Abstract
We examine how the timing of trust violations affects cooperation and solidarity, including trust and relational cohesion. Past studies that used repeated Prisoner’s Dilemmas suggest that trust violations are more harmful when they occur in early rather than later interactions. We argue that this effect of early trust violations depends on cultural and individual differences in generalized trust. A laboratory study from high- and low-trust cultures (the United States vs. Japan) supported our claim. First, early trust violations were more harmful than late trust violations, but only for Americans; the pattern reversed for Japanese. Second, these patterns were mediated by individual differences in generalized trust. Finally, generalized trust also moderated the effect of trust violations in the United States but not Japan. By demonstrating that generalized trust is not only lower but also less important in low-trust cultures, our research advances our understanding of how culture affects the development of solidarity in exchange relations.
In new exchange relations between strangers, early interactions pose a vexing dilemma. On the one hand, it is reasonable to hedge one’s trust and withhold cooperation from someone unfamiliar who may turn out to be untrustworthy. On the other hand, withholding cooperation might signal distrust that undermines the development of a cohesive relationship. Consistent with the latter possibility, research using repeated Prisoner’s Dilemmas demonstrates that “defections” (acts of mistrust or noncooperation that violate trust) are more harmful—inciting distrust and anger and reducing cooperation in subsequent interactions—the earlier they occur in the course of repeated interactions (Bottom et al. 2002; Komorita and Mechling 1967; Lount et al. 2008; Schilke, Reimann, and Cook 2013). In other words, the primacy effect of early trust violations seems to loom larger than the recency effect of the negative impressions people may form about each other or the relationship from trust violations in more recent (i.e., later) interactions. But are early trust violations always more harmful than later ones? In many relations, trust develops gradually, from initial acts of partial trust that grow over time toward commitment and mutual identification (Blau 1964; Kurzban et al. 2001; Lewicki and Bunker 1996). Some relations may even thrive after “getting off on the wrong foot,” namely, after overt acts of initial mistrust (Aronson and Linder 1965; Kuwabara and Sheldon 2012). The present research seeks to shed light on why some relations withstand mistrust better than others and to better understand the nature of commitment formation by examining how the timing of trust violations affects cooperation and feelings of social solidarity, namely, particularized trust and relational cohesion, after early versus late trust violations in two-person social exchange. Particularized trust concerns expectations of benign behavior based on inferences about a partner’s personal traits and intentions (Yamagishi and Yamagishi 1994). Cohesion refers to the sense of unity or “we-ness” based on the strength of relational bonds between actors. Together, particularized trust and cohesion are the interpersonal (person-to-person) and relational (person-to-relation) components of solidarity that promotes cooperation and commitment in exchange relations (Lawler and Yoon 1996; Molm, Collett, and Schaefer 2007).
We claim that the primacy effect of early trust violations on cooperation and solidarity depends on cultural and individual differences in generalized trust. In contrast to particularized trust people develop toward specific others over time, generalized trust refers to people’s propensity to trust others in general, including strangers (McKnight, Cummings, and Chervany 1998; Uslaner 2002). Generalized trust plays a critical role in establishing new relations, predisposing people to take “leaps of faith” toward new partners, and breaking down the dilemma of placing initial trust without the benefit of prior knowledge. The trade-off, of course, is greater vulnerability to mistrust by unknown others who may prove to be untrustworthy. Not only are people with high generalized trust (“high trustors”) more likely to have their trust broken than low trustors, they also place greater stakes on mutual cooperation in early interactions in the first place, which may influence how they experience early versus late trust violations. By comparison, we argue that people with low generalized trust (“low trustors”) are more likely to anticipate early trust violations and underplay their consequences because their behavioral strategy is building particularized trust over time instead of using generalized trust to jumpstart new relations with strangers (Simpson and McGrimmon 2007; Yamagishi, Cook, and Watabe 1998). Whereas high generalized trust exposes people to greater social risk, low generalized trust therefore shields people behaviorally and psychologically from the full blunt of early trust violations. If so, the primacy effect of early trust violations should be less likely to hold for people with low generalized trust.
The goal of the present research is to use the same basic design from past studies (Lount et al. 2008 in particular) to test our claim about how generalized trust affects the effect of early versus late trust violations on cooperation and solidarity while replicating past findings. In the laboratory study, participants played multiple rounds of the Prisoner’s Dilemma against a simulated partner programmed to defect in early or later rounds. To examine the role of generalized trust, we measured individual differences in generalized trust and conducted the study in the United States and Japan, two cultures that vary in generalized trust (Yamagishi et al. 1998).
While social exchange theorists have spilt much ink on how people form committed relations to safeguard against potential risk of malfeasance (e.g., Kollock 1994; Molm, Takahashi, and Peterson 2000; Simpson and McGrimmon 2007), how people react to actual trust violations has received far less attention. 1 This presents a critical gap in our understanding of relational commitment; for instance, why do some relations become resilient to occasional mistrust while others remain fragile, even as they show equally high levels of cooperative interactions (Molm, Schaefer, and Collett 2009)? Here, we consider how generalized trust might affect the nature of commitment. Does high generalized trust soften the blow of a defection or sharpen its sting (when it occurs early vs. late in a relationship)?
In the following sections, we review past research that shows that early trust violations are more harmful. We then discuss why the effect of the timing of violations might depend on individual and cultural differences in generalized trust. After deriving hypotheses about exactly how generalized trust affects (i.e., mediates and moderates) the effects of early versus late violations, we describe our laboratory study and results.
Theory and Hypotheses
The Effects of Early Versus Late Trust Violations
The substantive context of our research is two-person relations in which people must engage in a fixed number of repeated mixed-motive interactions without exit options or formal means to enforce cooperation. Such situations are pervasive in organizational settings where people are often assigned to projects or transactions with new partners. They also describe many situations in which exiting is impractical, such as dealing with a new neighbor or in-laws. We use the Prisoner’s Dilemma (PD) to model the persistent tension people often face in such relations: whether to pursue one’s own self-interest (i.e., work on personal projects, blast music to wake up in the morning) or to take on shared duties and responsibilities (i.e., sacrifice one’s personal time on group projects, keep noise down). In PD, each interaction consists of two players making simultaneous decisions to “cooperate” or “defect” (Figure 1). The payoff to each player depends on their joint decisions, such that mutual cooperation yields the greatest collective outcome (3 points to each player) while mutual defection leads to the worst collective outcome (2 to each player). The dilemma for each player is deciding whether to cooperate, knowing that the other player might be tempted to defect, committing a trust violation that yields the maximum individual payoff to the defector (4 points) but the worst individual payoff to the cooperator (1 point).

Prisoner’s Dilemma
While trust violations are harmful to the victim whether they occur early or late, it is unclear which should prevail or why—holding the payoffs constant across time—the timing of trust violations might matter at all. Lount et al. (2008) suggest two possibilities. On the one hand, primacy effect suggests that first impressions from early events have disproportionate influence on how we perceive subsequent events. From this view, early trust violations should cast a longer and darker shadow of distrust over the subsequent interactions compared to later trust violations. On the other hand, there are reasons to expect trust violations to be more harmful the later they occur. Contrast effect suggests that trust violations that occur after a period of positive interactions may stand out more, inflicting more harm than early trust violations. Similarly, recency effect suggests that people weigh recent events (e.g., trust violations occurring in later or more recent interactions) more heavily.
To test between these possibilities, Lount et al. (2008) asked participants to play a series of repeated PDs against simulated partners programmed to defect in early or later rounds. They found that participants were more likely to withhold cooperation in the final rounds and report greater distrust in the postexperimental questionnaire when trust violations occurred in earlier rounds. They also found convergent patterns in subjective measures of trust, a dimension of relational solidarity. Komorita and Mechling (1967) found similar results, showing that participants were slower to return to mutual cooperation when trust violations occurred after 4 rounds rather than 10 rounds of mutual cooperation from the partner. Schilke et al. (2013) replicate these patterns and provide neural evidence that early versus late trust violations activate different brain systems.
But are early trust violations always more harmful? Our claim is that these patterns depend on generalized trust. To wit, all three past studies on the timing of trust violations recruited participants from the United States, a high-trust culture. There are reasons to believe, however, that the primacy effect of early trust violations may weaken or even reverse in low-trust cultures.
The Role of Generalized Trust
The starting point for our argument is that the timing of trust violations affects different bases of relationships over time. As exchange relations typically evolve from generalized trust toward mutual strangers to particularized trust and commitment (Weber, Malhotra, and Murnighan 2005), trust violations are more likely to affect generalized trust than particularized trust in early interactions, whereas they are increasingly perceived as threats to the relationship itself rather than to the welfare of each individual party the later they occur (Lewicki and Bunker 1996). This points to a more sociorelational reason why the timing of trust violations might matter: in the prospect of building a relationship through repeated exchanges, when trust violations occur therefore matters not simply because they are perceived as recent versus past incidents but because they threaten different types of trust and in different stages of relationship formation.
This effect of early trust violations for generalized trust, we argue, is more pronounced for high trustors who place high stakes in the possibility of immediate cooperation. Because their success, welfare, and worldview depend on taking a leap of faith in the benevolence of strangers, high trustors should react more quickly and strongly than low trustors to what actually happens in early interactions. In comparison, the outcomes of early interactions are less critical to low trustors who focus on building commitment based on particularized trust rather than generalized trust, precisely because they view early interactions to be risky and unpredictable (Simpson and McGrimmon 2007; Yamagishi et al. 1998). Their goal is to (quickly) move past the initial stage of exchange relations to ensure commitment. Thus, low trustors may be more likely to tolerate or dismiss trust violations that occur in early interactions as an inevitable price of building new relations or a matter-of-fact sign that they have yet to establish mutual commitment. However, as relationships evolve beyond very early interactions, and generalized trust is gradually replaced by particularized trust based on cumulative knowledge and mutual identification, late trust violations are more likely to affect low trustors who are more eager to form and maintain commitment than high trustors. In other words, early and late trust violations may have different, and perhaps opposite, effects on high and low trustors.
Generalized Trust in the United States and Japan
Our goal is to examine the previously described logic by comparing Americans and Japanese as relatively high and low trustors. Surveys (Branzei, Vertinsky, and Camp 2007; Yamagishi and Yamagishi 1994), field studies (Hagen and Choe 1998), and laboratory studies (Buchan, Croson, and Dawes 2002; Cook et al. 2005) have consistently identified the United States as a high-trust society compared to Japan. We thus predict early versus late trust violations to have different effects on cooperation and solidarity for Americans and Japanese.
Hypothesis 1: For Americans, early trust violations are more harmful than late trust violations to cooperation and solidarity.
Hypothesis 2: For Japanese, early trust violations are less harmful than late trust violations to cooperation and solidarity.
We also reasoned that if Americans and Japanese respond differently to early trust violations because of generalized trust, these cross-cultural patterns may be attributed to different levels of generalized trust between the United States and Japan. Figure 2 illustrates our argument: culture influences the level of generalized trust, which affects how people react to early versus late defections, influencing the development of cooperation and solidarity. In other words, the effects of culture are mediated by generalized trust before interacting with the timing of trust violations, implying moderated mediation. This mediation is not causal in an experimental or proximate sense since assignment to culture is not random but distally causal in the sense that culture leads to different levels of generalized trust.

A Moderated-Mediation Model of the Effects of Culture, the Timing of Trust Violations, and Generalized Trust Affect Cooperation and Solidarity
Hypothesis 3: Generalized trust mediates the effect of culture on cooperation and solidarity; this mediation is moderated by the timing of trust violations.
Finally, we predict generalized trust to matter less in Japan than the United States. According to a socioecological view of trust (Yamagishi, Hashimoto, and Schug 2008; Yamagishi and Walker 2003), generalized trust is not simply a dispositional trait that varies across individuals but a “default strategy” people develop to explore new exchange opportunities outside of familiar relations. In a heterogeneous society with high social mobility where long-term relations are more ephemeral, generalized trust plays an important role in facilitating social exchange with unfamiliar partners rather than committing to particular relations (Macy and Sato 2002; Yamagishi et al. 1998). By comparison, in societies characterized by collectivism, stable social networks, and high levels of institutional order such as Japan, generalized trust is less critical for initiating new relations. Although building interpersonal trust quickly may be important in Japan as well, dense social ties and institutional order allow Japanese to rely more on referrals from mutual contacts or formal enforcement instead of their own faith or judgment in navigating new exchange opportunities (Yamagishi and Yamagishi 1994; Yuki et al. 2005). For them, a more critical concern than establishing new relations quickly is whether they can build and sustain committed relations over time (Kuwabara et al. 2008; Schug, Yuki, and Maddux 2010; Yamagishi and Yamagishi 1994). Somewhat paradoxically, Japanese can also be less cautious than Americans because strong norms, stable social networks, and low crime rates provide enough assurance and security to let their guard down in many situations (Miller and Mitamura 2003). Therefore, if Japanese offer cooperation to a stranger in early interactions, it is likely on the basis of assurance (Yamagishi and Yamagishi 1994) or the structural incentives of repeated interactions or institutional enforcement rather than generalized trust in the benevolence of an unknown other, compared to Americans.
In short, we argue that the timing of trust violations may have different consequences in the United States and Japan because Americans and Japanese differ not only in how trusting they are in general but also in whether or how much they actually rely on generalized trust to initiate or manage social exchange. Americans might respond to early versus late trust violations differently depending on how dispositionally trusting they are. By comparison, the idea that the Japanese cultural script focuses on the assurance of commitment rather than generalized trust suggests that Japanese would respond to early versus late trust violations regardless of their generalized trust. If so, the interaction of generalized trust and the timing of trust violations depicted in Figure 2 should be stronger in the United States than Japan.
Hypothesis 4: Generalized trust moderates the effects of early versus late trust violations on cooperation and solidarity in the United States more than in Japan.
Altogether, our hypotheses, if supported, would suggest that the effects of culture predicted by Hypotheses 1 and 2 can be explained by cultural differences in generalized trust between Americans and Japanese (Hypothesis 3). Moreover, this cultural difference is not simply a matter of higher generalized trust among Americans; generalized trust also matters more for Americans than for Japanese (Hypothesis 4). A novel implication is that cultural differences in generalized trust are not simply a matter of individual differences merely aggregated within each culture. Rather, some cultures might also differ in whether individual differences in generalized trust matter at all. In other words, generalized trust may differ both quantitatively and qualitatively across cultures.
Method
Design and Materials
Our interest is in how the effects of early versus late trust violations on cooperation and feelings of solidarity may be conditioned by generalized trust. The timing of trust violations was experimentally manipulated in a 2 (timing of trust violations: early vs. late) × 2 (culture: the U.S. vs. Japan) between-subjects design using PDs against a simulated partner. To evaluate cultural differences in generalized trust, we recruited 93 university students from the United States (44.1 percent men, age = 22.26 ± 4.42, all U.S. born) and 112 from Japan (50.2 percent men, age = 20.40 ± 2.60, all native Japanese) via mailing lists for cash payment based on performance. To measure generalized trust, participants were asked to complete at least 24 hours before the actual experiment a questionnaire containing the six-item generalized trust scale (Yamagishi and Yamagishi 1994; “Most people are basically honest [trustworthy; good and kind; trustful of others],” “I am trustful,” “Most people will respond in kind when they are trusted by others,” all 7-point Likert scales, 1 = completely disagree, 7 = completely agree), which we used to perform mediation and moderation analyses for Hypotheses 3 and 4. The scale achieved α = .78 in the United States and .84 in Japan.
The experiment took place in a laboratory at an elite metropolitan university in each country using identical procedures and payment schedule. The experimental materials were translated and reverse-translated by three bilingual researchers. The Japanese sample contained more men and was slightly younger, but controlling for these differences did not change our results. Although we made great efforts to run the study in the United States and Japan under equivalent conditions, we do not use culture as a main predictor in our hypothesis tests because as the hypotheses imply, our interest is not in direct comparisons between countries but how the effects of early versus late trust violations differ across countries.
Procedures
Each experimental session consisted of 4 to 10 participants. Upon entering the laboratory, they were seated separately in a cubicle with a computer terminal. After signing the consent form and reading detailed instructions to ensure complete understanding of the experimental protocols, participants were asked to complete “up to 30 rounds” of PDs with an anonymous partner, programmed (unbeknownst to participants) to cooperate in all rounds except 1 and 2 (early trust violations) or 10 and 11 (late trust violations). The exchange rounds actually ended after 22 rounds. 2 At the beginning of round 18, the computer screen announced that there were 5 more rounds; this served to create an “end-game” effect and heighten the salience of trust (Lount et al. 2008). After the last round, participants were asked to complete a postexperimental questionnaire on solidarity. Finally, participants were debriefed and paid based on performance (total points earned).
Following past research (e.g., Komorita and Mechling 1967; Lount et al. 2008), we operationalized trust violations by programming the partner to defect twice in a row because a single isolated trust violation may be interpreted as accidental. We then excluded participants who defected in both rounds 1 and 2 or 10 and 11 (N = 20 in the U.S., 36 in Japan) in order to focus on people who actually suffered a trust violation at least once. In additional analyses for robustness tests, we controlled for whether participants experienced a trust violation in each round, in the first round only, or the second only; they had no effects and are not reported.
The exit survey included two questions to measure suspicion: “Did you suspect that your partner was not an actual human” and “Did your suspicion affect how you made your decisions?” (1 = no, 2 = somewhat, 3 = very much). Because our hypotheses presume social interactions between actual humans, we removed participants who indicated 3 on both questions (N = 7 in the U.S. and 9 in Japan). The final sample size is 32 in the early condition and 34 in the late condition in the United States and 31 and 36 in Japan. In the final sample, we found no difference in age or gender across conditions or countries.
Dependent Variables
Following Lount et al. (2008), our behavioral dependent measure is average cooperation in the last five rounds after the end-game announcement. To measure feelings of solidarity, we administered a postexperimental questionnaire on particularized trust and relational cohesion; all items were 7-point Likert-type scales. Three items on particularized trust asked how much they trust their partner, how trustworthy they perceived their partners to be, and how honest they perceived their partners to be (α > .95 in both countries). Four items on relational cohesion, from Lawler and Yoon (1996, 1998) asked, “How close or distant do you feel toward your partner?” “How cohesive or uncohesive do you feel toward your partner?” “Would you describe the exchange relationship as team- or self-oriented?” and “How much do you agree that your relationship grew stronger over time?” (α > .88 in both countries).
Results
Table 1 reports the means and standard deviations of our dependent variables: average cooperation in the final five rounds and self-reported feelings of particularized trust and cohesion as measures of solidarity. To test Hypotheses 1 and 2, we submitted our data to a 2 (timing of trust violations: early vs. late) × 2 (culture: the U.S. vs. Japan) ANOVA. Gender, age, and race did not affect our main results in any significant or consistent ways and are not discussed. Results are in Table 2. Consistent with our main claim, we found an interaction effect of the timing of trust violations and culture, while neither main effect was significant, suggesting that the timing of trust violations had different effects in the United States versus Japan.
Average Cooperation in the Last Five Rounds and Feelings of Particularized Trust and Cohesion in the Exit Survey
Last Five Rounds and Feelings of Particularized Trust
p < .05. **p < .01, two-tailed tests.
Consistent with Hypothesis 1 and prior studies (Lount et al. 2008; Schilke et al. 2013), early trust violations resulted in lower cooperation in the final five rounds (M = .35, SD = .35) than late trust violations (M = .56, SD = .34), t(64) = 2.44, p = .018, d = .60, in the United States. However, consistent with Hypothesis 2, participants in Japan cooperated more in the final five rounds after early trust violations (M = .60, SD = .38) than late trust violations (M = .37, SD = .30), t(65) = 2.84, p = .006, d = .70.
The postexperimental questionnaire provides additional support for the idea that the timing of trust violations has different effects in the United States and Japan. A two-way ANOVA found a significant effect of timing of trust violations × culture on particularized trust (Table 2). Although the difference between early (M = 5.07, SD = 1.16) and late defection (M = 4.75, SD = .95) did not reach significance in the United States, t(64) = .71, p = .21, d = .31, participants in Japan reported greater trust toward their partners after early (M = 5.86, SD = 1.02) than late defection (M = 4.46, SD = 1.58), t(65) = 4.21, p = .001, d = 1.03. Similarly, a two-way ANOVA found a significant effect of timing of trust violations × culture on cohesion (Table 2). In the United States, the difference between early (M = 4.90, SD = 1.41) versus late trust violations (M = 4.53, SD = 1.30) was not significant, t(64) = 1.11, p = .27, d = .27, but in Japan, participants reported greater cohesion after early trust violations (M = 5.57, SD = 1.23) than late trust violations (M = 4.06, SD = 1.49), t(65) = 4.49, p < .001, d = 1.10. These results provide clear support for Hypothesis 2 for Japan and partial support for Hypothesis 1; we discuss this issue in Discussion.
Mediation by Generalized Trust
So far, we have tested the interaction effect of culture and the timing of trust violations on cooperation and solidarity: culture × timing of trust violations → DVs. Hypothesis 3 predicts that the effect of culture is mediated by generalized trust, but this mediation is moderated by the timing of trust violations: culture → generalized trust × timing of trust violations → DVs. The logic is that how Americans and Japanese react to early versus late trust violations depends on their generalized trust. To test this idea, we examined their responses to the generalized trust scale, collected prior to the behavioral experiment. Consistent with past research (e.g., Yamagishi and Yamagishi 1994), Japanese participants indicated lower generalized trust (M = 4.18, SD = .94) than Americans (M = 4.63, SD = .75), t(131) = 3.21, p = .002, d = .56.
Following Preacher, Rucker, and Hayes (2007), we used structural equation modeling with bootstrapping rather than the Baron-Kenny procedure to test the moderated mediation; the main idea is to estimate direct and indirect paths simultaneously rather than in discrete steps, which can yield unreliable results. Bootstrapping is used to estimate 95 percent bias-corrected confidence intervals (CI) around the estimated indirect effects; mediation is said to occur if zero falls outside of the confidence interval. The mediator, generalized trust, was centered to facilitate interpretation and reduce collinearity. 3 Following this approach, the full structural equation model (Table 3) found a significant effect of early trust violations × generalized trust, b = –.03, SE = .01, p = .004, and a nonsignificant effect of Japan, b = .01, SE = .06, p = .85, suggesting indirect-only mediation (Zhao, Lynch, and Chen 2010). Our bootstrap analysis (5,000 iterations) found a positive effect of the indirect path (Japan → generalized trust × early trust violations → cooperation = .058, 95 percent CI = .02 to .14). Similar results obtained for particularized trust (indirect path = .20, 95 percent CI = .04 to .46) and cohesion (indirect path = .23, 95 percent CI = .01 to .57). Overall, these results suggest that how high versus low trustors react to early versus late trust violations can account for how Americans and Japanese react to early versus late trust violations, supporting Hypothesis 3.
Structural Equation Modeling Testing Moderated Mediation by Generalized Trust
Note: N = 133. Standard errors in parentheses.
p < .05. **p < .01, two-tailed tests.
Moderation by Generalized Trust
Hypothesis 4 predicts that not only is generalized trust lower in Japan, but it matters less in Japanese exchange relations. To test this claim, we first submitted the generalized trust scores (centered within each culture) to ordinary least squares (OLS) regression predicting cooperation in the last five rounds as a function of early trust violations. As shown in Table 4, early trust violations × generalized trust is significant and negative, b = –.05, SE = .02, p = .006, in the United States, suggesting that Americans were less likely to cooperate after early trust violations the higher they scored in generalized trust. Given the dependent variable in percentage, this suggests that one unit increase in generalized trust (on a 7-point scale) reduces cooperation in the final rounds by 5 percent after early trust violations. In contrast, we find no effect of generalized trust or its interaction for Japanese, both ps > .74. We conducted a post-estimation linear contrast since a significant coefficient may not differ significantly from a nonsignificant coefficient. We found that the effect of early trust violations × generalized trust in the United States is significantly different from its effect in Japan, χ2 = 4.52, p = .033. Finally, we regressed particularized trust and cohesion as dependent variables and found similar patterns (Table 4). These results support Hypothesis 4 and suggest that generalized trust played a significant role in American exchange relations but not Japan. 4
Coefficients from Ordinary Least Squares Regression Showing Effects of Early Trust Violations and Generalized Trust on Average Cooperation in the Last Five Rounds and Feelings of Particularized Trust and Cohesion, by Culture
Note: Standard errors in parentheses. All two-tailed tests.
p < .05. **p < .01.
Suspicious Participants
Our aforementioned results preclude seven American and nine Japanese participants who indicated maximum suspicion on two control questions in the postexperimental survey. Among these participants, two did not complete the exit questionnaire on particularized trust and cohesion, and three did not complete the generalized trust questionnaire. However, a rank-sum test found no difference between the United States and Japan in the proportion of very suspicious participants, p = .58, and these participants did not differ in generalized trust from those retained in the sample in either country, ps > .32. Including very suspicious participants in our analyses found substantively similar results. For Hypotheses 1 and 2, early trust violations × Japan is significant for cooperation, F(1, 145) = 9.39, p = .003; particularized trust, F(1, 143) = 7.66, p = .01; and cohesion, F(1, 143) = 5.65, p = .02. For Hypothesis 3, generalized trust mediates the effect of culture on cooperation (.05, 95 percent CI = .002 to .13), particularized trust (.18, 95 percent CI = .02 to .38), and cohesion (.22, 95 percent CI = .04 to 47). Finally, for Hypothesis 4, generalized trust moderated the effect of early trust violations on cooperation (b = –.06, SE = .02, p = .002), particularized trust (b = –.14, SE = .06, p = .03), and cohesion (b = –.22, SE = .07, p = .002) in the United States only, not Japan. Overall, the consistency of these results suggests that our main results are largely unbiased by suspicious participants.
Conclusion and Discussion
The present research extends past work on the timing of trust violations (Komorita and Mechling 1967; Lount et al. 2008; Schilke et al. 2013) by identifying a key moderator of the effect of early trust violations: how people react to early versus late trust violations depends on how much they rely on generalized trust to initiate exchange relations. For people from the United States (a high-trust culture), early trust violations were more harmful than late trust violations. However, for people from Japan (a low-trust culture), early trust violations were less harmful than late trust violations. These results challenge the implication that full cooperation in early interactions is critical for the development of committed relations. This may be true for high trustors who are relatively quick to judge and trust strangers. By comparison, low trustors seem to tolerate some hedging or mistrust from each other until they establish a certain rhythm of exchange. Once committed, however, they stand to pay a greater price for withdrawing cooperation.
By linking these ideas about generalized trust to culture, our results deepen our understanding of cultural differences in trust and commitment formation. While the idea that commitment formation plays a more significant role than generalized trust in Japan compared to the United States has been well documented (Cook et al. 2005; Kuwabara et al. 2008; Yamagishi et al. 1998), our research suggests that this commitment does not mean simply tolerating trust violations for the sake of maintaining a relationship. In fact, Japanese were, with respect to cooperation in the last rounds as well as self-reported particularized trust and cohesion, less tolerant of late than early trust violations. To the extent that late trust violations are more likely to threaten the stability of mutual commitment, this may seem contrary to the very idea of commitment. Past research suggests, however, that in societies characterized by high relational stability such as Japan, commitment in exchange relations may be valued not for its own sake but only to the extent that mutual monitoring and sanctioning in repeated interactions provide “assurance” by creating structural incentives against malfeasance; once such incentives are removed or undermined, Japanese tend be less likely to remain committed to their relations or groups (Yamagishi 1988). Thus, for Japanese, commitment as a safeguard against social risk may be more fragile than resilient, crumbling as soon as it begins to show signs of wear. A novel implication is that people who typically value commitment—low trustors—may react more strongly than high trustors to late trust violations that disrupt stable exchange relations. By comparison, it is Americans, exposed to high relational mobility and social diversity, who may be more motivated to invest in stable relationships by raising the stakes of exchange and making themselves vulnerable to mistrust (Cook et al. 2005; Schug et al. 2010).
We also found that individual differences in generalized trust moderated the effects of early trust violations, but only among Americans, supporting the idea that generalized trust matters more for Americans than Japanese. This speaks to an important question in cultural psychology: are cultural differences reducible to individual differences? Na et al. (2010) argue that they are not. Individual differences can capture certain aspects of culture, but cultures also differ from each other in how they organize, interpret, and enact different individual traits and dispositions (also see Leung and Cohen 2011). By analogy of gender, men and women may differ not only in certain individual traits (e.g., assertiveness) on average, but also in how they express it (Babcock and Laschever 2003). Similarly, our research suggests that people from low-trust cultures are not simply lower in generalized trust; they also seem to rely less on generalized trust.
It is worth noting that we found effects of early versus late trust violations on cooperation in both countries and on self-reported solidarity in Japan but not the United States. It is unclear if this is an anomaly in our study or a real pattern for Americans since past studies did not measure solidarity, only cooperation, and our argument does not specify why early trust violations are more harmful than late trust violations for Americans, only why such patterns may be culture specific and thus differ in Japan. One possibility is that compared to round-by-round cooperation, solidarity is more retrospective and global in scope; relational bonds develop as people look back at a certain course of interactions or recent events, whereas cooperation is much more prospective or projective, anticipating immediate future cooperation in each round. This means generalized trust may not affect solidarity to the same extent that it affects cooperation because feelings of solidarity do not directly depend on future cooperation. This also suggests that solidarity may be less sensitive to the exact timing of trust violations than to overall cooperation or recent events. In other words, the timing of trust violations did not affect solidarity for Americans, perhaps because late trust violations are more harmful to solidarity in both cultures, but its effect was tempered by the reverse pattern for cooperation in the United States. Although speculative, these ideas point to important differences in the nature of cooperation versus solidarity and calls for further research.
In order to present unambiguous cases of trust violations, we used PDs with binary decisions, allowing only cooperation or defection and precluding partial trust. Although this was crucial for isolating the timing of trust violations from other contextual factors or endogenous processes that may confound our results, trust violations can occur under a variety of conditions. For instance, in many situations, people can hedge, cooperating little by little, or avoid interacting altogether with people they distrust rather than simply defecting against them. More research is needed to understand whether our findings might generalize to other situations or other forms of trust violations.
More generally, we hope to call attention to the need for more research on how the precise dynamics of exchange affects the development of commitment and solidarity. Although social exchange theorists have paid careful attention to endogenous processes that transform repeated exchanges into an exchange relationship imbued with feelings of trust and cohesion (Kuwabara 2011; Lawler 2001; Molm, Collett, et al. 2007; Schaefer and Kornienko 2009), research so far has focused by and large on aggregate or time-invariant properties of repeated exchanges, such as the structural form of exchange (Molm, Collett, et al. 2007) or overall frequency of interactions (Lawler, Thye, and Yoon 2000). However, how exchange patterns unfold or evolve over time is also a salient feature of repeated exchanges that can influence how actors make sense of their exchange partners or relations (Kuwabara and Sheldon 2012; Molm, Schaefer, et al. 2007). Our research shows that the timing of trust violations—whether they occur in early or late interactions—can also affect the development of relational solidarity.
Footnotes
Acknowledgements
Kuwabara, Vogt, Watabe, and Komiya designed and performed data collection. Kuwabara and Vogt analyzed data and wrote the paper. Kuwabara and Vogt contributed equally to this work. We thank Joel Brockner and anonymous reviewers for meticulous comments on earlier drafts.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: We thank the Swiss National Science Foundation Grant #100014-130127(1) and Columbia University ISERP Grant #467859 for research.
1
A notable exception is Molm, Collett, and Schaefer (2006), which examined how actors reacted to unfair exchange outcomes in reciprocal versus negotiated exchange relations.
2
We use early and late in relative terms. This is consistent with the finding that the primacy effect of early trust violations is monotonic: Lount et al. (2008) found trust violations to be increasingly less harmful the later they occurred (in rounds 1–2 vs. 6–7 vs. 11–12), and
found trust violations in round 4 to be more harmful than trust violations in round 10. It is possible that given much longer interactions than 22 rounds, late trust violations may eventually outweigh the effect of early trust violations, even for high trustors. Our goal, however, is not to define this tipping point but to show that the relative effects of early versus late trust violations vary across culture and individuals. Thus, in our case, early trust violations were programmed to occur immediately (in rounds 1 and 2) when generalized trust is expected to play a particularly crucial role in building new relations, thus providing the most direct test for the primacy effect of early trust violations.
3
The measure was centered at the grand mean of the United States and Japanese samples, combined, since centering within each culture partials out between-culture differences.
4
An alternative explanation is that the Japanese are more homogeneous in generalized trust or devoid of high trustors. This explanation is not convincing to us since the standard deviation in generalized trust was actually higher in Japan than the United States, .94 versus .75.
