Online Program Home
My Program

Abstract Details

Activity Number: 296 - SPEED: Biometrics - Methods and Application, Part 1
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
Sponsor: Biometrics Section
Abstract #305346 Presentation
Title: Relative Risk Estimation in Clustered/Longitudinal Data Using Generalized Estimating Equations (GEE)
Author(s): Chao Zhu* and David W Hosmer and Jim Stankovich and Karen Wills and Leigh Blizzard
Companies: Menzies Institute for Medical Research, University of Tasmania and University of Vermont and School of Medicine, University of Tasmania, Central Clinical School, Monash University and Menzies Institute for Medical Research, University of Tasmania and Menzies Institute for Medical Research, University of Tasmania
Keywords: Generalized estimating equations (GEE),; binomial; log link; boundary vector; convergence; clustered/longitudinal data
Abstract:

The risk ratio (RR) is a ratio measure of effect for randomized controlled trials and cohort studies. It is possible to estimate the RR in clustered/longitudinal data by fitting a log binomial model using generalized estimating equations (GEE). However, the algorithm may fail to converge even from admissible starting values. Previous studies of the log binomial GEE have focused on convergence rates and the selection of a working correlation matrix. There is no published account of the causes of non-convergence or remedies for it. Our investigations with simulated data suggest that convergence issues for the log binomial GEE arise because the solution lies on a boundary of the allowable parameter space for a model of probabilities. This is the reason that the log binomial model for independent responses may fail to converge. Petersen and Deddens (2010) sketched an “exact” method for resolving convergence difficulties for the log binomial model. We extend the exact method to the log binomial GEE, and provide the details to implement it. The properties of the exact estimator are investigated by simulation, and the results are compared with those of the Poisson GEE.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program