An Examination of Link Function Choice in Ordinal Regression Models

• In Event: 32.067 - Multiple Linear Regression: The General Linear Model SIG Paper Session: Special Adaptations of the General Linear Model to Best Reflect the Research Question

Abstract

The effect of link function choice in the fitting of ordinal regression models is assessed, using both an extant data set (Pew Research Center, 2014) and simulated data. Additionally, appropriateness of link, link specificity, variability in obtained results—with particular attention to palindromic invariance of indices of determination/discrimination—are examined.

Authors

• Thomas J. Smith, Northern Illinois University

• David A. Walker, Northern Illinois University

• Cornelius McKenna, Kishwaukee College