Attention Constraints and Learning in Categories

When different stimuli belong to the same category, learning about their attributes should be guided by this categorical structure. Here, we demonstrate how an adaptive response to attention constraints can bias learning toward shared qualities and away from individual differences. In three preregistered experiments using an information sampling paradigm with mousetracking, we find that people preferentially attend to information at the category level when idiosyncratic variation is low, when time constraints are more severe, and when the category contains more members. While attention is more diffuse across all information sources than predicted by Bayesian theory, there are signs of convergence toward this optimal benchmark with experience. Our results thus indicate a novel way in which a focus on categories can be driven by rational principles.