THEORY-917: Chou, Wilson and Smith read about eight computational studies of exploration in reward-based bandit tasks, mostly in sub-clinical samples, with mixed results and none replicated, as suggesting that directed exploration rises with trait or cognitive anxiety and falls with state or somatic anxiety; none they review links random exploration to anxiety

Source

Chou, Wilson & Smith, LIT-1483, read in full in NOTE-1127 from the PsyArXiv preprint, version 1 (24 June 2024): pp. 15–20 and 28–29. The version of record is closed.

What was actually shown

The review reports:

• More directed exploration with trait anxiety in a dynamic bandit (Aberg et al. 2022) and with generalized anxiety symptoms (Yan et al. 2023); more with cognitive anxiety and less with somatic anxiety in a spatial bandit under risk (Witte et al. 2024); less with somatic anxiety in two datasets of a two-armed bandit (Fan et al. 2023b).
• In the Horizon Task, less directed exploration with state anxiety, partly because anxious participants sought information when it was not useful (Smith et al. 2022c).
• Higher directed exploration in anxious depression than in low-anxiety controls, reduced to control levels by a breathing-based state-anxiety induction (Li et al. 2024; Goldman et al. 2024).
• No study linking random exploration to anxiety (p. 19).

The authors’ summary: “many results could be seen to suggest that trait/cognitive anxiety measures promote exploration, while state/somatic anxiety deters exploration. However, findings are mixed” (p. 18).

What could have come out otherwise: anxiety could have gone with less exploration across the board, as an avoidance account would most simply predict. In these reward tasks it did not.

What this does not say

• That the pattern is established. A positive finding “did not replicate in a second sample within a confirmatory study” (p. 18), a preliminary report found no association with worry or intolerance of uncertainty, and no replication of the main findings is reported.
• Anything about exploration under threat. These tasks reward points; nothing aversive is at stake.
• Clinical anxiety. Most samples are sub-clinical.
• Why. The authors’ explanations (exaggerated uncertainty in trait or cognitive anxiety, felt urgency in state or somatic anxiety) are offered, not tested.

Connections

• CLAIM-446. Narrows the claim. Its defensive authority over evidence acquisition predicts less exploration; that fits the state and somatic facets, not the trait and cognitive ones, at least for reward. The Horizon result, information sought when it cannot be used, is the claim’s “poorly governed information seeking” in one study.
• THEORY-907. The foraging half of the same review, where the direction is consistent.

Note of 2026-10-30: the dropped clause

The old title ended “and random exploration shows no relation to anxiety”. That clause is not kept as a Proposed reading (ADR-047, kind 3). It was imprecise: the review reports that “no studies to date have specifically linked random exploration to anxiety” (p. 19), which is an absence of evidence, and NOTE-1127 rates it so. The old wording turned it into a finding of no relation, which the review does not claim. The rest of the old title, the facet pattern in directed exploration, is kept as a Proposed reading that extends this one.