Neural basis of increased costly norm enforcement under adversity

Abstract

Humans are willing to punish norm violations even at a substantial personal cost. Using fMRI and a variant of the ultimatum game and functional magnetic resonance imaging, we investigated how the brain differentially responds to fairness in loss and gain domains. Participants (responders) received offers from anonymous partners indicating a division of an amount of monetary gain or loss. If they accept, both get their shares according to the division; if they reject, both get nothing or lose the entire stake. We used a computational model to derive perceived fairness of offers and participant-specific inequity aversion. Behaviorally, participants were more likely to reject unfair offers in the loss (vs gain) domain. Neurally, the positive correlation between fairness and activation in ventral striatum was reduced, whereas the negative correlations between fairness and activations in dorsolateral prefrontal cortex were enhanced in the loss domain. Moreover, rejection-related dorsal striatum activation was higher in the loss domain. Furthermore, the gain-loss domain modulates costly punishment only when unfair behavior was directed toward the participants and not when it was directed toward others. These findings provide neural and computational accounts of increased costly norm enforcement under adversity and advanced our understanding of the context-dependent nature of fairness preference.

Keywords: computational modeling; costly norm enforcement; fMRI; fairness; ultimatum game.

Fig. 1

Sequence of events and timing in a trial. Each trial began by presenting the offer to the participant for 6 s. The participant was told to evaluate the offer but not to press any button at this moment. After a 2 s interval, the participant had to decide whether to accept or to reject the offer by pressing one of two buttons. After another interval, the duration of which varied from 1 to 3 s, the outcome of this trial was presented. Upon acceptance, the amount of gain or loss would be divided according to the proposer’s offer. Upon rejection, both the participant and the proposer would get nothing (in the gain domain) or have to pain the full price (in the loss domain).

Fig. 2

Behavioral and modeling results from the UG. (A) Points indicate rejection rates in gain (red) and loss domains (blue), respectively. Lines are rejection rates as a logistic function fitted to those points. (B) The ‘envy’ parameter (α) derived from the computational model (Wright et al., 2011) in gain and loss domains. Error bars represent s.e.m.

Fig. 3

Positive effect of SU modulated by frame. (A) Whole-brain exploratory analysis of the contrast ‘Gain₊ [masked (excl.) by Loss₊]’. (B) Beta values corresponding to 10 types of offers (based on GLM 3) extracted from the VS peak. Error bars represent s.e.m.

Fig. 4

Negative effect of SU modulated by frame. (A) Whole-brain level exploratory analysis of the contrast ‘Loss₋ [masked (excl.) by Gain₋]’. (B) Beta values corresponding to 10 types of offers (based on GLM 3) extracted from the DLPFC peak. (C) The difference in the mean beta values in the gain and loss domain predicted the differences in rejection rates between the loss and gain domain (r = 0.80, P < 0.001). ***P < 0.001 (two-tailed). Error bars represent s.e.m.

Fig. 5

Neural effects of interaction between choice and frame. (A) ROI-based analysis of the contrast ‘Loss (rej-acc) − Gain (rej-acc)’. SVC revealed an activation cluster in the left DS, whose rejection-induced activation was higher in the loss compared with gain domain. (B) Activation timecourse extracted from a 6 mm sphere around the maximum coordinates indicates that this interaction effect was driven by the amplified activation difference in the loss relative to the gain domain. (C) The differences in beta estimates extracted from the activation maximum (Loss − Gain) predicted the increases in rejection rate in the loss relative to the gain domain (r = 0.67, P < 0.05). Note, the white and grey dots are outliers identified by robust regression and they are down-weighted in computing the correlation coefficients (Wager et al., 2005).

Fig. 6

Effect of fairness and domain on third-party punishment behavior. The amount spent on punishment increased as the offer fairness decreased. Gain–loss domain did not modulate third-party punishment. Error bars represent s.e.m.