The Probabilistic Reward Task (PRT; Pizzagalli et al., 2005) challenges participants to make a difficult distinction between two very similar (and briefly presented) stimuli, and it rewards correction identification of one (“rich”) stimulus three times more often than correct identifications of the other (“lean”) stimulus. The asymmetric reinforcement induces a rich response bias, and several datasets (although not all) have reported weaker response biases in depressed adults. Consequently, weak response bias has been proposed as a behavioral marker of anhedonia, and the PRT appears in the RDoC matrix as a validated probe of Reward Learning. Furthermore, PRT data have been successfully fit with models of reinforcement learning (RL; Huys et al., 2013). These lines of work have established the PRT as a translational tool for studying RL in healthy and psychiatric samples. Recently, however, we described two aspects of behavior in the PRT that had gone unnoticed. First, the response bias effect is dependent on response time (RT): it is significantly stronger when RTs are fast versus short. This indicates that response-outcome associations may play as strong a role as stimulus-locked prediction errors in the PRT. Second, discriminability—accuracy in identifying the two stimuli—better predicts reward totals in the PRT than does response bias. This calls into question the idea that a weak bias is necessarily a sign of anhedonia, and it underlines the importance of perception. Both these findings can be explained by the drift diffusion model (DDM), and in this talk I will describe our efforts to use the DDM to model PRT data in healthy controls and adults with Major Depressive Disorder (and, if time allows, social anxiety disorder). Overall, this work demonstrates the complexity of behavior in the PRT, highlights the impact of depression on decision-making, and underscores the value of deep dives into behavioral data.

 

Dr. Daniel Dillon

Associate Professor at Harvard Medical School

Daniel Dillon – Modeling the Probabilistic Reward Task: Disentangling Effects of Depression on Decision-Making vs. Reinforcement Learning