When does the drift diffusion model fail in risky choice? A quantitative test of response time predictions

We tested whether the drift diffusion model (DDM) can account for response times (RTs) in risky choice—an application where the model’s fit to data has received limited scrutiny. We compiled 14 publicly available datasets (N(participants) = 1,388, N(choices) = 199,157). We fitted a hierarchical integrative DDM with drift rates governed by cumulative prospect theory (CPT-DDM). A core DDM prediction is a negative relationship between choice strength and RT. Datasets varied considerably in the degree to which they showed this pattern, and model fit varied accordingly. Parameter estimates from the DDM-CPT model showed high concordance with estimates from a pure CPT model ignoring response times (concordance correlations: 0.73–0.87). This indicates that the two models infer similar underlying preferences, even when the CPT-DDM provides a poor account of the RT data. The general fit of the CPT-DDM was rather poor in experiments using accept-reject choices and 2AFC tasks with simple lotteries. On the other hand, in experiments where participants chose between two lotteries with more than one non-zero outcome, the CPT-DDM provides a satisfying fit to observed data. We conclude that while the CPT-DDM can capture choice–RT relationships in some settings, task features matter critically. A more comprehensive model integrating information acquisition and decision dynamics is needed to fully account for RTs in risky choice.