Individual Differences in Attention During Category Learning

A central idea in many successful models of category learning-including the Generalized Context Model (GCM)-is that people selectively attend to those dimensions of stimuli that are relevant for dividing them into categories. We use the GCM to re-examine some previously analyzed category learning data, but extend the modeling to allow for individual differences. Our modeling suggests a very different psychological interpretation of the data from the standard account. Rather than concluding that people attend to both dimensions, because they are both relevant to the category structure, we conclude that it is possible there are two groups of people, both of whom attend to only one of the dimensions. We discuss the need to allow for individual differences in models of category learning, and argue for hierarchical mixture models as a way of achieving this flexibility in accounting for people's cognition.