Hierarchical Bayesian Modeling of Individual Differences in Texture Discrimination

The ability to identify a target texture in a visual display is a basic capability of the human visual system. Traditionally, as with much psychophysical modeling in the cognitive sciences, models of texture discrimination have been fit to individual subject data. By estimating model parameters independently, this approach emphasizes individual differences, and does not model the similarities between subjects. We consider alternative assumptions about individual differences in texture discrimination, using a standard model and previously studied data. In particular, we show how a hierarchical Bayesian approach can capture both the similarities and differences over subjects in a theoretically satisfying way. We also show that the hierarchical Bayesian approach has a number of methodological advantages over existing analyses, including improving parameter estimation when data are either sparse or missing, and improving model predictions when generalizing to new or different stimuli