Masters Oral Defense: "Generalized ML Approach to Modeling Restricted Categorical Choice for the Case of The Spatial Random Utility"

Elena Labzina, Washington University in Saint Louis

Abstract: Multinomial logistic regression model (MNL) is a powerful and easily tractable way for measuring the probabilistic impact of input variables on individual categorical choices. Importantly for the case of arbitrary data, compared to another classical model, multinomial probit (MNP), MNL has better convergence properties. Unfortunately, MNL assumes that all subjects of the study have the same choice sets, while in political science and economics this condition is frequently violated. Probably, the most graphical example of varying choice sets (VCS) is partially contested elections. Furthermore, the MNL explicitly implies the Independence of the Irregular Alternatives (IIA) assumption requiring i.i.d errors, which contrasts MNL from the multinomial probit (MNP) and mixed logit (MXL) models. In the case of VCS, the errors are not i.i.d and IIA is clearly violated. However, neither MNP nor MXL allows estimating particular parameters for distinct choice sets that is critical if the aim is to compare the selection process conditional on the choice restrictions. This text argues that the MNL still proposes the best opportunity to model categorical choice given VCS. For that, it advances the theory of MNL adjusting this classical model for the case of VCS. Second, the paper proposes a way to calculate and evaluate the model that poses minimal data restrictions. Finally, this research provides several examples of the model's application.  

Host: Jose Figueroa-Lopez