Exploration Heuristics During Anxiety – an Online Study

Abstract

Aims

Every day, we may choose something new randomly (random exploration) or select something new with no prior information (de-novo exploration). The link between exploration and anxiety has only been studied using trait-like anxiety questionnaires, but an experimental manipulation of anxiety could have different results. Individual differences (e.g., sex or novelty-seeking (NS) trait) also impact specific exploration strategies. Thus, we examined if anxiety manipulation in a task would influence different exploration strategies while also looking at sex, NS bias and trait anxiety.

Methods

117 healthy subjects (58 female) completed online questionnaires (novelty-seeking dimension of the Cloninger's Tridimentional Personality Questionnaire (TPQ-NS), trait anxiety questionnaire (STAI)) and performed the Maggie's farm task. This task allows to review different exploration strategies, promoting exploration via the number of available choices (horizon). The threat of aversive stimuli (loud noises appearing at random times) was used to emulate anxiety, in a between-subject design. Comparing computational models of exploration, the best-fitting model (evaluated by Bayesian Information Criterion) in our data was a Thompson model with an ε-greedy element (random exploration) and a novelty bonus η (de-novo exploration). We used repeated-measures ANOVA, comparing the effect of horizon on the ε and η parameters with the anxiety category as a between-subject factor. We used partial Pearson's correlations of ε and η derivatives (mean and standardised-difference (SD) across horizon) with STAI and TPQ-NS measures correcting for participant's reported stress levels and anxiety category. Partial correlations analyses were repeated after splitting the data by sex.

Results

There was no between-subject effect of anxiety category on the horizon of either ε (F(1,1) = 0.253, p = 0.6) or η (F(1,1) = 0.305, p = 0.58). SD of ε was negatively correlated with TPQ-NS (r = −0.184, p = 0.050) but no other partial correlation was significant. When splitting by sex, SD of ε was negatively correlated with the STAI score (r = −0.341, p = 0.01) in females and the TPQ-NS score in males (r = −0.275, p = 0.038). The mean η positively correlated with the STAI score (r = 0.318, p = 0.016) in males.

Conclusion

While the experimentally modulated anxiety did not affect the exploration parameters, individual differences in NS and trait anxiety are suggested to affect random and de-novo exploration in a sex-dependent manner. Imaging research, or research into anxiety population could help further solidify these results in the future.