Polytomous Outcome Regressions

Abstract

In clinical research outcome categories like disease severity levels are very common, and statistics has great difficulty to analyze categories instead of continuously measured outcomes. Polytomous outcome regressions are regressions with categorical rather than continuous outcomes. Five methods of analysis are reviewed in this chapter. With multinomial regression the outcome categories are equally present in the data. With ordinal regression one or two outcome categories are underpresented. With negative binomial and Poisson regressions data are assessed as multivariate models with multiple dummy outcome variables. Random intercept regression is like multinomial but provides better power. With logit loglinear regression first and second order interactions of the predictors on the outcome categories are assessed. With hierarchical loglinear regression third and fourth order interactions of the predictors on the outcome categories are so.