Measurement error in linear mixed models
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Measurement error of exposure is widely acknowledged as pervasive and often important source of bias or misleading results in much research. There has been an abundance of interest in the topic of covariate measurement error and there is an extensive literature in measurement error models and methods. Clustered data can be defined as data in which the observations are grouped into disjoint classes, called clusters, according to some classification criterion. Mixed models were developed to handle clustered data and have received a great deal of attention in the statistical literature for the past years because of the flexibility they offer in handling the unbalanced clustered data that arise in many areas of investigation. In this thesis we consider both linear and linear mixed effect models with measureĀment error. Three methods are compared through simulation studies, namely the naive method, the two-step approach, and the likelihood estimation method. The naive method ignores covariate measurement error in models, the regression calibration method, as a two-step approach, is a commonly used simple approach and may be applicable to alĀmost any regression models, and the Expectation Maximization (EM) algorithm, as a likelihood method, treats random effects as missing data. Naive approaches are shown to be inadequate to be used when covariates are subject to error. Both the regression calibration method and the EM method appear to be good, but the regression calibration method is much simpler than the EM method. We illustrate these methods in HIV study data.
Description
Bibliography: p. 46-50
Keywords
Citation
Lawal, O. A. (2012). Measurement error in linear mixed models (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/5014