Statistical modeling for human behavioral task data in computational psychiatry

Abstract
Human behavioral tasks offer a window into reward learning, attention, and decision-making in psychiatric disorders. Translating observed choices and response times into interpretable cognitive characteristics, however, requires models that account for individual differences and changing strategies. Motivated by major depressive disorder, this talk presents a progression of statistical approaches for understanding these behavioral processes. I will discuss how reinforcement learning models can be extended to distinguish engaged decision-making from lapses, how drift-diffusion models incorporate response times and the speed–accuracy trade-off, and how shared latent factors connect information across multiple tasks. Applications to the probabilistic reward and Flanker tasks illustrate how these approaches reveal differences in engagement and decision dynamics between individuals with depression and healthy controls. I will conclude with opportunities for integrating behavioral tasks and neuroimaging data. Together, these developments highlight the role of statistical modeling in characterizing cognitive heterogeneity and advancing understanding of mental disorders through computational psychiatry.