Fig. 4 |. Memory reactivation decoding results.

a, Blockwise decoding results showing reactivation (y axis) as a function of age plotted continuously (x axis). Age significantly improved the model fit beyond the base model including only repetition (AIC_base = 24.24, AIC_age = 23.31, χ²(9) = 18.93, P = 0.03); in the better-fitting model with age, there was a significant age-by-repetition interaction (χ²(6) = 16.55, P = 0.011). Repetition one serves as a baseline, as AB study occurs prior to encountering any overlapping (BC) face or scene content, and is not plotted here for the sake of simplicity. The coloured inverted triangles along the x axis indicate age points at which model predictions are shown in the subsequent panels. Adults maintain reactivation of related content across encoding repetitions, whereas adolescents show reactivation only on repetition two (light purple line; not significant on repetition three, shown in dark purple). Children show no significant evidence of reactivation. b, Model predictions from a visualized at four age points (10, 15, 20 and 25 years) across all three repetitions (light to dark). The asterisks denote age points and repetitions for which the model predictions are significantly above 0, indicating reliable reactivation according to the better-fitting model. The plots in a and b, represent 252 observations across N = 84 participants. c, Applying our classifier to individual trials rather than blocks yielded reactivation scores associated with each repetition of each specific pair. We found evidence for developmental differences (ages are shown by line colour and correspond to b) in the direction of the within-participant relationship between reactivation change from repetitions two to three (x axis) and subsequent inference performance (y axis; interaction: χ²(3) = 8.13, P = 0.043). Specifically, while adults (coral and orange) were more likely to get an inference decision correct when reactivation increased from repetitions two to three (>0 on the x axis), children (10 yr, purple) showed the opposite pattern—reactivation decreases (<0 on the x axis) were associated with a greater probability of correct inference at younger ages. There was also a main effect of age, such that inference accuracy was greater for older than for younger participants (χ²(3) = 20.60, P < 0.001). The figure displays model predictions at specific, user-defined age points; however, within the model, age was treated continuously. The plot in c represents 2,528 observations across N = 84 participants.