Fig. 2: Reversal learning behavior and psychometric fits.
From: Interacting with volatile environments stabilizes hidden-state inference and its brain signatures
a Response reversal curves. Fraction of hidden state \(s_2\) reported behaviorally (y axis) in the four trials preceding (left) and following (right) a reversal from \(s_1\) to \(s_2\) (x axis). The thin dotted line indicates the position of the reversal. Dots indicate the observed data (means ± SEM, n = 24 participants), whereas lines and shaded error bars indicate best-fitting saturating exponential functions (means ± SEM, n = 24 participants). b Best-fitting parameters of saturating exponential functions in the cue-based and outcome-based conditions. Black dots and error bars indicate group-level means ± SEM, whereas colored dots indicate participant-level estimates (n = 24 participants). Left: the reversal time constant is longer and the asymptotic reversal rate is higher in the outcome-based condition. Right: the two parameters correlate positively across participants. The thin dotted line indicates the best-fitting regression line and the shaded area its 95% confidence interval. c Response repetition curves. Fraction of response repetitions (y axis) as a function of the evidence provided by the intervening sequence in favor of the previous response (x axis, expressed as logLR). Positive evidence indicates evidence consistent with the previous response, whereas negative evidence indicates evidence conflicting with the previous response. The thin dotted line indicates perfectly uncertain (null) evidence. Dots indicate the observed data (means ± SEM, n = 24 participants), whereas lines and shaded error bars indicate best-fitting sigmoid functions (means ± SEM, n = 24 participants). d Best-fitting parameters of sigmoid functions in the cue-based and outcome-based conditions. Left: the PSE is increased in the outcome-based condition, whereas the sensitivity to evidence is matched across conditions. Right: the two parameters correlate slightly negatively across participants. Significant effect at ***p < 0.001; n.s., a nonsignificant effect (paired two-sided t-tests, d.f. = 23, no correction for multiple comparisons). The error band indicates the 95% confidence interval for the regression line fitted using ordinary least squares. Source data are provided as a Source Data file.
