2026 Volume 33 Issue 3 Pages 420-435
Decisions from experience (DfE) show systematic behavioral patterns that differ from decisions from description (DfD), a phenomenon known as the description–experience gap. In DfE, probability inference plays a central role, however the normative Bayesian inference model assumes exchangeability of experiences and does not account for “recency”, the tendency to overweight recent samples. We propose a forgetting Bayesian inference model that introduces temporal discounting into the likelihood, to capture recency in probability inference. We also integrate this model with Cumulative Prospect Theory (CPT), to examine heterogeneity in the probability weighting function. Bayesian model comparison in a repeated decision-making task with 65 university students, indicated that the forgetting Bayesian inference model achieved higher predictive power than conventional models. Allowing heterogeneity in CPT’s probability weighting function yielded the best fit. These findings show that incorporating forgetting and heterogeneity provides a more refined account of DfE.
