Two Types of Pessimism in Anxiety: a Computational Psychiatry Approach

Abstract

Anxiety disorders are highly heterogeneous, and computational psychiatry offers a path toward identifying the mechanistic distortions that give rise to specific symptom profiles. A recent influential account proposes that a single deviation from optimal value computation, namely pessimism about one’s own future actions, can explain a wide range of anxious behaviors (Zorowitz et al., 2020). However, clinical theory has long distinguished between two conceptually distinct vulnerabilities: doubt about one’s own competence (self-efficacy) and doubt about the environment’s responsiveness (perceived control). We propose that these accounts constitute two computationally separable forms of pessimism. To cement our double dissociation hypothesis, we extend the single-parameter pessimistic Bellman equation to a two-parameter model in which one controls pessimism about self-efficacy by distorting the policy, and the other regulates pessimism about environmental dynamics by distorting the assumed state transition probabilities. Additionally, we designed a two-stage sequential decision task to engage these parameters differently and administered it online to a subclinical population sample, alongside the Negative Problem Orientation Questionnaire (NPOQ) and the Anxiety Control Questionnaire (ACQ-R) which are meant to capture two types of pessimism. Using hierarchical Bayesian model fitting, we see preliminary evidence for a directionally coherent dissociation. A full learning model that included a trial-by-trial learning rate did not succeed at reproducing this pattern, likely due to a trade-off between parameters during model fitting. Together, these results provide preliminary, model-contingent support for two dissociable pessimistic mechanisms in anxiety and suggest that long-standing clinical distinctions between self-efficacy and controllability may correspond to separable computational signatures.