Social manipulation of preference in the human brain

Abstract

Our preferences are influenced by what other people like, but depend critically on how we feel about those people, a classical psychological effect called "cognitive balance." Here, we manipulated preferences for goods by telling participants the preferences of strongly liked or disliked groups of other people. Participants' preferences converged to those of the liked group, but diverged from the disliked group. Activation of dorsomedial prefrontal cortex (dmPFC) tracked the discrepancy between one's own preference and its social ideal and was associated with subsequent preference change (toward the liked and away from the disliked group), even several months later. A follow-up study found overlapping activation in this same region of dmPFC with negative monetary outcomes, but no overlap with nearby activations induced by response conflict. A single social encounter can thus result in long-lasting preference change, a mechanism that recruits dmPFC and that may reflect the aversive nature of cognitive imbalance.

Figure 1

Balance theory and experimental protocol. (a) Arrows represent the direction of evaluation together with indicated valence (+, like; −, dislike). Any imbalanced state has an odd number of negative (−) attitudes. The figure shows an example of an imbalanced state (e.g., one negative attitude) that would motivate a change in one’s evaluation of the object (towards increased preference in this example). (b) Present experiment. Participants were students at the California Institute of Technology (Caltech). Their attitude toward others was manipulated by using a validated liked group (fellow Caltech students) and disliked group (sex offenders). Participants rated their preferences for t-shirts and were subsequently given feedback about the other group’s preferences for the same t-shirts. (c) During the first preference rating task, subjects rated 174 t-shirt designs using a 14-point scale. Immediately after rating a t-shirt, subjects viewed their own preference and the preference of one of the two groups (either Caltech students or sex offenders), dichotomized as liked or disliked. Thumbs-up corresponded to ratings ≥ 8 and thumbs-down to ratings ≤ 7 on the 14-point scale. (d) Eight possible combinations of subjects’ preference and others’ preference (plus two control conditions). Four of them represent imbalanced states (highlighted by red squares) according to balance theory.

Figure 2

Preference change and dmPFC activation induced by cognitive imbalance. (a) Self-reported preference change between second and first ratings. Red arrows indicate imbalanced conditions. (b) dmPFC regions significantly correlated with the degree of cognitive imbalance (Cognitive Imbalance Index; CII) in each trial. (c) Breakdown of activation patterns in dmPFC during the feedback period of the first preference rating tasks. Beta values were extracted using a leave-one-subject-out (LOSO) cross-validation procedure from the nearest local maximum from the peak activation identified by the conjunction analysis (see Methods for more details). Especially high activations were observed in imbalanced conditions (red arrows). Means and S.E.M. shown.

Figure 3

Pooled within-subject correlations between preference change and dmPFC activation. dmPFC activations significantly predicted subsequent preference change, (A) several minutes after viewing others’ preferences (18 subjects X 8 conditions = 144 data points) and, (B) even after four months (15 subjects X 8 conditions = 120 data points). Y-axis indicates preference change for each condition in the predicted direction (i.e., higher value indicates preference increase in Caltech students-like or sex offenders-dislike conditions, and preference decrease in Caltech students-dislike or sex offenders-like conditions). For both preference changes and brain activations, subject-mean centering was performed to remove between-subjects variance before computing correlations. * p < 0.05, *** p < 0.001 (after Bonferroni correction for 14 tests, see Table 2).

Figure 4

pMFC areas sensitive to response conflict and negative outcome. (a) Response conflict-related areas were localized by the contrast of Interference vs. Control conditions in the MSIT task (Bush and Shin, 2006), which activated the pre-SMA (x = −6, y = 12, z = 44). Negative outcome-related areas were localized by the contrast of Miss vs. Hit feedback in the MIDT task (Knutson et al., 2000) which activated the posterior part of dmPFC (x = 6, y = 30, z = 46). The yellow outline indicates the dmPFC areas significantly correlated with CII in our balance task (cf. Figure 2b). (b) Activation patterns in pre-SMA and posterior dmPFC during the two localizer tasks. Beta values were extracted using a leave-one-subject-out (LOSO) cross-validation procedure from the local maxima from the peak activation identified by the Interference vs. Control contrast (pre-SMA) and the Miss vs. Hit contrast (posterior dmPFC) (see Methods for more details). Activation patterns in (c) pre-SMA and (d) posterior dmPFC during the feedback period of the t-shirt rating task. Means and S.E.M. shown.