Logistic Regression and Generalized Linear Models for Binary Survey Variables
This chapter introduces the fundamental concepts of regression for categorical data based on Generalized Linear Models (GLMs) that are important to understand for effective application of the techniques to survey data. It provides a systematic review of the stages in fitting the logistic regression model to complex sample survey data. The chapter presents example logistic regression analyses based on the Health and Retirement Study (HRS) and National Comorbidity Survey Replication (NCS-R) data that illustrate typical applications of the model-building steps to actual survey data sets. It focuses on the important underlying concepts of GLMs as they apply to logistic regression and probit regression. The two most common link functions used to model binary survey variables are the logit and the probit. The chapter concludes with a comparative application of the logistic, probit, and complementary-log-log (C-L-L) regression techniques to the problem of modeling the probability of alcohol dependency in US adults.