Abstract
Recent work shows that both reward and punishment systems increase short-term cooperation in social dilemmas. Yet, a growing body of research finds that punishment systems generate a range of negative side effects, including an undermining of trust in fellow group members’ cooperative intentions. The present work asks whether reward systems can generate the same positive effects as punishment systems (increased cooperation) without the negative side effects (decreased interpersonal trust) or whether reward systems also lead to detrimental effects on trust. In two experiments we find that once removed, reward systems, like punishment systems, reduced trust to levels below a control group who never experienced sanctions. This research highlights the detrimental effects of punishment and reward systems on intragroup trust and thus shows that while reward systems can generate the same positive effects as punishment systems, they also generate the same negative side effects.
Introduction
At least since Hobbes ([1651] 1968), social scientists have suggested that sanctioning systems are an effective means to solve the problem of social order (e.g., Fehr and Gachter 2002; Gürerk, Irlenbusch, and Rockenbach 2006; Herrmann, Thöni, and Gächter 2008; Yamagishi 1986; see Shinada and Yamagishi 2008 for a review). In general, this research shows that punishment systems enhance short-term cooperation—when present, negative sanctions motivate individuals to sacrifice for the greater good at a personal cost. But a growing body of work highlights a dark side to punishment. A number of studies indicate that while increasing cooperation in the short run, punishment systems may produce schisms among group members: they create cycles of recrimination (Nikiforakis 2008), foster counterpunishment (Denant-Boèmont, Masclet, and Noussair 2007), and most importantly for our purposes, they undermine trust (Mulder et al. 2006). Trust refers to expectations regarding others’ benign intent and likelihood of cooperating (Yamagishi 2001). Punishment systems undermine trust because they lead group members to believe that others would not otherwise act cooperatively.
The negative effects of punishment systems on trust may be crucial given the strong link between trust and other positive group-level outcomes. Trust is associated with feelings of group solidarity (Molm, Collett, and Schaefer 2007), affective commitment (Lawler, Thye, and Yoon 2009), and social identification (Tanis and Postmes 2005), among other positive factors (see also Poteete, Janssen, and Ostrom 2010). The erosion of trust and its correlates may precipitate cracks and fissures in the group’s structure, making it weak and unstable (Markovsky and Lawler 1994). Accordingly, although punishments may yield short-term increases in cooperation, trust and positive regard for the group may be necessary for cooperation over the long term: for generating “more widespread, more enduring, and lower-cost cooperation” (Lawler et al. 2009:176). In short, rather than a boon to collective action efforts, recent studies indicate that punishment systems may be counterproductive or even detrimental to those efforts.
Meanwhile, a separate stream of research has investigated the impact of reward systems for cooperation and collective action (Andreoni, Harbaugh, and Versterlund 2003; Komorita and Barth 1985; Rand et al. 2009; Sefton, Shupp, and Walker 2007; Vyrastekova and van Soest 2008). A recent meta-analysis of this literature found that rewards yielded short-term cooperation levels comparable to punishments (Balliet, Mulder, and Van Lange 2011). This work suggests that carrots may be viable alternatives to sticks, at least for increasing short-term cooperation.
Yet, an important question remains: Do rewards systems, like punishment systems, undermine intragroup trust and therefore potentially hamper sustained cooperation? 1 Some work suggests that the use of rewards versus punishments tends to evoke less negative emotions, retaliation, and psychological reactance (Tjosvold and Tjosvold 1995; Wit and Wilke 1990). Research also indicates that those who reward others may be viewed more favorably than punishers, indicating that reward and not punishment systems are more likely to produce sustained cooperation (Kiyonari and Barclay 2008). But no research has addressed whether punishment and reward systems differ in their effects on trust. The present research is the first to investigate punishment and reward systems with regard to their trust-undermining effects. We also address whether these trust-undermining effects manifest in reduced cooperation.
To study the impact of sanctioning systems on trust and cooperation, we use a social dilemmas paradigm in two laboratory experiments. Social dilemmas are situations where individual and collective interests conflict. At their core, social dilemmas represent the problem of social order, and successful resolutions require cooperation (see Kollock 1998). In the first experiment, we address the impact of punishment and reward systems on interpersonal trust and cooperation by observing their effects when present and once they are removed. The second experiment is a replication of Study 1 using a different sample. Moreover, in the second study the punishment and reward systems were weaker than in Study 1, which allowed us to assess whether the impact on trust and cooperation occurs even when sanctions are less consequential.
Theory and Hypotheses
In the current work we study the undermining effect of sanctions on trust mainly by means of the “removing the sanction method” (see Mulder et al. 2006). The idea behind this method is that if a sanctioning system indeed undermines trust that others have cooperative intentions, this should become evident in the trust levels after the removal of that sanction. When a sanctioning system is present, trust may either be based on the expectation that others have cooperative intentions or on the perception that their cooperation is motivated by the sanction. When sanctions are absent, system-based trust is irrelevant; once removed, the only basis for trust is perceptions of others’ benign intent. Therefore, with the discontinuance of a sanctioning system the possible side effects of incentives are most evident: if trust in others is lower when a previously present sanctioning system has been removed compared to a situation in which there has not been a sanctioning system at all, one can conclude that the sanctioning system has undermined trust. In the following, when we refer to the impact of sanctions on trust and cooperation, we are specifically referring to the effects that emerge after the removal of the sanctioning system.
Apart from giving us the analytical leverage to disentangle these very different bases of trust, studying the effects of sanction removal also allows us to draw parallels to real-world sanctioning systems. That is, monitoring systems are often imperfect (Hechter 1987); group members sometimes have opportunities to act outside the eye of a sanctioning regime. Thus, studying what happens if a sanction is removed is relevant for real-world situations where formal sanctions are in place but group members’ behavior can go undetected.
Punishments, Rewards, and Attributions About Others
The effects of punishment and reward systems on trust and cooperation may stem from the information these systems signal to individuals about other group members. One approach suggests a main effect of external inducements. Sanctioning systems (whether positive or negative) generate adverse attributions about others and once removed undermine trust and cooperative behavior. A second argument suggests that the valence of inducements (i.e., as punishments or rewards) differentially colors the views individuals have about other group members. Rewards promote positive sentiments, but punishments engender negative feelings. Once removed, rewards may not be subject to the same deleterious effects as punishments.
The first argument contends that sanctions (both positive and negative) produce negative attributions about others. When present, sanctioning systems may imply that group members are untrustworthy and may lead group members to assume that others are externally but not internally motivated to cooperate. Individuals may wonder why a sanctioning system is necessary unless group members require external motivation (see Galbiati, Schlag, and van der Weele 2013). That is, group members may reason that others cooperate in order to receive a reward or avoid punishment. Such motivation would indicate that others are self-interested rather than group-oriented, which may call into question others’ trustworthiness. If sanctions are removed or individuals interact outside the sanctioning system, these expectations may reduce trust in others’ willingness to cooperate. In line with this argument, considerable research spanning several decades has shown that positive sanctions lead people to assume that others are externally (rather than internally) motivated (see Deci, Koestner, and Ryan 1999). Additionally, Mulder et al. (2006) found that in the presence of a negative sanction, people tended to attribute others’ cooperation to the punishment system rather than an intrinsic motivation to cooperate. As such, the presence of reward systems, just like the presence of punishment systems, may lead individuals to believe that others are externally motivated, and these perceptions may hinder trust and cooperation. This “attributional” explanation thus suggests that both reward and punishment systems have undermining effects once removed.
A second approach suggests that while both positive and negative sanctioning systems produce attributions about others, the valence of the sanction may alter the types of attributions actors make about others. In the case of punishment, people might assume that cooperation is problematic and a punishment system is necessary to curtail free-riding (Cialdini 1996; Mulder et al. 2005). More specifically, punishment systems may lead people to reason that others would be unwilling to cooperate and that pursuing self-interest is normative absent negative sanctions (Galbiati et al. 2013). As such, once removed, these systems may produce negative effects on trust and cooperation. This is because punishments are often associated with retribution and just desserts (Carlsmith, Darley, and Robinson 2002; Darley and Pittman 2003). Punishments may therefore evoke negative sentiments or expectations even without evidence of others’ wrongdoing and may communicate that there are “bad apples” in the group whose behavior must be curtailed via negative sanctions (Kerr et al. 2009). Empirical work is consistent with these arguments (see De Dreu, Giebels, and Van de Vliert 1998; Fehr and List 2004).
Following this “affective” explanation, reward systems may exert different effects on trust and cooperation compared to punishment systems. Whereas punishments can highlight group members’ egoism, rewards are typically given to those who sacrifice for the group. By highlighting more cooperative behaviors, reward systems might make salient the goodness in people. If so, they should enhance perceptions that others are motivated to cooperate and consequently, the feeling that they are trustworthy. Indeed, previous work suggests that at least in the short term, reward systems have a range of positive effects including increased solidarity, cohesion, and positive emotions (Friedkin 2004; Markovsky and Lawler 1994; Thibaut and Kelley 1959). As Markovsky and Lawler (1994:133) state: “Groups that attempt to control members through the administration of positive rather than negative sanctions are more likely to produce positive emotional responses and thereby foster attractive bonds among members and the group. Thus, positive sanctions promote cohesion and solidarity.” Likewise, Greitemeyer and Weiner (2003, 2008) suggest that people ascribe more benevolent intentions to behavior linked to a reward compared to a punishment. Thus, given that rewards likely bolster more positive feelings than punishments, the “affective” view suggests that reward systems will be less likely than punishment systems to undermine trust and cooperation once the system is discontinued.
Rewards, Punishments, and Trust
We predict that in groups where a punishment system is present, trust is based to a large extent on the sanctioning system. People believe that cooperation is driven by the motivation to avoid punishment. If so, once the punishment system is removed, trust levels will decrease. Punishment systems reduce trust because they signal that others are externally, versus internally, motivated. The attribution argument given previously leads to a similar prediction for rewards. As extrinsic motivators, a reward system will mitigate trust. An alternative prediction for reward systems, based on the affective argument, suggests instead that reward systems emphasize people’s goodness. This reasoning leads to a competing prediction: the trust decline due to the removal of reward systems will be less severe compared to the removal of a punishment system.
Hypothesis 1: Trust will be lower when a punishment system is removed compared to when it is present.
Hypothesis 2a: Trust will be lower when a reward system is removed compared to when it is present.
Hypothesis 2b: Trust will remain stable when a reward system is removed compared to when it is present.
Hypotheses 1a through 2b focus on the within-subjects effects of sanctioning systems, but we also address between-groups effects and expect that punishment systems will undermine trust. In groups where punishments have been removed, trust will be lower compared to a control group where a punishment system is never used. As stated earlier, we test this argument via the “removing the sanction” method. Hypotheses 3 through 4b specify these predictions:
Hypothesis 3: Trust will be lower when a punishment system is removed compared to when it is never used.
Hypothesis 4a: Trust will be lower when a reward system is removed compared to when it is never used.
Hypothesis 4b: Removal of a reward system will not undermine trust compared to a control group.
Rewards, Punishments, and Cooperation
If sanctioning systems undermine trust, we should also expect to observe reduced cooperation. As we argued in the introduction, trust may be necessary for sustained cooperation. Our reasoning stems from a large literature showing a strong and positive relationship between trust and cooperation in social dilemmas (e.g., Buchan, Croson, and Dawes 2002; De Cremer and Stouten 2003; Orbell, Dawes, and Schwartz-Shea 1994). According to this research, trust motivates cooperation by reducing individuals’ fear of exploitation from others, making cooperation less risky (Yamagishi and Sato 1986).
Hypothesis 5: There will be a positive association between trust and cooperation.
If trust predicts cooperation, then we should expect the same findings for cooperation as we do for trust. If the attribution argument presented earlier is correct, within-subjects tests should show lower cooperation once positive and negative sanctioning systems are removed compared to when they are present. Likewise, between-subjects analyses should indicate that cooperation is lower after the removal of sanctioning systems compared to a control group; sanctioning systems should undermine cooperation. If the affective argument is accurate, removal of a reward versus punishment system should not reduce cooperation. Also, unlike punishment systems, removal of a reward system should not undermine cooperation. Hypotheses 6 through 7b specify the within-subjects predictions, while the between-subjects predictions are given by Hypotheses 8 through 9b.
Hypothesis 6: Cooperation will be lower when a punishment system is removed compared to when it is present.
Hypothesis 7a: Cooperation will be lower when a reward system is removed compared to when it is present.
Hypothesis 7b: Cooperation will remain stable when a reward system is removed compared to when it is present.
Hypothesis 8: Cooperation will be lower when punishment systems are removed compared to when they are never used.
Hypothesis 9a: Cooperation will be lower when reward systems are removed compared to when they are never used.
Hypothesis 9b: Removal of reward systems will not undermine cooperation compared to a control group.
We also expect that trust will mediate the relationship between sanctions and cooperation. The effect that sanctioning systems have on cooperation will work through trust.
Hypothesis 10: Trust will mediate the relationship between punishment and reward systems and cooperation.
Study 1
Methods
Experimental design, participants, and procedure
The experiment consisted of three conditions (punishments, rewards, control). Data were collected at two universities in the Netherlands. Participants were students from various disciplines. In total, 129 (79 female) students completed the study in exchange for EUR 5 (60 at one location and 69 at the other). The experiment lasted approximately 30 minutes.
One to eight participants completed the experiment at a time. After arriving at the laboratory they were seated in soundproof individual cubicles containing a computer workstation. Communication with others was prohibited. All instructions and interactions took place via the computer. After reading and signing a consent form, participants read instructions explaining the public goods dilemma. Thereafter, participants played two one-shot public goods games with fictitious others. Trust and cooperation were measured at two points during the experiment.
Public goods dilemma
Participants learned that they would be randomly assigned to a group and that each group member would receive a personal endowment of 100 points. Each point was worth 5 cents. The experimental setup led participants to believe that they would interact with three other participants. Instructions stated: “In this session, you will work together with three other participants in this lab. So, you and three others are part of the same group. The other three are in the same type of cubicle as you are.” Through the instructions and experiment, we aimed to create the impression that they were interacting with three other participants from the same experimental session. Thus, participants believed that their own and others’ contributions would affect the amount of money they earned during the experiment, namely, their pay was not guaranteed. Because we were not interested in group dynamics or group-level outcomes, following related work (e.g., Mulder et al. 2006; Simpson 2003), participants were told they were in groups of four, but this was not actually the case. All participants received the same amount (EUR 5) at the conclusion of the study.
Instructions informed participants that the choice facing group members was how much of their endowment to contribute to a “group fund” and how much to keep. The total amount contributed to the group fund was doubled by the experimenter and distributed equally, regardless of what each contributed. Thus, consistent with the structure of social dilemmas, it was in each individual’s self-interest to contribute nothing. But in following this logic, no public good would be produced, and all would be worse off than if they had all donated.
Participants also learned that contribution decisions would be made anonymously and that no one would receive information about others’ decisions. Additionally, participants were not informed as to who was punished or rewarded. After making their first (phase 1) public goods decision, participants were told they would make a second, similar decision about how much to contribute to a public good (phase 2). Endowments were replenished for phase 2.
Phase 1 decision
Following related work (Mulder et al. 2006), those in the punishment condition were told that the two lowest contributing group members would receive an automatic punishment. The sanctioning system would reduce their earnings by 5 Euros (i.e., their entire starting endowment). Instructions stated: “An automatic punishment will be assessed to the two lowest contributors in your group. Specifically, the two individuals who contribute the least will automatically lose 5 Euros.” Participants were also told that point deductions resulting from punishment would be applied at the end of the study, and participants were not informed as to the outcomes of the first decision before moving on to phase 2. That is, we held constant all information across conditions other than the manipulation (i.e., presence of a punishment or reward). Doing so allowed us to compare the effect of the mere presence of a punishment system to the control and reward conditions.
The reward condition mirrored the punishment condition. Specifically, participants were told that the two highest contributors would automatically receive a 5 Euros reward. Participants were not informed as to whether or not they would receive a reward until after the study was complete. As in the punishment condition, they did not learn about the outcomes of the phase 1 decision before moving on to phase 2.
Participants in the control condition received no instructions about punishments or rewards. Thus, they made decisions in a standard public goods game.
After receiving information about the public goods game, but before making decisions, we measured participants’ level of trust in other group members. Our trust measure follows Mulder et al. (2006) and asks: “Do you trust that other group members will contribute to the group fund?” Response categories ranged from 1 = not at all to 7 = completely. After the trust question, participants were asked how many points, if any, they wished to contribute to the public good. Contribution amounts measured phase 1 cooperation.
Phase 2 decision
For this decision, participants learned that they would remain in the same group and again decide how much of a 100 point endowment to contribute to a public good. Participants were not aware of the second phase during phase 1. As in phase 1, the total amount contributed would be doubled and divided equally. Following Mulder et al. (2006), the difference between the first and second phase was that sanctioning systems were discontinued. They were told: “Unlike your first decision, there will be no punishments/rewards. Regardless of contribution decisions, no one will be rewarded or punished. More specifically, the two lowest/highest contributors will receive no punishment/reward.” As in the first phase, participants were told that their phase 2 decision would be anonymous. Those in the control condition were told that the phase 2 decision was identical to phase 1. After receiving instructions, we measured trust with the same question from phase 1. After completing the items, participants made their contribution decision, which measured cooperation at phase 2. Thereafter, participants were debriefed and paid.
Results
Much previous work shows inconsistent effects of gender on trust (see Croson and Gneezy 2009) and cooperation (see Simpson and Van Vugt 2009). Gender is therefore relevant to both of our dependent measures. Thus, we include gender as a covariate in all analyses but report findings related to gender only when significant.
Trust Results
Hypotheses 1, 2a, and 2b
The first two hypotheses (1 and 2a) state that trust will be lower when punishment and reward systems are removed compared to when they are present. A competing hypothesis predicts that trust will remain stable when a reward system is present versus absent (2b). We tested the hypotheses using a 3 (punishment, reward, control) × 2 (phase) ANOVA with phase as a within-subjects factor. This analysis yielded a marginal effect of phase, F(1, 125) = 3.15, p = .08, showing that trust was lower in phase 2 (M = 4.38; SD = 1.60) compared to phase 1 (M = 4.89; SD = 1.25). This main effect was qualified by a phase × condition interaction, F(2, 125) = 10.79, p ≤ .001, demonstrating that trust decreased in the punishment, t(43) = 2.94, p ≤ .01, and reward conditions, t(42) = 3.66, p ≤ .001, but not in the control (p = .34) (see Table 1 for means; all tests are two-tailed). These findings show that trust decreased significantly once reward and punishment systems were removed. Therefore, Hypotheses 1 and 2a are supported, and Hypothesis 2b is disconfirmed. Findings also indicated a significant phase × gender interaction, F(1, 125) = 6.57, p ≤ .01, such that trust decreased more for men t(49) = 3.92, p ≤ .001 than women t(78) = 2.66, p ≤ .01. The condition × gender interaction was not significant.
Study 1 Trust and Cooperation in Phase 1 and Phase 2
Note: Within columns, means that do not share a letter in the superscript are significantly different, p < .05, pair-wise comparisons with t-tests. All tests are two-tailed.
We also performed a 2 (punishment vs. reward) × 2 (phase) ANOVA with phase as a within-subjects factor to determine if the trust decline from phase 1 to phase 2 differed in the punishment and reward conditions. Results indicated a significant main effect of phase, F(1, 83) = 9.47, p ≤ .001, showing that trust levels were lower in phase 2 (M = 4.13; SD = 1.68) than phase 1 (M = 4.98; SD = 1.23) when the control group was excluded. The phase × sanction interaction was nonsignificant (p = .58); the decline in trust from phase 1 to phase 2 was similar in the punishment and reward groups. In other words, punishment and reward systems were equally detrimental to trust when they were removed than when they were present. The phase × gender interaction was significant, F(1, 125) = 6.57, p ≤ .01, when excluding the control.
Hypotheses 3, 4a, and 4b
Hypotheses 3 and 4a predict that punishment and reward systems undermine trust. Hypothesis 4b predicts no difference for those in a condition where a reward system is removed versus a control. For these hypotheses, phase 2 trust is the dependent measure. Results of a one-way ANOVA showed a significant effect of condition, F(2, 128) = 5.55, p ≤ .01. Pair-wise comparisons showed a marginally significant difference between the punishment (M = 4.32; SD = 1.70) and control (M = 4.88; SD = 1.31) conditions, t(84) = 1.70, p = .09. Participants in the reward (M = 3.95; SD = 1.66) condition reported significantly lower trust compared to those in the control, t(84) = 2.88, p ≤ .01. There was no difference between the reward and punishment condition (p = .31). These findings supported Hypotheses 3 and 4a but disconfirmed Hypothesis 4b. Once again, reward and punishment systems had similar negative effects on trust. The results also indicated a gender effect on phase 2 trust, F(1, 128) = 9.41, p ≤ .01, with women (M = 4.65; SD = 1.37) trusting more than men (M = 3.96; SD = 1.84), t(127) = 2.45, p ≤ .01. Prior evidence on the effects of gender on trust is mixed. Some studies find that women are less trusting than men (e.g., Buchan, Croson, and Solnick 2008), and others find no effects of gender on trust (e.g., Croson and Buchan 1999). While we did not observe any main effects of gender on trust, t(41) = –.55, p = .42, we did find that women were less susceptible than men to the trust-undermining effects of sanctions. That is, women’s faith in others’ cooperation did not decline as severely as men’s trust once sanctions were removed.
We also assessed whether there were differences across conditions on phase 1 trust. Results of a one-way ANOVA showed a nonsignificant effect (p = .43), demonstrating no differences in trust when sanctions were present. 2 Taken together, these findings suggest that the presence of sanctioning systems did not significantly bolster trust (compared to a control group), but removing sanctioning systems significantly undermined trust.
Cooperation Results
Hypothesis 5
Consistent with much prior work, we predicted an association between trust and cooperation. Correlational analyses demonstrated a strong and positive relationship between trust and cooperation in phase 1 (r = .41; p ≤ .001) and phase 2 (r = .61; p ≤ .001). These findings support Hypothesis 5.
Hypotheses 6, 7a, and 7b
We tested the hypotheses using a 3 (punishment, reward, control) × 2 (phase) ANOVA with phase as a within-subjects factor. There was a significant effect of phase, F(1, 125) = 5.89, p ≤ .01, showing that cooperation was lower in phase 2 (M = 56.19; SD = 34.96) compared to phase 1 (M = 70.05; SD = 31.45). This main effect was qualified by a phase × condition interaction, F(2, 125) = 6.89, p ≤ .001, demonstrating that cooperation decreased in the punishment, t(42) = 5.21, p ≤ .001, and reward conditions, t(42) = 3.28, p ≤ .01, but not in the control (p = .54) (see Table 1 for means). Thus, Hypotheses 6 and 7a are supported, and Hypothesis 7b is disconfirmed. Similar to the trust results, findings also indicated a phase × gender interaction, F(1, 125) = 10.13, p ≤ .01, such that cooperation decreased slightly more for men, t(49) = 4.61, p ≤ .001, than women, t(78) = 3.67, p ≤ .001. That is, women’s cooperation was not as sensitive to the removal of sanctions compared to men. Gender did not interact with condition.
We also performed a 2 (punishment vs. reward) × 2 (phase) ANOVA with phase as a within-subjects factor to determine if the cooperation decline from phase 1 to phase 2 differed in the punishment and reward conditions. Results indicated a main effect of phase, F(1, 83) = 9.03, p ≤ .01, showing that cooperation levels were lower in phase 2 (M = 52.69; SD = 34.89) than phase 1 (M = 71.98; SD = 31.31) excluding the control group. The phase × sanction interaction was nonsignificant (p = .62), indicating that the decline in trust from phase 1 to phase 2 was similar in the punishment and reward conditions. Punishment and reward systems were equally detrimental to cooperation when they were removed versus present. The phase × gender interaction was also significant, F(1, 125) = 12.55, p ≤ .001, when excluding the control group.
Hypotheses 8, 9a, and 9b
To test between-subjects effects, we used a one-way ANOVA on phase 2 cooperation. Results yielded a nonsignificant effect (p = .11). To further explore the relationship, we performed a less conservative test using orthogonal contrasts on sanctions (punishment and reward) versus the control and punishments versus rewards. There was a significant difference between the sanctions (M = 52.69; SD = 34.89) and control conditions (M = 63.18; SD = 34.45), F(1, 128) = 3.75, p ≤ .05. Sanctioning systems therefore had an undermining effect on cooperation compared to the control group. Analyses indicated no difference in cooperation between the punishment (M = 55.60; SD = 33.43) and reward conditions (M = 49.79; SD = 36.45) (p = .43). Thus, reward and punishment systems had similar negative effects on cooperation. These findings lend some support to Hypotheses 8 and 9a and indicate that the undermining effects of sanctioning systems have negative consequences for not only trust but also cooperation.
Hypothesis 10
To assess whether the negative effect of sanctioning systems on cooperation was mediated by trust, we performed mediation analyses (see Hayes 2013). Specifically, we focused on the decrease in trust from phase 1 to phase 2 as a mediator between sanctions and cooperation. Results showed significant mediation (z = -2.80; p ≤ .01). Thus, the removal of sanctions and resulting lower trust had a negative impact on cooperation. This finding is consistent with the argument that trust has a direct impact on cooperation. Declining trust precipitates decreased cooperation levels. 3
We also addressed whether sanctioning systems, when present, enhanced cooperation. Results from a one-way ANOVA on phase 1 cooperation indicated a nonsignificant effect (p = .43). In addition, we performed orthogonal contrasts, which showed that sanctions (punishment and reward) (M = 71.98; SD = 31.31) did not boost cooperation compared to the control condition (M = 66.18; SD = 31.74) (p = .27). The difference between rewards (M = 67.53; SD = 35.45) and punishments (M = 76.44; SD = 26.19) was also nonsignificant (p = .19). These findings contrast some prior work showing that sanctions enhance short-term cooperation (e.g., Fehr and Gächter 2002) but are consistent with other research indicating that when present, strong systems (which we use) may do little to bolster cooperation (see van der Weele 2009). We return to this issue in more detail in the following.
Discussion
At the outset of the article, we outlined two ways in which trust may be affected by sanctions. One line of reasoning (leading to Hypotheses 1, 2a, 3, and 4a) was that trust is similarly undermined by punishments and rewards due to attributional effects. An affective argument (leading to Hypotheses 2b and 4b) is that rewards may have less negative effects on trust because they evoke less negative sentiments than punishments. The findings from Study 1 were that trust was lower when punishment and reward systems were discontinued compared to when they were present. More importantly, both punishment and reward systems undermined trust compared to a control group who never experienced sanctions. This pattern of findings supports the attributional reasoning that external inducements, whether punishment or reward systems, undermine trust as they lead individuals to expect that others are externally (rather than internally) motivated.
Findings from Study 1 also indicated a significant undermining effect of cooperation when sanctioning systems were removed. Thus, sanctions were detrimental to cooperation, and these effects appeared immediately after the removal of the sanctioning system. In addition, mediation analysis indicated that the negative impact of sanctions on cooperation worked through declining trust. This provides some evidence that sanctioning systems might also negatively impact cooperation.
It was striking that, when present, the sanctioning systems we implemented did not bolster cooperation. This finding runs counter to some prior research indicating that (negative) sanctions have a positive effect on cooperative behavior (see Fehr and Gächter 2002; Gürerk et al. 2006). Other work, however, suggests that strong sanctions give mixed messages: inducements motivate free-riders to cooperate, but pursuit of self-interest must be rampant to warrant such strong incentives (van der Weele 2009). In our experiment, sanctions were strong, as participants believed they could receive a punishment (reward) equivalent to their entire starting endowment. Such strong incentives may have signaled (as hypothesized) that free-riding is common in these scenarios, which may have created a norm of noncooperation. Alternatively, a large sanction may have motivated reactance, where a prohibited behavior becomes more attractive precisely because it is prohibited (Brehm 1966). Strong sanctions exert pressure to conform, which group members may have resisted. Cooperation, however, was not lower than the control group when sanctioning systems were present. This may have been because the inducement buoyed expectations and behavior. Consequently, compared to those in a control group, members of groups with sanctions were not more likely to cooperate.
The primary findings from Study 1 were that the removal of positive and negative sanctioning systems undermined trust. Removing sanctions also undermined cooperation, and this effect was mediated by trust. An important implication may be that rewards are not a viable alternative to punishments as selective incentives. In a second study, we test the robustness of these key findings using weaker sanctioning systems. After all, when present, the sanctioning systems used in Study 1 did not bolster cooperation and were therefore ineffective. As such, the question remains whether effective sanctioning systems undermine trust and cooperation. As we argued previously, the relative size of the sanction is one possible explanation for why sanctions did not increase cooperation when present. We use less drastic sanctions in Study 2 to explore whether weaker sanctions can bolster short-term cooperation and whether these weaker sanctions display trust- and cooperation-undermining effects. If replicated, the findings may suggest that group member trust and cooperation is sensitive to weaker inducements; regimes need not employ drastic sanctions to have negative consequences.
Study 2
Methods
Design and participants
Data were collected in the United States at universities in the southeast and southwest. Participants at both sites were recruited from Introductory Sociology classes in exchange for course credit and raffle tickets for a drawing for four $100 Amazon gift cards. Our use of a raffle as an indirect monetary incentive follows previous social dilemmas research (e.g., Batson et al. 1999; Van Vugt and DeCremer 1999). Participants were led to believe that they would receive one raffle ticket for every 10 points accumulated during the experiment. But in reality, all participants had an equal opportunity to receive a gift card; each participant’s name appeared once in the raffle regardless of the number of points earned. A total of 200 students (147 female) participated in the study (68 at one location and 132 at the other). Six (3 percent) reported suspicion about others and were therefore excluded from the analyses.
The design of Study 2 was identical to Study 1 except for the magnitude of the sanctions. In Study 2, participants in the punishment condition were told the two lowest contributors would lose 60 percent of their starting endowment, compared to 100 percent in Study 1. Likewise, in the reward condition, the two highest contributors could earn a reward equivalent to 60 percent of their starting endowment. All other procedures were identical to Study 1.
Results
Trust Results
Hypotheses 1, 2a, and 2b
As in Study 1, all analyses control for gender. We only report gender effects when they are significant. Consistent with Study 1 findings, within-subjects analyses indicated a main effect of phase, F(1, 191) = 38.37, p < .001, with trust being lower in phase 2 (M = 3.84; SD = 1.39) compared to phase 1 (M = 4.42; SD = 1.14). This effect was qualified by a phase × condition interaction, F(2, 191) = 13.29, p < .001, with a decrease in the punishment, t(62) = 8.19, p < .001, and reward conditions, t(65) = 5.22, p < .001, but not in the control (p = .86). 4 See Table 2 for means. Hypothesis 1 and 2a are again supported.
Study 2 Trust and Cooperation in Phase 1 and Phase 2
Note: Within columns, means that do not share a letter in the superscript are significantly different, p < .05, pair-wise comparisons with t-tests. All tests are two-tailed.
Hypotheses 3, 4a, and 4b
Between-subjects analyses on trust in phase 2 yielded a significant effect, F(2, 193) = 8.15, p < .001. Participants reported lower trust in both the punishment and reward conditions compared to those in the control (p < .01 for each) (see Table 2 phase 2 column). The difference between reward and punishment groups was nonsignificant (p = .52). Similar to Study 1 findings, these findings supported Hypotheses 3 and 4a; once removed, rewards and punishments undermined trust.
Results from a one-way ANOVA on phase 1 trust were nonsignificant (p = .38), indicating no difference between conditions. Consistent with Study 1, when present, rewards and punishments provided no boost in trust (compared to the control condition) (see Table 2 phase 1 column for means). Thus, weaker sanctions were no more useful for increasing trust than the strong sanctions used in Study 1.
Cooperation Results
Hypotheses 5, 6, 7a, and 7b
Once again, trust was correlated with cooperation in phase 1 (r = .33; p < .001) and phase 2 (r = .44; p < .001), which supported Hypothesis 5. As in Study 1, within-subjects tests showed a significant effect of phase, F(1, 190) = 36.36, p ≤ .001, with lower cooperation in phase 2 (M = 52.70; SD = 29.39) compared to phase 1 (M = 63.47; SD = 24.81). This main effect was qualified by a phase × condition interaction, F(2, 190) = 9.10, p ≤ .001, demonstrating that cooperation decreased in the punishment, t(62) = 4.19, p ≤ .001, and reward conditions, t(65) = 3.96, p ≤ .001, but not in the control (p = .59) (see Table 2). Once again, Hypotheses 6 and 7a are supported, and Hypothesis 7b is disconfirmed.
Hypotheses 8, 9a, 9b, and 10
Between-subjects comparisons yielded no difference between conditions on phase 2 cooperation (p = .75). As in Study 1, we also performed orthogonal contrasts by comparing sanctions (rewards and punishments) to the control. Findings indicated no difference (p = .68). Thus, unlike Study 1, once removed, sanctions had no immediate undermining effect on cooperation. That is, weaker sanctions, although undermining trust, did not negatively impact cooperation. Mediation analyses, however, indicated an indirect effect of sanctions on cooperation via reduced trust (z = -2.86; p ≤ .01). 5
We also address cooperation when sanctions were present. Between-subjects analyses on phase 1 cooperation indicated a significant difference between conditions, F(2, 193) = 5.67, p ≤ .01. Cooperation was higher in both the reward and punishment conditions compared to the control group (p < .05 for each; see Table 2). The difference between the punishment and reward groups was nonsignificant (p = .17). Here, the presence of punishments and rewards increased short-term cooperation. The weaker sanctions used in Study 2 were more effective than the strong sanctions used in Study 1. When present, weaker sanctions enhanced cooperation and once removed did not undermine cooperative behavior. However, the weaker sanctions did undermine trust, which may precipitate lower cooperation in future interactions. That is, the effect of undermined trust generated by weaker sanctions may not manifest immediately. We return to this issue in greater detail in the following.
General Discussion
Results from our experiments yielded several key findings related to trust. First, across both studies we found that reward and punishment systems were equally detrimental to trust. Trust declined significantly when rewards and punishments were removed compared to when they were present. More important, we found that both positive and negative sanctioning systems undermined trust compared to a control group never exposed to incentives. To our knowledge, this research is the first to establish that reward and punishment systems have similar trust-undermining effects.
These findings shed new light on the viability of rewards as alternatives to punishments in social dilemmas. Recent work (Balliet et al. 2011) shows that when present, reward and punishment systems are equally effective at promoting short-term cooperation. We found that once removed, both positive and negative inducements had detrimental effects on trust. Installing and maintaining sanctioning systems is often costly, and group members frequently face situations in which their behaviors are undetected. Our results indicate that in such situations group members will likely have decreased trust in others’ willingness to cooperate. The trust literature points to several negative consequences of distrust, including decreased productivity and efficiency (Levi, Moe, and Buckley 2004) and a heightened need for formal contracts and costly monitoring (Malhotra and Murnighan 2002).
Our finding that not only punishment systems but also reward systems undermine trust adds to the existing literature. First, some have suggested that rewards are more apt than punishments to foster positive emotions (Markovsky and Lawler 1994). In this view, it could be argued that rewards highlight the cooperative side of human nature more than punishments and may therefore have less detrimental effects on trust. Yet results from both of our studies suggest that reward systems may be just as detrimental to trust. Thus, findings were inconsistent with an affective explanation, which suggested that punishment and reward systems elicit different emotional responses. This explanation suggested that a reward system would not undermine trust and cooperation. Results instead supported the attribution explanation. This approach predicted similar undermining effects for reward and punishment systems because, as external inducements, each spurs individuals to attribute others’ behaviors to external and not internal motivations. Second, the undermining effect of both punishment and reward systems suggests that the previously found trust-undermining effects of punishments are not due to their retributive nature, highlighting the “badness” of people. After all, rewards are nonretributive and exerted the same effects as punishments, once removed. Thus, it is more likely that sanctioning systems, whether they are punishments or rewards, are mainly seen as an extrinsic motivation that drives others to cooperate and thus drives out trust.
Results from our experiments also address the relationship between sanctioning systems and cooperation. Once removed, the strong sanctions in Study 1 undermined cooperation. Mediation analyses showed that the negative effect of sanctioning systems on cooperation worked through trust. Weaker sanctions did not have these cooperation-undermining effects. Plus, the experiments demonstrated that strong sanctions did not bolster cooperation, but weak sanctions increased cooperative behavior. These findings are consistent with work addressing the signaling power of sanctions. Using tax evasion as an example, van der Weele (2009:104) contends that increasing fines to curtail the problem may in fact exacerbate it: “Being tough on tax evasion sends a mixed message: although evaders are being punished, they must be numerous to be taken so seriously.” Strong sanctions may unintentionally suggest that bad behavior is common. As such, they may do little to promote cooperation above levels found in groups without sanctions. Alternatively, reactance may have led people to cooperate less as they resisted strong pressure to conform. Our findings indicate that strong sanctions were not only unable to bolster short-term cooperation, they reduced cooperation once removed.
The aforementioned argument regarding the nonsignificant effect of strong sanctions on cooperation in Study 1 is speculative. It is noteworthy that the baseline level of cooperation is considerably higher in Study 1 (M = 66.18; SD = 31.74) than in Study 2 (M = 55.78; SD = 25.73), hence the marginal effect of sanctions is, not surprisingly, smaller in Study 1. Thus, the observed effects may be attributable to differences in samples. Findings from Boone, Declerck, and Kiyonari (2010) are relevant to this argument. They found that the effect of extrinsic incentives is less pronounced among those already willing to cooperate. It may be that the Study 1 sample included more prosocials than Study 2. 6 It is also possible that cultural differences between the Dutch (Study 1) and American (Study 2) participants are responsible for the observed differences. Indeed, work by Herrmann et al. (2008) and Balliet et al. (2011) show considerable cross-cultural difference in the effects of incentives on cooperation. In any case, more work is needed to clarify factors influencing the impact of sanction size on cooperation in social dilemmas.
In Study 2, weaker sanctions undermined trust but did not have cooperation-undermining effects. We expected that cooperation results would mirror those of trust. Why would individuals cooperate if they do not trust? There are at least two explanations. First, it is important to consider when and how the trust-undermining effects of sanctioning systems manifest in reduced cooperation. It may be that the removal of sanctioning systems produces reduced cooperation immediately (see Gneezy and Rustichini 2000) or after multiple interactions (see van der Weele 2009). The weaker sanctions used in Study 2 may have required additional interactions to generate negative effects on cooperation.
A second explanation for the lack of correspondence between trust and cooperation in Study 2 is that cooperative decisions are often based on more than trust alone. They are also guided by internalized norms of benevolence (Biel and Thogersen 2007; Kerr 1995), or personal convictions that one should cooperate, even if there is doubt about others’ cooperative intentions. This may explain why undermined trust did not translate directly into undermined cooperation. For some, low trust may not be sufficient to withhold contributions. That is, some may have a preference for cooperation that is relatively immune to expectations of others’ behavior.
Limitations and Directions for Future Research
Earlier, we suggested that sanctions might undermine trust via an attribution process that others are externally motivated, which is consistent with prior work (Mulder et al. 2006). However, we did not directly measure this attribution mechanism. Nevertheless, because each study indicates that reward and punishment systems both undermine trust, the attribution explanation is supported and the affective explanation is not. A possible avenue for future work would be to show more direct evidence of the attribution explanation by including measures of whether people attribute fellow group members’ cooperation to external versus internal factors.
A second limitation is that the experimental design did not allow us to investigate when undermined trust manifests in reduced cooperation. We used a series of one-shot interactions that measured cooperation when sanctions were present and immediately after they were removed. As we discuss previously, the strong sanctions used in Study 1 immediately undermined cooperation, but the weaker sanctions in Study 2 did not have an immediate effect. Future work should address whether weaker sanctions, once removed, reduce cooperation over repeated interactions.
A more general issue for future research is how we might explain the level of cooperation we observe in the real world in light of results suggesting that sanctioning systems drive out trust (and potentially cooperation). Results from our experiments are consistent with prior work on cooperation in real-world groups. Ostrom (2000) reviews research showing that externally levied punishment and reward systems, like those used in our experiments, are ineffective at promoting cooperation. As she suggests, “evidence has accumulated that externally imposed rules tend to ‘crowd out’ endogenous cooperative behavior” (Ostrom 2000:147). Our findings dovetail with Ostrom’s (2000) argument and indicate that groups subject to external sanctions may show lower intragroup trust. Field research instead demonstrates that cooperation levels are highest when sanctioning derives from sources inside the group and is delivered by a single individual or small contingent of selected monitors (see Ostrom 2000 for a discussion). Sanctions in these groups are directed at behavior that breaks mutually agreed upon rules. According to this perspective, cooperation may require sanctions, but sanctions imposed internally by a small group or single individual are likely to be the most effective (rather than externally imposed sanctions studied in our experiments). Of course, more work needs to be done to better understand the relative impact of external versus internal sanctions on trust and cooperation, which may help uncover the bases for successful collective actions.
Conclusion
Some research shows that sanctioning systems can be beneficial (e.g., Yamagishi 1986). Our work, on the other hand, highlights that positive and negative sanctioning systems, particularly those enforced externally, can undermine trust. Research has shown a variety of negative consequences of distrust, including hindering the development of long-term cooperative and reciprocal relations (Ostrom and Walker 2003; Serva, Fuller, and Mayer 2005) and decreases in group productivity and efficiency (Levi et al. 2004). As both punishments and rewards may hinder group members’ trust, it may be prudent to install a punishment or reward system only if trust and cooperation are relatively low. In this case, such a system will, at least in the short run, induce cooperation through system-based trust, or an “assurance system” (Yamagishi, Cook, and Watabe 1998); low trusters may expect others to cooperate when a sanction is present. Of course, if the incentives are removed, trust and cooperation may plummet, but the benefits of short-term cooperation may outweigh the potential negative drawbacks in groups lacking trust. However, if trust already exists between group members in the absence of sanctions, installing punishment or reward systems may do more harm than good. The mere presence of such a system may increase the need to apply them strictly via the use of formal contracts and monitoring (Malhotra and Murnighan 2002); sanctions can erode the trust that existed in the absence of incentives.
Footnotes
Acknowledgements
The authors are grateful for research support from the faculty and graduate students at the Center for Community Research and Development at Baylor University and Stevie Chamberlain, Marissa Gastelle, Michael Hahn, and Daniela Negraia at the Laboratory of Social Research at the University of South Carolina. An earlier version of this paper was presented at the 2010 meetings of the American Sociological Association in Atlanta, GA.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support for the project was provided by a summer sabbatical from Baylor University to the first author and a National Science Foundation grant (SES-1058236) to the third author.