Figure 2.

A graphical Bayesian approach for parameter estimation of the reinforcement learning model for the dynamic reward task. In this graphical model, nodes represent variables of interest and arrows indicate dependencies among these variables. Note that shaded nodes represent observed variables. Specifically, R_{i, j−1} is the reward received by subject i in trial j – 1, and Ch_{i, j} is the observed choice of subject i in trial j. The parameters α_i and β_i represent the learning rate and choice perseveration for subject i, respectively. Each parameter was assumed from a normally distributed group-level population with respective means and standard deviations. In our implementation, both μ_α and σ_α were assigned an (non-informative) uniform distribution between 0 and 1 for the prior. For β, a uniform prior between 0 and 10 was assigned to μ_β. For σ_β, we assigned a uniform prior between 0 and 5.