ABSTRACT
Frequently outcomes of interest will take the form of several different events that have no natural order or ranking to them. For example, the choice of a political party in an election or the choice of career following graduation are all nominal categorical outcomes. This chapter covers common options for how to model and explain such outcomes. Using an example of the choice of a health insurance product by company employees, the chapter starts with the concept of stratified binomial models which model each category as a binary outcome versus the other categories. The chapter then proceeds to describe the multinomial approach, where the likelihood or relative risk of each category is determined against a common reference category. The chapter covers the interpretation of the coefficients of a multinomial model relative to its reference, how to change the reference and how to simplify the model without risking the loss of important variables.
