Statistics and Its Interface

Volume 12 (2019)

Number 2

Analysis of panel data with misclassified covariates

Pages: 309 – 320

DOI: https://dx.doi.org/10.4310/SII.2019.v12.n2.a11

Authors

Grace Yi (Department of Statistics and Actuarial Science, University of Waterloo, Ontario, Canada)

Wenqing He (Department of Statistical and Actuarial Sciences, University of Western Ontario, London, On., Canada)

Feng He (Department of Statistics and Actuarial Science, University of Waterloo, Ontario, Canada)

Abstract

Markov models are commonly used to describe the disease progression, and the likelihood method is usually used to perform inference for such models. However, in the presence of measurement error in the variables, standard inference procedures are no longer valid. In this article, we analytically show that the model is not even identifiable when binary covariates are subject to misclassification. To overcome model nonidentifiability, we consider scenarios where the misclassification probabilities are known, or the main/validation study design is available, and consequently, we propose estimation procedures for Markov models with binary covariates subject to misclassification. Simulation studies are conducted to evaluate the performance of the proposed methods and the consequence of the naive analysis which ignores the misclassification. Our proposed methods are illustrated by the application to the data arising from a psoriatic arthritic study.

Keywords

identifiability, main/validation study design, Markov models, misclassification, panel data

Received 14 September 2017

Published 11 March 2019