Fig 1.
Before receiving any feedback from session 2, participants were asked to use a slider (bottom left) to estimate the probability that they were in the more popular half of students. Participants were then shown a pair of profiles, containing their profile and another participant’s profile, and whether or not they had been chosen by a third participant during session 2 (depicted by a black outline, see central display). After receiving this piece of feedback, participants were again asked to estimate the probability that they were in the more popular half of students (top right). Participants were given twenty pieces of feedback in total and were asked to report their updated judgement following each piece. Participants then repeated this task but observing an anonymous ‘other’ participant whose profile they were shown.
Fig 2.
Relationship between parameter estimates for the prior mean (μ0) and bias in belief updating (b) and depression-specific and anxiety-specific factor scores.
Participants’ standardized scores for the general factor, anxiety-specific factor and depression-specific factor (x axes) are plotted against parameter estimates for the prior mean (μ0) and bias in belief updating (b), obtained using the winning model: Model 3 the “biased RW” model. μ0 is the estimated mean of the participant’s belief distribution at the start of the feedback period and b represents bias in updating in response to positive or negative feedback (b>1 corresponds to a positive bias, whereas b<1 corresponds to a negative bias). P-values for Pearson correlations were obtained using two-tailed permutation-tests with 10,000 samples.
Fig 3.
Evidence for a confirmation bias for self-referential beliefs.
Parameter estimates for the prior mean (μ0; y-axis) and bias in belief updating (b; x-axis), obtained using the winning model (Model 3: the “biased RW” model), are significantly correlated across participants (r = 0.5, p<0.001). This can be thought of as a form of confirmation bias, with new information (positive versus negative feedback) being incorporated more strongly into participants’ beliefs when it aligns with prior beliefs.
Fig 4.
Testing for Depressive Realism.
We used the percentage of times that a participant was actually chosen as a potential partner in session 2 (‘true % selected’) to investigate if depression-related differences in starting beliefs (μ0) might actually reflect accurate perceptions of genuine differences in ‘profile popularity’, i.e., popularity as a potential internship partner. (a) Scores on the depression-specific latent factor were significantly negatively correlated with starting beliefs, r(64) = -0.33, p = 0.006. (b) Scores on the depression-specific factor showed a non-significant negative association with profile popularity (as indexed by true % selected, y-axis), r(64) = -0.17, p = 0.17. (c) Participants had poor insight into their profile popularity; correlation between true % selected and starting belief: r(64) = 0.06, p = 0.62. (d) Partialing out profile popularity (i.e., the true % selected for each participant) had little impact on the significant relationship between depression-specific factor scores and starting beliefs (partial r = -0.33, p = 0.007). A causal mediation analysis [32] confirmed that the relationship between scores on the depression-specific factor and negative starting beliefs was not mediated by profile popularity; Average Causal Mediation Effect (ACME) = 0.0002, p = 0.97.
Fig 5.
Evidence for a confirmation bias for other-referent beliefs.
Parameter estimates for the prior mean (μ0; y-axis) and bias in belief updating (b; x-axis), obtained using the winning model (Model 3: the “biased RW” model), are significantly correlated across participants (r = 0.41, p<0.001); this relationship also holds without the outlier, r = 0.44, p<0.001). Thus, participants, as a group, tend not only to show a confirmation bias when updating beliefs related to themselves (as shown in Fig 4) but also when updating their beliefs about other participants. Here, a confirmation bias means that positive/negative feedback is incorporated more strongly into beliefs when it aligns with prior beliefs.