Parametrically Detailed Computational Analyses of Human Foraging Behavior

The overall goal of this project is the development and validation of parametric mathematical assays of human decision-making, based on an online information-foraging task (WebSurf). Decisional impairments in general are common in mental illnesses, but the exact pattern of deficits varies within and between diagnostic categories. Those deficits often involve multiple decision-making systems and the interactions between those systems. For instance, patients with obsessive-compulsive disorders rely overly on habitual/procedural action (leading to ritualizing behavior), but also show impairment in change-of-mind systems (inability to interrupt rituals) and deliberation (?analysis paralysis? in the face of uncertainty and a chance of negative outcomes). A major challenge in computational psychiatry is the need for tasks/paradigms that measure these multi- system dysfunctions, including interactions between systems. A further need is tasks that are viable for clinical settings, i.e. that are valid for repeated-measures use, sensitive to clinical-level impairment, and usable without highly trained experimenters present. We propose to address these needs with WebSurf, an information-foraging task developed by co-PIs MacDonald and Redish. These investigators and their colleagues have used WebSurf (and its rodent version, Restaurant Row) to demonstrate a common ?sunk costs? fallacy across rodents and humans, to identify the neural basis of regret, and to quantify differences in rule-based decision making in patients with eating disorders. Those studies have demonstrated WebSurf?s general utility as a cross-species paradigm and shown the richness of parametric descriptions that can be extracted from task behavior. They have also identified difficulties with the base version of the task, including needs for greater subject engagement and higher trial counts. As importantly, although Restaurant Row appears to elicit stable day-to-day behavior in mice, we do not yet know if the same is true for humans. We will close these gaps in task validation by assessing the performance of multiple variants using Amazon?s Mechanical Turk platform. Data from those variants, as well as ongoing data collection with the baseline task in our psychiatric clinics, will validate newer and more robust approaches to decision parameter estimation (Aim 1), grounded in Bayesian hierarchical modeling. They will demonstrate repeated- measures stability (Aim 2) and ability to describe variation between and within clinical populations (Aim 3). Executing these Aims will build on WebSurf?s success as a (reverse) translatable experimental paradigm, demonstrating a tool for clinical computational psychiatry. Our team?s broad experience includes computational science, experimental psychology and neuroscience, and clinical psychiatry, making us well-suited both to perform the Aims and apply the results in future psychiatric neuroscience studies.