Essays On Discrete Choice Models
Doctor of Philosophy
Discrete choice models have long been a cornerstone of marketing research. The beginning of what is currently known as the Multinomial Probit Model was published by Thurstone in 1927. Luce's choice axiom in 1959 is a landmark in the development of the Multinomial Logit Model that was popularized by McFadden in 1973 when he showed that the model could be derived from the primitives of random utility maximization. Tversky introduced the Elimination By Aspects (EBA) model in 1972 to provide a more general theory of choice. These and developments of numerous researchers have been further propelled by constantly decreasing computing costs. The result is that discrete choice analysis research has continued to flourish. Surprisingly, while discrete choice models have found applicability in a wide variety of academic and industrial research, several implications of the underlying theories have not been tested. In our first essay, we examine how individual consumers respond to choice tasks when asked for the most preferred choice, least preferred choice, or rank ordered choices. We find that the mathematical and statistical implications of qualitative choice models do not fully explain differences in response for the assigned conditions. In our second essay, we examine how the EBA model performs when predicting choices for individuals relative to the more commonly used Logit and Probit models. We find that EBA outperforms both Probit and Logit for in-sample datasets and provides comparable performance to Probit on holdout data. In our third essay, we examine a unique dataset on choice data collected from practicing Venture Capitalists. The analysis examines the influence of branding on Venture Capitalists during deal screening. By developing a new methodology to incorporate demographic information with respondent choices, we find the surprising result of a differential impact of branding based on the level of responsibility of the respondent. Each essay demonstrates that while discrete choice analysis is well-studied field, more research is needed to further refine and improve our understanding of how choices are made.