Fitting Proportional Odds Models for Ordinal Response Variables in Educational Research: A Comparison of Multiple Packages in R

Abstract

Although researchers assume that software packages should produce the same results (e.g., model parameter estimates) for the same model, it is not always the case for ordinal regression models. The MASS, ordinal, rms, and VGAM packages in R may use different forms to express the proportional odds model (PO) for ordinal response variables and parameterize it differently, so researchers can easily become confused when they interpret the results. However, there are no studies specifically to address this issue. The purpose of the study was to investigate how to implement the PO model in educational research with multiple packages in R and compare the differences among these packages. Furthermore, it compares the results from the R packages and those from other general purpose software programs, SAS, SPSS Statistics, and Stata with the aim to help researchers to understand the performance of each package.