We used a probabilistic instrumental learning paradigm where participants had to learn by trial and error which response led to the disappearance of a neutral or angry face. 20 high socially anxious and 20 non-socially anxious individuals with an average level of social anxiety learned to avoid an angry or a neutral face by choosing one of two cues. Each of the cues led to the disappearance of the face either with high or low reinforcement probability.
Groups learned to choose the more effective cue across trials and did not differ with regard to self-report valence, arousal for the faces or the a posteriori estimated reinforcement probability for both cues. High socially anxious individuals as compared to the controls chose the high probability cue significantly more often and were slower particularly when the neutral face could be avoided. Notably, HSA engaged in more avoidance responding to the neutral as compared to the angry face early on during the experiment.
Due to the experimental design, the observed avoidance behavior most likely reflects the motivation for avoidance rather than contingency learning per se.
In social anxiety, neutral faces might be processed as ambiguous social cues and strongly motivate avoidance behavior.