Salience-Biased Nested Logit

Abstract

This paper introduces a two-level nested stochastic choice model in which nest probabilities are driven by salience. A category comprises alternatives that might be costly to gather information about, and we implicitly assume that market leaders are easier to familiarize oneself with. By learning about those alternatives more affordably, the items with the highest probability within each category become their respective saliences when selecting the category. Formally, a partition of the available options defines the collection of nests (categories), while a Luce function assigns weights to all alternatives. These two components represent the salience-biased nested logit (SBNL) model, which differs from the standard nested logit (NL) model primarily because the nest probabilities are determined solely by the highest probability within each category, which defines the corresponding salient alternative in our approach. Like the NL model, the Luce model is applicable within categories. While SBNL usually violates regularity, which leads to a form of market leader effect, we can develop a specific case of our model within the conventional random utility framework and demonstrate its broad applicability in practice under a standard parametric specification for utility. This results in a well-specified method for estimating the model’s parameters using individual or aggregate market data. It serves as an additional tool for analyzing market shares and clarifying how price elasticities may display different patterns according to marginal effects on demand stemming from variations in the prices of market share leaders (the salient ones) compared to price changes in non-leader alternatives.