No Place to Hide: When Shame Causes Proselfs to Cooperate

Abstract

Shame is considered a social emotion with action tendencies that elicit socially beneficial behavior. Yet, unlike other social emotions, prior experimental studies do not indicate that incidental shame boosts prosocial behavior. Based on the affect as information theory, we hypothesize that incidental feelings of shame can increase cooperation, but only for self-interested individuals, and only in a context where shame is relevant with regards to its action tendency. To test this hypothesis, cooperation levels are compared between a simultaneous prisoner's dilemma (where “defect” may result from multiple motives) and a sequential prisoner's dilemma (where “second player defect” is the result of intentional greediness). As hypothesized, shame positively affected proselfs in a sequential prisoner's dilemma. Hence ashamed proselfs become inclined to cooperate when they believe they have no way to hide their greediness, and not necessarily because they want to make up for earlier wrong-doing.

Keywords:

• cooperation
• prisoner's dilemma
• shame
• social dilemma
• social emotions
• social value orientation

Notes

1. There are a number of other explanatory mechanisms in the literature that incorporate the role of emotions to account for why actual decision-making departs from rational models. For example, “the adaptive toolbox of fast and frugal heurstics” provides an evolutionary approach to understand how and why emotions are so important in guiding decision-makers (Muramatsu & Hanoch, 2005).

2. The letters L, K, S and P were chosen at random to denote the options that participants in an experiment can choose given the pay-off matrix shown in Figure 1. We purposefully avoided using the more typical ”option A versus option B” (or option 1 versus 2) to avoid possible experiential biases. For similar reasons, we avoided using “C” for cooperate and “D” for defect.

3. One could argue that it is shameful not to trust a partner, but that implies the partner is personally acquainted and deserving of trust. This is unlikely to be the case in one-time, anonymous interactions.

4. During the experiments, participants played an additional coordination game. The pay-off structure in this game does not offer an incentive to defect. The coordination game was added for the purpose of testing a hypothesis that is unrelated to the present study and is reported elsewhere (Boone et al., 2010). The order of the coordination game and the simultaneous prisoner's dilemma were alternated and counterbalanced. No feedback of the partner decision was given in either game.

5. As a robustness check, we repeated the main analyses of interest without the control variables, which yielded very similar results. As hypothesized, the interactive effect of Shame*SVO is not significant in the simPD (B = .19, SE = .89, p < .83), but it is significant in the seqPD (B = −1.81, SE = .91, p < .009). The effect of the three-way interaction Shame*SVO*Game remains marginally significant (B = 2.00 SE = 1.15, p < .082).

6. To substantiate that concern for reputation affects especially proselfs, two independent judges analyzed the content of the shame essays (n = 59) and evaluated whether or not the participants reported losing face in public (reports of losing face = 1, 0 otherwise). The judges agreed on 75% of the cases. A third judge scored the 25% of the essays on which no agreement was reached, and the score for these remaining essays was determined by the 2/3 majority. There were 21 essays written by proselfs and 32 by prosocials (the remaining 6 shame essays were written by unclassified participants). Proself reported significantly more “losing face” than prosocials (χ ² (1, 55) = 5.02, p < .025). While a number of proselfs who reported losing face still defected, there was not a single cooperating proself who did not report losing face. Hence the idea of losing face might be a necessary but not sufficient condition for proselfs to cooperate. (Note that, because the frequency of cooperating proselfs reporting losing face = 0, the interactive effect of losing face and SVO could not be tested statistically with frequency tables. The latter do not allow for 0 cells.)

7. The negative effect of shame on cooperation for prosocials only (combining the data from both games) is marginally significant (Logistic regression on data pooled for both games, clustered on each individual, with cooperation as dependent variable, age and sex as control variables, and shame as independent variable: B = –.72, SE = .40, p < .070).