Fitting Proportional Odds Models for Complex Sample Survey Data with SAS, IBM SPSS, Stata, and R

Abstract

An ordinal logistic regression model with complex sampling designs is different from a conventional proportional odds model since the former needs to take weights and design effects in account. While general-purpose statistical packages, such as SAS, IBM SPSS, Stata, and R, are all capable of fitting proportional odds models with complex survey data, they may use different techniques to estimate the models and have different features. The purpose of this article was to illustrate the use of SAS, IBM SPSS, Stata, and R to fit proportional odds models with complex survey data, and to compare the features and results for fitting the models using SAS PROC SURVEYLOGISTIC, IBM SPSS CSORDINAL, Stata svy: ologit, and R survey. The linearization method was used to estimate the sampling variance.