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

Abstract

This study examined how the values of three distinct R^2 analogs used in ordinal regression vary as  functions of number of predictors, choice of link function, and distribution type. Results indicated that Nagelkerke’s (1991) index most closely approximated OLS R^2, and the use of a complementary log-log  link function resulted in particularly variant values of the R^2 analogs.