Conditional Logistic Regression Estimators for Ordinal or Multinomial Outcomes and Complex Survey Data

• Activity Number: 560
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Babette A Brumback*+ and Zhulin He and Hao Zheng and Amy Dailey and Zhuangyu Cai
• Companies: University of Florida and University of Florida and University of Florida and Gettysburg College and University of Florida
• Keywords: weighted composite conditional likelihood ; complex survey data ; proportional odds model ; baseline category logit model ; conditional logistic regression ; clustered data ; weighted composite conditional likelihood ; complex survey data ; proportional odds model ; baseline category logit model ; conditional logistic regression ; clustered data

Abstract

In order to adjust individual-level covariate effects for confounding due to unmeasured neighborhood characteristics, we extend conditional logistic regression estimators for use with ordinal or multinomial outcomes and complex survey data. For ordinal outcomes we use a proportional odds model, whereas for multinomial outcomes we use a baseline category logit model; in both models we include a neighborhood-specific intercept. Our estimators are consistent even when the within-neighborhood sample sizes are small and the sampling bias is strongly informative. The key to this consistency is our use of sampling design joint probabilities for each within-neighborhood pair. The estimators and asymptotic sampling distributions we present can be computed using standard logistic regression software for complex survey data, such as SAS PROC SURVEYLOGISTIC. We validate the methods using a simulation study, and we apply the methods to data from the 2008 Florida Behavioral Risk Factor Surveillance System survey, in order to investigate disparities in frequency of dental cleaning both unadjusted and adjusted for confounding by neighborhood.