Fig. 5: Results from Experiment 2 replicated key findings from Experiment 1.

A Over the course of each block, participants (n = 34 children; n = 39 adults) learned to make more optimal responses to stimuli in both the category-predictive and exemplar-predictive conditions, though performance was better in category-predictive relative to exemplar-predictive blocks. B In the category-predictive condition, participants increasingly generalized learned category responses to respond optimally to novel stimuli. C Participants across age groups demonstrated higher category-level choice weights in category-predictive blocks. Choice-weight magnitudes reflect normally distributed parameter estimates, which were exponentiated within each model. Thus, negative values reflect low, positive choice weights. D Participants demonstrated better memory for stimuli from the exemplar-predictive versus category-predictive blocks of the task. E Participants who earned the most points in the exemplar-predictive blocks also demonstrated better memory for exemplar-level information encountered in those blocks. Participants are binned into equal-sized performance groups based on the number of points earned in each block condition for visualization purposes only. In panels (A–E), horizontal lines reflect individual participant means. The points and error bars indicate group means ± 1 SEM. Statistical analyses were conducted using mixed-effects models that assessed effects within participants while accounting for variation across them. F Participants who weighted exemplar-level information most strongly during learning also demonstrated better category and better exemplar memory. The strength of this relation between learning and memory increased with increasing age. Participants who weighted category-level information most strongly demonstrated better category memory but not better exemplar memory. The plots depict marginal effects from linear-mixed-effects models examining the effects of age group, block condition, specificity (exemplar and category), choice weight magnitude (exemplar or category), and their interactions on memory performance, as indexed by AUC. The shaded regions depict 95% confidence intervals.