Multinomial and ordinal logistic regression

References

Cite

Share

Abstract

Standard logistic regression handles binary outcomes such as disease/no disease. Researchers sometimes collapse outcomes with more than two groups into binary variables to enable use of this familiar model. However, this practice risks losing information and statistical power. In fact, logistic regression is not limited to binary outcomes: Multinomial logistic regression can handle outcomes with more than two groups; and ordinal logistic regression can handle outcomes with more than two ordered groups. This article aims to demystify these two models.