|
Abstract:
|
When we are interested in comparing two independent groups that have ordered categorical measurements, we can use a measure of stochastic superiority to describe the effect size. This measure embodies the relative size of outcomes, rather than actual magnitude. We investigate three confidence interval methods: the Wald-type, score-type, and likelihood ratio-based, which are based on the Mann-Whitney statistic allowing ties. These confidence intervals are obtained by considering data from two independent multinomial distributions. We consider an unrestricted and a parametric model, depending on whether there exists a relationship between cell probabilities. Simulation studies show it is better to use an unrestricted model under several conditions.
|
