Measurement Error in Analysis of Longitudinal Data

Abstract

We model covariate measurement error in longitudinal data, study its effects on naive analyses, and present a method for consistent estimation in its presence. Covariates in longitudinal data can be decomposed into between- and within-cluster components, and we demonstrate that measurement error in covariates can lead to difference between the effects of these two components. By extending existing results for the problem of different between- and within- cluster covariate effects, we provide a general assessment of how measurement error gives rise to biases in the estimation of regression coefficients. Furthermore, we propose the Decomposition-and-Recombination method, which produces consistent estimators for the regression coefficients of interest in the presence of measurement error. The performance of the estimators in small samples is evaluated by means of simulations, and the method is applied to analyzing data from a longitudinal study. * This is a joint research with Lei Shen, Mari Palta, and Soomin Park of Department of Statistics, University of Wisconsin-Madison. Keywords: Correlated data; Marginal models, Generalized estimating equations; Linear mixed effects models.