Fig. 1: Overview of the experimental procedures, design of feedback task, and measures of interest.
From: Brain mediators of biased social learning of self-perception in social anxiety disorder
A Overview. While in the MRI scanner, participants were told to prepare a short (3 minutes) speech to be given to two judges (confederates) who would evaluate their performance and give them feedback later. Participants had 4 min for speech preparation and then gave their speech using the scanner interphone. If the participant remained silent for more than 20 seconds, one of the judges prompted them via the scanner interphone (“Please continue”). The feedback task followed the speech after a brief break. B Trial design of the feedback task. Each of 52 trials started with a short presentation of an evaluative cue, written in a first-person perspective (e.g., “I looked anxious”, “I was convincing”). Participants rated how much they thought this applied to their speech performance, and then received feedback from the judges on the same visual analogue scale (VAS) but written in third-person perspective (e.g., “He looked anxious”, “She was convincing”). This feedback was drawn from a distribution centered on the participant’s self-evaluation, resulting in a feedback mismatch (ΔEval, difference between social feedback and initial self-evaluation) that was either more positive or more negative than the self-evaluation. C Hypothesized group differences in positive versus negative adjustments. We predicted that HC (in blue) would learn more (reflected in steeper slopes/larger beta weights) from positive compared to negative feedback, while SAD (in red) would learn more based on negative compared to positive feedback. D Measuring state self-esteem (‘How do you feel about yourself?’) at the end of each trial allowed us to fit an adapted Rescorla-Wagner learning model that described how positive and negative social feedback impacted self-esteem over time (‘affective updating’). The plots show the time course of state self-esteem (rating data and modeled data) from two example subjects, with high positive and high negative affective updating biases, respectively.
