context is highly specific to fulfilling that particular order. On other websites, individuals can read ratings of individuals with whom they will interact in the future. For example, students use ratemyprofessor.com (and similar schoolspecific sites) when choosing courses and forming expectations about the professor’s behavior. Lawyers and clients use ratephillyjudges.com to ascertain the characteristics of the judge they will face in court proceedings. Understanding how individuals incorporate multiple others’ numerical ratings in their trust decisions will have increasing relevance as such websites proliferate and reputation information becomes more widely available in this form. In this study, I manipulated all three dimensions of reputation: favorableness, breadth and consensus. By using numerical ratings, I was able to manipulate the three dimensions orthogonally. As my dependent variable, I measured a trust-based decision: whether to take a monetary risk when the reward depends on the target’s promise fulfillment. I also measured perceived trustworthiness and confidence in perceptions of trustworthiness. Procedure 208 participants completed the study as one of multiple studies in which they were compensated $10 for the entire one-hour session. This was the first study in the session. Participants completed the study in one of 30 sessions over a two-week period. Lab participants were ostensibly paired with negotiation students to play a trust game (e.g., Croson & Buchan, 1999; Schweitzer, Hershey & Bradlow, 2006). In the game, participants were initially told that both they and their partner were given $8 and they needed to decide whether to pass their $8 to their partner. They read that by sending $8, the money would double and the counterpart received $16 (in addition to their original $8). The counterpart could return all $16 or return any amount between 0 and $16, in one dollar increments. If the participant kept the $8, the counterpart made no decisions and received no money. Participants were told that their counterparts would use their $8 it in a different individual decision-making game. All participants ultimately left with $8 bonus in addition to the $10 show-up fee. The negotiation students (N = 62) with whom participants were paired had volunteered to participate in the lab sessions for extra credit. Each student was involved in three or four sessions. They arrived at the lab at the same time as the participants. After everyone received general instructions, the negotiation students were sent to a lab next door to complete a different task, and the participants were paired with fictional counterparts. The negotiation students were aware that they were not paired with the participants, and they essentially acted as confederates in this experiment. They did, however, complete other studies for extra credit. To make the participants’ decision more explicitly one of trust in their counterpart (and not general risk preferences or expectations of reciprocity), participants read a message from their counterpart promising positive behavior. Participants were told that their partner could choose to send one of three messages indicating the range of money they planned to return, which could be as low as zero and as high as 200% the amount they had originally sent. The three message options were 0 99% of original money sent, 100% 149% of money sent, or 150%-200% of money sent. Participants were also told their counterpart could choose to send no message at all and that messages were not binding (i.e., the counterpart could return less than they indicated in the message). All participants read that the counterpart would return 150-200% of the money they had originally sent. Reputation Manipulation. Lab participants were told that their partners were from a negotiations class and that each partner had been evaluated by his or her 14 classmates in terms of their honesty. Participants were also told that the evaluations were made in the first half of the semester, and as a result not everyone had a perception of everyone else’s honesty. For each partner, the participant saw 14 ratings columns across the screen. When a given rater did not have a perception of the counterpart, participants saw a blank rating