ABSTRACT
Ordered category or ordinal outcomes are arguably the most common outcomes in people analytics. Most measurement scales have an ordinal form, such as performance ratings, survey responses, academic grades and ranked preference voting systems. This chapter covers the most common regression technique for modeling and explaining ordinal outcomes: proportional odds logistic regression. A simple, easily interpretable model is derived and then applied to an example related to the levels of discipline of players in soccer games. As in previous chapters, time is spent on the design, simplification and interpretation of this model and on assessing its fit and goodness-of-fit. Importantly, the chapter also covers the key assumption that underlies the proportional odds model and shows ways to determine whether or not that assumption is violated for any given data set. The chapter also briefly reviews alternative approaches to modeling ordinal outcomes when the proportional odds assumption is violated.
