<p>Parameter recovery results for the hierarchical Bayesian single-feature policy models.</p>

The dashed red line indicates the identity line corresponding to perfect recovery, while the solid black line shows the fitted regression line with 95% confidence intervals. Recovery performance was quantified using regression slopes and Pearson correlations between true and recovered parameter values. Across models, parameter recoveries demonstrated generally good calibration and moderate-to-strong correlations, indicating reliable recovery of both central tendency and inter-individual variability parameters.

Recovery analyses used the same hierarchical Bayesian logistic regression framework as the empirical analyses, including non-centered parameterization, weakly informative priors, and partial pooling across participants. Recovery fits were estimated using 4 Markov chains with 500 tuning and 500 posterior sampling iterations per chain (target_accept = 0.95). Convergence diagnostics produced frequent and effective sample size warnings for datasets generated by very simple single-feature policies, particularly when these datasets were fit using more flexible hierarchical models.

This likely reflects weak parameter identifiability and quasi-separable synthetic choice patterns arising from highly deterministic low-dimensional policies rather than failures of overall model recovery. Despite these warnings, the recovery analyses showed robust model discriminability and reliable recovery of the principal group-level parameters.</p> <p>(PDF)</p>