ABSTRACT
This chapter discusses the local, cumulative, and continuation odds ratios as association measures used to account for the ordering. It presents some models for ordinal data that use the logit link. The chapter explains methods to test for effects and model fit. It also discusses methods for confidence interval estimation. The chapter illustrates the general result that tests that account for ordered data through constraints on the alternative hypothesis have higher power than tests that do not put any constraint on the alternative hypothesis. It outlines models for cumulative logits with proportional odds. The chapter suggests that the score test for a proportional odds model, which is equivalent to the midrank version of the Wilcoxon-Mann-Whitney test.
