Choice Modeling With Context Effects: Generalization for Observational Data

Abstract

Established procedures of analyzing the effect of context on choice consider simple, compact environments in laboratory settings. However, these approaches severely limit the study of context effects and, as a consequence, the applicability of their findings. In this paper, we generalize existing approaches in modeling choice with the aim of developing a toolbox for the analysis of observational data. We consider three main context measures: attraction, compromise, and similarity. The proposed methodology hinges on ex ante calculation of context features for every alternative in multioption, multiattribute choice sets. This approach minimizes the computational complications of estimating the resulting choice model. The proposed approach is applied to air traveler choice data using an extensive observational dataset. This yields the first examination of all three context effects simultaneously in a large observational dataset. We discuss the consequences of product (re)design based on the results of the empirical exercise to showcase the potential use of the developed methodology in managerial practice.