The Autocorrelated Bayesian Decision Maker: Mental Sampling Processes Capture Preference, Confidence and Response Time in Risky Choice

Modern models of preferential decision making have achieved substantial success in predicting choice behaviour both qualitatively and quantitatively, but often overlook related aspects such as response time or confidence. In parallel, recently proposed mental sampling frameworks have offered broad explanations for multiple aspects of behaviour including probability judgments, perceptual estimates and numerical forecasts, but have not yet been applied to preferential choice. Here, we seek to unite these lines of research by extending one such framework, the Autocorrelated Bayesian Sampler (ABS; Zhu et al., 2024), to decisions between alternatives which differ in the value of their potential outcomes. We present a new version of the ABS adapted to such decisions, the Autocorrelated Bayesian Decision Maker (ABDM), able to predict not just choices, but also reaction times and confidence in selections. We compare the predictions of the ABDM against extant choice and response time data in risky choice tasks, finding strong correspondence in both. We also contrast the new model against leading explanations for behaviour in such tasks, showing the ABDM is competitive in choice and ahead in response times. We thus propose this model as a broader approach to decision making capable of explaining multiple measures from a common process.