Guilty by association: How group-based (collective) guilt arises in the brain

People do not only feel guilty for transgressions of social norms/expectations that they are causally responsible for, but they also feel guilty for transgressions committed by those they identify as in-group (i.e., collective or group-based guilt). However, the neurocognitive basis of group-based guilt and its relation to personal guilt are unknown. To address these questions, we combined functional MRI with an interaction-based minimal group paradigm in which participants either directly caused harm to victims (i.e., personal guilt), or observed in-group members cause harm to the victims (i.e., group-based guilt). In three experiments (N = 90), we demonstrated that perceived shared responsibility with in-group members in the transgression predicted behavioral and neural manifestations of group-based guilt. Multivariate pattern analysis of the functional MRI data showed that group-based guilt recruited a similar brain representation in anterior middle cingulate cortex as personal guilt. These results have broaden our understanding of how group membership is integrated into social emotions.


307
More importantly, the classifier can also discriminate the two group-based guilt Intergroup Emotion Theory (Smith, 1993;Smith and Mackie, 2015). According to this 357 theory, group members feel emotions in response to events affecting other in-group 358 members as if those events were happening to themselves. An implication of this 359 theory is that group-based emotions and personal emotions may share common 360 neurocognitive substrates (Rydell et al., 2008). Our findings provide direct neural 361 evidence for this hypothesis: aMCC responses to an in-group partners' transgressions 362 (relative to an out-group's transgression) was positively associated with increased 363 ratings of guilt, suggesting that aMCC is involved in experiencing group-based guilt.
Our findings shed new light on the role of aMCC in social-affective processing, 366 extending its functional significance to inter-group processes (Lamm,Decety,and 367 Singer ,  is typically between a scenario depicting inter-group harm and a scenario depicting no 386 harm). Thus, emotion measures obtained in these studies may be moderated by 387 different levels of empathy or compassion for the victims 388 Brown et al., 2008;Doosje et al., 1998;McGarty et al., 2005). To address these 389 pitfalls, in the current study, we assessed group-based guilt by comparing 390 transgressions committed by in-group partners with identical transgressions 391 committed by out-group partners (In-group_ Observe > Out-group_ Observe).

392
Because the nature and extent of transgression is matched, any differential effect can 393 thus be attributed to group membership (i.e., in-group/out-group).  Overview. In the current experiment, six participants of the same sex were recruited on 445 each experimental session (none of them had known one another before the session).

Materials and Methods
Upon arrival, participants were told that all six of them were predetermined to be 447 assigned to group A (Transgressor-group) and six further co-players (confederates of 448 the experimenter) in another room were assigned to group B (Victim-group). The task 449 consisted of two phases. In the first, minimal group manipulation phase, the six 450 participants of group A were randomly divided into two sub-groups of three members 451 each to build in-group/out-group context; in the second phase, the participants played 452 multiple rounds of a dot-estimation game either with two in-group partners or two out-453 group partners. The victims would receive electric shocks depending on the 454 performance of the participant and /or his/her partners (Fig. 1). The participants were 455 explicitly told that the victims could not reciprocate the electric shocks.

457
Minimal group manipulation. In the first phase, the six participants of group A were 458 randomly divided into two sub-groups of three members each (a "Yellow Group" and 459 a "Blue Group"). They were asked to wear a yellow or a blue T-shirt corresponding to 460 their group membership. Each sub-group was required to work together to solve a 461 "winter survival problem" (Johnson and Johnson, 1991) to enhance group identity. The immediately after the mini-group manipulation. Player As were explicitly told that 473 player Bs were also divided into two 3-member groups.  To identify brain areas that are shared by personal guilt and group-based guilt, we 648 performed a conjunction analysis (Price and Friston, 1997)  influences of one of these two contrasts (Price and Friston, 1997). This conjunction was 657 formulated as conjunction null hypothesis (Friston et al., 2005;Nichols et al., 2005) 658 and should therefore only yield activations that are significantly present in both original 659 contrasts of the conjunction. The null hypothesis for "conjunction null hypothesis" is 660 that "not all contrasts activated this voxel." If the conjunction results are significant, 661 the null hypothesis is rejected and the conclusion is that "all contrasts activated this 662 voxel." That is, conjunctions represent a logical 'and', requiring both contrasts to be 663 separately significant for the conjunction to be significant.

664
Compared to the univariate analysis, the Multivariate pattern analysis (MVPA) could 667 increase the amount of information that can be decoded from brain activity (i.e. spatial choice test (Chang et al., 2015;Wager et al., 2013;Woo et al., 2014 to and identified more with the in-group partners than the out-group partners: closeness, The pattern of monetary allocation (compensation behavior) in Experiment 2 was similar to the pattern of guilt rating in Experiment 1 (Fig. 3B). Participants compensated more when the harm was caused by in-group partners than out-group partners, β = 0.38, SE = 0.11, t = 3.53, p < 0.001, and when they themselves committed the harm than when they just observe the harm, β = 0.73, SE = 0.09, t = 7.44, p < 0.001. The Group × Objective responsibility interaction (β = 0.16, SE = 0.08, t = 2.14, p = 0.04) showed that the participants allocated more in the In-group_  (Koban et al., 2013;Yu et al., 2014). We conceptually replicated this pattern for personal guilt by a mediation analyses (Preacher and Hayes, 2008 To confirm the stability of the behavioral patterns observed in Experiment 2, we examined the online monetary allocation in Experiment 3, and found that its patterns was the same as the pattern in the fMRI experiment. Specifically, the participants compensated more when the harm was caused by in-group partners than out-group partners, β = 0.52, SE = 0.11, t = 4.68, p < 0.001 and when they themselves committed the harm than when they were not, β = 0.79, SE = 0.14, t = 5.63, p < 0.001 (see Fig.   S1). The Group × Objective responsibility was significant, β = 0.11, SE = 0.05, t = 2.28, p = 0.02. Further tests showed that the participants allocated more in the In-group_ Observe condition than in the Out-group_ Observe condition, β = 0.63, SE = 0.13, t = 4.96, p < 0.001; the participants also allocated more in the In-group_ Commit condition than in the Out-group_ Commit condition, β = 0.41, SE = 0.11, t = 3.56, p < 0.001. The allocation difference in the Observe conditions (In-group_ Observe > Out-group_ Observe, 1.31) was bigger than the Commit conditions (In-group_ Commit > Out-group_ Commit, 0.84), suggesting that the group membership exerted a larger modulation on monetary allocation when in the Observe conditions than in the Commit conditions. Moreover, consistent with the fMRI experiment, the allocation difference in Observe conditions was positively associated with the corresponding post-scan guilt rating difference, r = 0.48, p = 0.003.
We also examined the role of the perceived responsibility sharing in experiencing group-based guilt in Experiment 3. Specifically, Consistent with Experiment 2, the Group membership by Objective responsibility two-way ANOVA analysis on the sense of responsibility was significant, F (1, 34) = 4.35, P = 0.04 (Table S1). Planned t tests showed that participants sensed a higher level of responsibility in the In-group_ Observe condition than in the Out-group_ Observe condition, t (34) = 28.75, p < 0.001, and the difference of the perceived shared responsibility between these two conditions was also positively correlated with the corresponding guilt rating difference, r = 0.41, p = 0.01. No significant effect was found for fear and angry emotion.