Clustered Odds Ratio

• Activity Number: 144
• Type: Contributed
• Date/Time: Monday, August 7, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Wanjie Sun*+ and Patricia Cleary and John M. Lachin
• Companies: The George Washington University and The George Washington University and The George Washington University
• Keywords: clustered odds ratio ; ALR ; GEE ; bias ; efficiency

Abstract

Clustered binary data occur frequently in many fields, especially in genetics where we try to estimate sib-sib Odds Ratio (OR) in presence of multiple siblings. Standard methods that ignore the within-cluster correlation underestimate the variance of OR, therefore, inflate the type-1 error rate. In this paper we compared the bias and efficiency of four methods: 1) ordinary OR, 2) clustered OR from Rao & Scott (1992), 3) clustered OR from first-order GEE (Liang & Zegar, 1986), and 4) clustered OR from second-order GEE, Alternative Logistic Regression (ALR) (Carey etc.1993). Simulation shows that ALR is the most consistent, and most efficient of the four. And the efficiency improves with growing cluster size n. So besides the fact that ALR models the main effect (y/x relationship) and association effect (OR/x relationship) simultaneously, it is also the most robust method of the four.