A Modified Cochran-Friedman Test with Missing Observations and Ordered Categorical Data

A modification of Friedman's statistic is proposed in order to extend Cochran's test to situations with more than two ordered categories of response. The modification arises from a need to handle missing observations in some categories. A Scheffe-type multiple comparison procedure that uses the asymptotic chi square approximation is also given. A motivating example which compares multiple evaluation criteria on breast-cancer data is presented.