Fig. 5. The impact of increasing number of trials on mean-level model-based estimates, its reliability and their association with individual differences (N = 716).

a Mean MBI estimated with (i) cumulative trials i.e., increasing trial numbers sequentially by 25 at a time, until 300 trials per participant was reached, and (ii) binned trials i.e., bins of 25 trials sampled sequentially through the task, in chronological order. Mean-level MBI decreased with more trials. Error bars reflect standard errors of the mean. b Split-half reliability co-efficient as a function of increasing trial number using (i) cumulative trials and (ii) binned trials. Using the cumulative trials, reliability of the MB estimate increased with additional trial collected per participant. Error bars reflect 95% confidence intervals c Model-based associations with compulsivity, age, gender and education using (i) cumulative trials and (ii) binned trials. Increasing the number of trials per participant did not significantly increase the association between model-based planning and individual differences. No statistical significant difference was found between model-based associations with compulsivity when estimated at participants’ first 25 vs 300 trials. Model-based associations with age and education became stronger with the addition of trials while the association between model-based and gender reduced. Data from the binned analysis demonstrated that this effect is in part driven by stronger signal in earlier versus later trials. d We repeated this analysis using a publicly available dataset of N = 1413 individuals who completed the traditional two-step task (200 trials). Here we found (i) the association between model-based planning and individual differences increased as trials collected per participant increased and (ii) this effect is driven by later trials compared to earlier trials. For c and d, error bars reflect standard errors of the mean.