Interpretation of individual differences in computational neuroscience using a latent input approach

Computational neuroscience offers a valuable opportunity to understand the neural mechanisms underlying behavior. However, interpreting individual differences in these mechanisms, such as developmental differences, is less straightforward. We illustrate this challenge through studies that examine individual differences in reinforcement learning. In these studies, a computational model generates an individual-specific prediction error regressor to model activity in a brain region of interest. Individual differences in the resulting regression weight are typically interpreted as individual differences in neural coding. We first demonstrate that the absence of individual differences in neural coding is not problematic, as such differences are already captured in the individual specific regressor. We then review that the presence of individual differences is typically interpreted as individual differences in the use of brain resources. However, through simulations, we illustrate that these differences could also stem from other factors such as the standardization of the prediction error, individual differences in brain networks outside the region of interest, individual differences in the duration of the prediction error response, individual differences in outcome valuation, and in overlooked individual differences in computational model parameters or the type of computational model. To clarify these interpretations, we provide several recommendations. In this manner we aim to advance the understanding and interpretation of individual differences in computational neuroscience.