Statistical analysis of ordinal categorical status after therapies.

Appropriate statistical analysis of clinical data based on ordinal categorical outcome scales is discussed. Chi-square analysis is inappropriate for testing the superiority of one therapy over another when outcome is ordinal categorical. Emphasized is estimation of clinically informative effect sizes after statistical significance has been attained with the Mann-Whitney U test. Little-used informative measures of effect size are related to estimation of the probability that a client given one therapy will have an outcome superior to that of a client given another therapy.