Mantel-Haenszel Estimators for Complex Survey Data

• Activity Number: 250
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Babette Brumback*+ and Zhulin He
• Companies: University of Florida and University of Florida
• Keywords: odds ratio ; complex sampling designs ; conditional logistic regression ; sparse-data limiting models ; cluster sampling

Abstract

The SUDAAN CROSSTAB procedure provides an adaptation of the Mantel-Haenszel estimator of a common odds ratio for complex survey data. The estimator is consistent for large-strata limiting models, in which the number of strata remains fixed and the sample sizes within strata increase to infinity. However, we show via simulation and counterexamples that the estimator is inconsistent for sparse-data limiting models, in which the number of strata (more appropriately termed as clusters) increases to infinity but the sample sizes within cluster remain fixed. We also propose an alternative estimator that is consistent for sparse-data limiting models satisfying a positivity condition, but not for large-strata limiting models. We compare the estimators with each other and with recent adaptations of conditional logistic regression for complex survey data.