Proceedings of the Annual Meeting of the Cognitive Science Society

Planning involves a process of considering future states before acting. To understand this process, researchers typically infer planning algorithms by fitting computational models to choices. However, different planning models often predict the same choices, despite relying on different computations. Reaction time can help distinguish among models, since different computations produce different temporal signatures. However, incorporating reaction time into fitting is challenging because analytical likelihoods are typically unavailable. Here we propose a likelihood-free method to estimate the density for choices and reaction times in multi-stage decision making. We validate the method through comparisons with analytical solutions, parameter recovery, and showing robust estimates relative to distribution-free and summary statistic approaches. Through a new human experiment and fitting evidence accumulation models from Solway and Botvinick (2015), we demonstrate that modeling the full distribution is important to explain human behavior. Overall, our method is a valuable tool for modeling reaction times in multi-stage decision-making.