Analysis of correlated ordered categorical data with rank measures of association for responses with ties

Rank measures of association are used to describe the variation among a set of groups for ordinal response data from two or more conditions in a repeated measures data structure. Linear models are fit to the respective measures by the ordinary least squares method for the full model and by the weighted least squares method for the reduced models. The resulting estimated parameters can be used for hypothesis tests or confidence intervals for comparisons among groups and for the evaluation of group × condition interaction. Ways of having these methods account for ties in rankings are also discussed. An example of a longitudinal study for comparison of four groups (two centers × two treatments) is provided.