Figure 3. The influence of decision confidence on subsequent decision bound.

(A) Model-free measures of response caution on trial n+one as function of confidence on trial n: mean RT, accuracy, and their product. Inset, distribution of empirical and fitted RTs. (B) Model-based estimate of decision bound and drift rates on trial n+one as function of confidence on trial n. Distributions show the group posteriors over parameter estimates. In all panels, as well as in all subsequent figures, ‘delta’ on y-axis refers to deviation of dependent variable from its value in the high-confidence condition (i.e., centered on zero for high-confidence). For model fits, trials from the high-confidence condition served as reference so that parameter estimates reflected deviations from their value on high-confidence. Statistical significance is reflected in overlap between posterior distributions over parameter estimates (Materials and methods).

Figure 3—figure supplement 1. The influence of decision confidence on subsequent decision bounds and drift rates, separately for blocks with post-decisional evidence presentation (A–B) and a post-decisional blank (C–D).

Due to a lack of data in one of the cells, these analyses were performed on the data of 27 and 24 participants, respectively. Same conventions as in Figure 3.

Figure 3—figure supplement 2. Simple effects of confidence on trial_n and confidence on trial_n+2 on the product of subsequent RTs and accuracy as a model-free measure of decision bound (Experiment 1).

Same conventions as in Figure 3.

Figure 3—figure supplement 3. Simple effects of confidence on trial_n and confidence on trial_n+2 on decision bound (A) and drift rate (B) on trial_n+1 (Experiment 1).

Same conventions as in Figure 3.

Figure 3—figure supplement 4. Complementary approach controlling for slow drifts in performance (Experiment 1).

Separate models were fitted estimating the decision bound and drift on trial_n and i) we compared low confidence on trial_n to high confidence on trial_n for which trial_n+2 was a low confidence trial, and ii) we compared perceived errors on trial_n to high confidence trials on trial_n for which trial_n+2 was a perceived error. Note that low confidence and perceived errors cannot be compared directly because they stem from different models. These results remained unchanged when only including correct trials (there were not enough trials when only including errors).