Properties of Empirical Bayes Estimators for Evaluating Questionnaire Data in Epidemiology Studies

• Activity Number: 237
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jaya M. Satagopan*+ and Qin Zhou and Susan Oliveria and Stephen Dusza and Martin Weinstock and Marianne Berwick and Allan Halpern
• Companies: Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Memorial Sloan-Kettering Cancer Center and Brown University and University of New Mexico and Memorial Sloan-Kettering Cancer Center
• Keywords: questionnaire ; regression ; dimension reduction ; empirical Bayes ; type I error ; bias

Abstract

Cancer epidemiology studies use questionnaires to examine environmental risk factors for disease using multiple related questions. A goal of these studies is to identify the individual questions associated with disease. Regressing the outcome on the responses to individual questions may lead to counter-intuitive results due to multi-collinearity issues. An alternative is a dimension reduction analysis, where pre-defined composite summaries - i.e. linear combinations of the individual questions - are used, and effects of the individual questions are elicited in a suitable manner from this analysis. But, this approach can lead to inflated type I error and bias if the composites are defined incorrectly. To address this issue, we consider a class of empirical Bayes estimators that combines estimates from the two analyses. We investigate their properties using simulations and real data.