Ordinal Logistic Regression as an alternative analysis strategy for the comparison of two independent samples

B. BILSKI
, B. DERRICK
, D. TOHER
, P. WHITE

Keywords

Ordinal logistic regression, Mann Whitney test, Welch t-test

Abstract

The two group between subjects design is pervasive with analyses often performed using the Mann Whitney Rank Sum test or using the Welch variant of the t-test. Using simulation it is shown that a dummy variable ordinal logistic regression (OLR) model provides an alternative analysis strategy for n >= 16 per group retaining Type I error robustness for both continuous and tied data. OLR is demonstrated to have comparable power to the Mann Whitney test under non-normal alternatives and with comparable power to the Welch t-test for normal distributed data. This opens the possibility for the ordinal logistic model to be a general analysis technique for higher order designs.