Figure 1
Instability of behavior and volatility of beliefs are markers of paranoia. We replicate prior work on linking behavioral markers from the probabilistic reversal learning task to paranoia. (A) Erratic win-switching behavior present in individuals who self-report having high levels of paranoia (t = –6.66, p < 0.001). (B) Elevated anticipation to volatility also present in high paranoia (t = –5.76, p < 0.001).
Figure 2
Lower metacognition present in individuals who report having paranoia. We compute a composite index of metacognitive structure by averaging across five structured dimensions: comprehension, judgment, evaluation, decision, and confidence. Individuals with higher levels of paranoia demonstrate significantly lower metacognitive structure (t = 5.98, p < 0.001), suggesting reduced ability to reflect on and structure their cognitive experience.
Figure 3
Metacognition attenuates the link between belief volatility and behavior instability. People who reflect more are less affected when beliefs become more volatile. (A) Predicted win-switch rate (WSR) from a binomial GLM fit: wsr ~ * MP + verbosity + cognition. Curves show WSR across belief volatility at low self-reflection (MP = –1 SD; orange) and high self-reflection (MP = +1 SD; blue), accounting for controls. The widening gap at higher illustrates a metacognitive buffer: increases in volatility translate to smaller increases in switching for participants who exhibit greater self-reflection. (B) Probability-scale contrast quantify that buffer. Bars show the absolute change in WSR (percentage points; pp) when increases from –1 to +1 SD, evaluated at low and high MP; error bars are 1,000-bootstrap percentile 95% CIs.