Fig. 2: Behavioral signatures of dynamic belief updating.
From: Individual differences in belief updating and phasic arousal are related to psychosis proneness
a Choice accuracies of n = 90 human participants (gray bars), with accuracies of the ideal observer (navy), perfect integrator (green) and deciding based on last sample heuristic (green) given the same stimulus sequences shown for reference. Red dots, accuracies of normative model fits. b Best-fitting parameters of normative model fits for individual participants (gray circles) and group means (black horizontal bars). c Time-resolved weight of evidence on choice, derived by regressing participants’ single-trial choices onto evidence strengths (log-likelihood ratios, LLRs) per sample position within each trial. d, e Modulation of evidence weighting, estimated via LLR interaction terms in the regression model, by two variables that the normative model prescribes sensitivity to: change-point probability (CPP, d), which captures the degree to which a new evidence sample conflicts with the current belief; and uncertainty (-|ψ | , e) in the belief prior to observing each new sample. In c–e: black, mean and s.e.m. (shaded areas) of participants’ data; significance bars, time points where weights differ from zero (P < 0.05, two-tailed cluster-based permutation test; evidence weighting and largest CPP and -|ψ| modulation clusters, P < 0.0001). Red shadings, mean +/− s.e.m. of normative model fits. Gray thin lines, individual participants. n = 90 participants for calculation of means and s.e.m. in all panels.
