Social Information Is Integrated into Value and Confidence Judgments According to Its Reliability

Abstract

How much we like something, whether it be a bottle of wine or a new film, is affected by the opinions of others. However, the social information that we receive can be contradictory and vary in its reliability. Here, we tested whether the brain incorporates these statistics when judging value and confidence. Participants provided value judgments about consumer goods in the presence of online reviews. We found that participants updated their initial value and confidence judgments in a Bayesian fashion, taking into account both the uncertainty of their initial beliefs and the reliability of the social information. Activity in dorsomedial prefrontal cortex tracked the degree of belief update. Analogous to how lower-level perceptual information is integrated, we found that the human brain integrates social information according to its reliability when judging value and confidence.SIGNIFICANCE STATEMENT The field of perceptual decision making has shown that the sensory system integrates different sources of information according to their respective reliability, as predicted by a Bayesian inference scheme. In this work, we hypothesized that a similar coding scheme is implemented by the human brain to process social signals and guide complex, value-based decisions. We provide experimental evidence that the human prefrontal cortex's activity is consistent with a Bayesian computation that integrates social information that differs in reliability and that this integration affects the neural representation of value and confidence.

Keywords: Bayesian; confidence; integration; social; value; vmPFC.

Figure 1.

A, Task: In part 1 (before scanning), the participant is presented with a series of products from the retail website Amazon (e.g., headphones). The participant enters his/her liking rating R1 followed by his/her confidence rating C1 in the liking rating (not shown the in figure schematic above). In the part 2 (inside of the scanner), the participant sees the same item again, this time together with real reviews from the Amazon website: the mean of the reviews (1–5 stars), the number of reviewers, and a 5-bar histogram showing the distribution of ratings across reviewers. At this stage, the participant is required to enter a new liking R2 and confidence rating C2. All effects predicted by the Bayesian account are significant in the appropriate direction. Shown are fixed effects coefficients from hierarchical linear regression models predicting rating update (R2 − R1) (B1), confidence update (C2 − C1) (B2), and second confidence rating (C2) (B3) for the following predictors: initial deviation from the group (M − R1), interaction between the initial deviation from the group and number of reviews (M − R1 N), interaction between the initial deviation from the group and first confidence rating (M − R1 C1), absolute difference in a participant's initial product rating and the group consensus (|R1 − M|), quadratic function of product rating (R2²). Error bars indicate 95% CI. ***p < 0.001. m.s., Median split.

Figure 2.

A, BOLD signal in mPFC/vmPFC correlates with monotonic increase in liking ratings (peak= −9,38,−11 mm, z = 4.21, p < 0.05, FWE corrected at cluster level). For illustration purposes only, percentage signal change in vmPFC (8 mm sphere centered at the peak of the main effect −9,38,−11) for 3 levels or rating level and confidence (low, medium, and high) are shown; a linear relation between percentage signal changes and rating level and a nonsignificant (linear or quadratic) relation between percentage signal changes and confidence level. B, Activity in mPFC (extending in vmPFC and dmPFC) tracked monotonically the increases in confidence ratings (peak = −9,56,31, z = 4.55, p < 0.05, FWE corrected at cluster level). For illustration purposes only, percentage signal change in mPFC/dmPFC (8 mm sphere centered at the peak of the main effect −9, 56, 31) for 3 levels or rating and confidence (low, medium, and high) are shown; a linear relation between percentage signal change and confidence levels and a significant quadratic relation between percentage signal change and rating levels. The histogram plots are not used for statistical inference (which was performed in the SPM framework); they are shown solely to illustrate the dynamic of the BOLD signal. Error bars indicate SEM. SPM maps are thresholded at p < 0.005 uncorrected for display purposes. C, Conjunction analysis for rating and confidence: activity in mPFC/vmPFC (peak activation at −12,59,4, z = 3.61, p < 0.05, small volume corrected at peak level using at 8 mm centered at −2,52,−2 from Lebreton et al., 2015).

Figure 3.

Spatial gradient analysis along the ventral–dorsal axis of mPFC (see colored dots) for a contrast between the parametric response to rating and the parametric response to confidence (R2 − C2). Data from seven anatomical locations (A) are mapped onto a line and the spatial regression slope is computed (B). Across participants, there is a robust gradient along the medial lane of PFC with response to rating expressed in the more ventral part and response to confidence represented in in the more dorsal part.

Figure 4.

A, Schematic representation of the Bayesian update of liking ratings in response to social information communicated through reviews. The KL divergence parameter indexes the impact of the reviews in shifting the liking rate from the first rating (made in the absence of review information) and the second rating (performed by the participants after seeing the Amazon reviews). B, BOLD signal in dmPFC (peak = −6,50,40) correlates with increase in KL divergence (z = 3.66, p < 0.05, FWE small volume corrected). Percentage signal change for three levels (low, medium, and high) of KL divergence is shown. The histogram plot is not used for statistical inference (which was performed in the SPM framework); it is shown solely to illustrate the dynamic of the BOLD signal. Error bars indicate SEM. C, Between-subject correlation between activity in dmPFC (8 mm ROI centered at −6,50,40) and the degree of resistance to social information (r = 0.77, p < 0.0005). This analysis shows that people who are less influenced by the opinion expressed by others in the reviews have overall more activity in this area.