Statistics and Its Interface

Volume 2 (2009)

Number 2

Analysis of multi-level correlated data in the framework of generalized estimating equations via xtmultcorr procedures in Stata and qls functions in Matlab

Pages: 187 – 196

DOI: http://dx.doi.org/10.4310/SII.2009.v2.n2.a8

Authors

Sarah J. Ratcliffe (Department of Biostatistics and Epidemiology, University of Pennsylvania School of Medicine, Philadelphia, Penn., U.S.A.)

Justine Shults (Department of Biostatistics and Epidemiology, University of Pennsylvania School of Medicine, Philadelphia, Penn., U.S.A.)

Abstract

Many medical studies yield data with multiple sources of correlation. For example, in a study of repeated measurements collected on each eye of spouses, three sources of correlation may be present, due to the fact that measurements within the same family will be more similar if they are measured on the same eye (left versus right), within the same person (husband versus wife), or at the same measurement occasion. This article reviews an algorithm for analysis of data with two or more sources of correlation (Shults, Whitt, Kumanyika, 2004) that can be implemented using quasi-least squares, an approach in the framework of generalized estimating equations. It then describes and demonstrates implementation of this algorithm with xtmultcorr procedures in Stata and the qls functions in Matlab. The Stata and Matlab procedures are available on the website for the Longitudinal Analysis for Diverse Populations project: http://www.cceb.upenn.edu/~sratclif/QLSproject.html.

Keywords

Cholesky decomposition, correlated data, generalized estimating equations, multi-level data, multivariate data, quasi-least squares

Full Text (PDF format)