An Examination of Ordinal Regression Goodness-of-Fit Indices Under Varied Sample Conditions and Link Functions

• In Event: Advanced Topics in Multiple Regression Analysis

Abstract

The present study examined three commonly-used pseudo R-squared values used in ordinal regression. Effects of (1) distinct distributional conditions (uniform distribution and positively/negatively skewed) and (2) distinct specified link functions indicated that the use of the complementary log-log link function consistently resulted in larger pseudo R-squared values than other link functions, and that link functions other than the complementary log-log function were relatively invariant to distributional characteristics of the ordinal outcome.

Authors

• Thomas J. Smith, Northern Illinois University

• Cornelius McKenna, Kishwaukee College