Discrete and Rank Ordered Choice Models with Heterogeneous Preferences and Consideration

Cornell University, Ithaca NY

Investigators

Abstract

Researchers use economic theory and data on individual choices to study what individuals want (preferences) in order to predict the welfare effects of policies. It is therefore essential that individuals’ preferences are correctly estimated. Existing methods of estimating preferences assume that the researcher observes the set of all alternatives and the characteristics of each alternative available to the individual, as well as the characteristics of all individuals making choices. However, the economics literature shows that individuals may not consider all alternatives available to them, or may not have access to all existing alternatives in the market in making choices. This mismatch between assumption and how people make choices may lead to mismeasurement of the estimates of individuals’ preferences hence welfare analyses and policy evaluations based on these estimated preferences may be wrong. This research project develops new methods to identify and estimate the distribution of preferences that recognize that there can be dependence between preferences and consideration sets, and that consideration sets may depend on individuals’ and alternatives’ characteristics. The methodology developed in this study will help to better infer people’s wants from data, and hence help policy makers design better policies to the benefit of consumers. The researchers will write and freely distribute computer software to implement the new methods, hence make it easier for others to use these novel methods. The results of this research project will improve prediction, improve policy making and therefore increase economic growth and improve the welfare of Americans. This research project develops a new class of models that combine preference heterogeneity and unobserved heterogeneity in consideration sets, with the goal of building discrete choice models that are: (i) applicable in many economic settings (e.g., school and college choice, insurance purchases, vehicle choice); (ii) identifiable under conditions that are similar to those required for identification of standard discrete choice models with homogeneous consideration sets; (iii) easy to compute; and (iv) flexible enough to explain a wide range of market phenomena that standard models have difficulty explaining. A novel feature of the class of models is the unrestricted dependence between heterogeneous consideration and heterogeneous preferences, as well as dependence between consideration and regressors. The contributions of this project include, establishing conditions for semi-nonparametric point identification on the new models; a maximum likelihood-based estimator that achieves significant computational gains relative to existing state-of-the-art models of discrete choice; new testing methods for model specification and a new measure of model's goodness of fit; and methods to carry out robust welfare analysis and study policy questions. The project will also develop and freely distribute software to implement the methods developed in this research. The results of this research project will improve prediction, improve policy making and therefore increase economic growth and improve the welfare of Americans. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.