Logistic Generalized Regression (LGREG) Estimator in Cluster Samples

• Activity Number: 143
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Timothy Kennel*+
• Companies: U.S. Census Bureau
• Keywords: Generalized Regression Estimator ; Logistic Regression ; Calibration ; Clustered Sample Designs ; GREG

Abstract

Statistical models are often used to assist estimation of descriptive statistics from surveys. Perhaps the most common estimator is the Generalized Regression Estimator (GREG) which is design consistent, uses a linear assisting model, and results in a set of calibrated weights. However, when the variable of interest is binary, binomial, or multinomial, it may be more appropriate to use a logistic assisting model instead of the standard linear model. In this paper we develop point and variance estimators for totals of finite population characteristics from a clustered sample assisted by a logistic regression model. Using a national Public Use Microdata set we compare the design-based properties of the new estimator to the GREG and the Horvitz-Thompson estimator under two clustered sample designs.