On the Welfare (Ir)Relevance of Two-Stage Models

TL;DR

This paper critically examines two-stage choice models, revealing significant identification issues that undermine their welfare analysis validity, and suggests that these problems stem from modeling approaches rather than the models themselves.

Contribution

It classifies two-stage models by their preference principles, identifies the root cause of their welfare-relevance issues, and demonstrates these problems with experimental data.

Findings

Half of the preferences under rational choice are not revealed in two-stage models.

Welfare-relevance of literature models is significantly lower than rational choice.

Issues stem from modeling the first stage, not the two-stage approach itself.

Abstract

In a two-stage model of choice a decision maker first shortlists a given menu and then applies her preferences. We show that a sizeable class of these models run into significant issues in terms of identification of preferences (welfare-relevance) and thus cannot be used for welfare analysis. We classify these models by their revealed preference principles and expose the principle that we deem to be the root of their identification issue. Taking our analysis to an experimental data, we observe that half of the alternatives that are revealed preferred to another under rational choice are left revealed preferred to nothing for any member of this class of models. Furthermore, the welfare-relevance of the specific models established in the literature are much worse. The model with the highest welfare-relevance produces a revealed preference relation with the average density of 2% (1 out of 45 possible comparisons revealed), while rational choice does 63% (28 out of 45 possible comparisons). We argue that the issue is not an inherent feature of two-stage models, and rather lies in the approach with which the first stage is modelled in the literature.