Categorical Data Analysis Using The SAS® System, 2nd Edition

15.13. Using GEE to Fit a Partial Proportional Odds Model: Univariate Outcome

Chapter 9, “Logistic Regression II: Polytomous Response,” describes the use of the proportional odds model for response outcomes that are ordinal. Instead of modeling logits as in logistic regression for a dichotomous response, you model cumulative logits. However, sometimes you have data with an ordinal outcome for which the proportional odds assumption doesn’t apply. You can use the GEE approach to fit a partial proportional odds model in which you assume proportional odds for some of the explanatory variables but not others. You form multiple response outcomes from your univariate outcome by forming logits corresponding to the different cutpoints of the ordinal ...