Tied Rank Analyses and the Grouped Continuous Model for Ordered Categorial Data

摘要

Using a marginal likelihood based on ranks adapted for ties, it is shown that inference based on ranks is asymptotically equivalent to inference based on a grouped continuous model when data are the tied, grouped continuous or the ordered categorical type. Approximations which make the rank analysis practical to use are given. Two numerical examples, involving a two-sample problem and a regression problem, are presented. Results are similar, but rank analysis is preferable to the group continuous model for moderate to large number of categories and small to moderate sample sizes.