Attention-Entropy Random Utility: Endogenous Attention and Context Effects in Discrete Choice

This paper introduces the attention-entropy random utility (AERU) model, a behavioral model of discrete choice in which a decision-maker endogenously allocates attention across subsets of attributes in order to increase subjective confidence by reducing ex post choice uncertainty, and subsequently chooses an option based solely on the attended information. By endogenizing attention, the decision problem is reformulated from "which alternative to choose" to "which informational cues to process," with the observed choice emerging as the outcome of this attentional allocation. The AERU framework nests random utility model (RUM)-like behavior under transparent conditions, yet it is not restricted by Luce's independence of irrelevant alternatives (IIA), order-independence, or regularity. This flexibility enables AERU to capture key context effects in a disciplined manner and to generate sharp, testable predictions regarding the conditions for each context effect. From an empirical standpoint, AERU preserves the parsimony of the multinomial logit, requiring only a single additional attention parameter. Employing a scalable estimation procedure based on block coordinate ascent combined with a quasi-Newton method, I provide results from computational experiments demonstrating that AERU can produce better in-sample and out-of-sample predictions. Overall, AERU provides a flexible, parsimonious, and interpretable model of boundedly rational choice with a clear behavioral foundation and implications for context effects.