Semiparametric Estimation in Transition Measurement Error Model

• Activity Number: 218
• Type: Contributed
• Date/Time: Tuesday, August 5, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Wenqin Pan*+ and Donglin Zeng and Xihong Lin
• Companies: University of Michigan and University of North Carolina and University of Michigan
• Keywords: measurement error ; longitudinal data ; semiparametric estimation ; transition model

Abstract

We propose a new class of models, transition measurement error models, for analyzing longitudinal data when one of the covariates is measured with error. Three approaches are considered to reduce the estimating bias due to measurement error and achieve valid inferences of parameter estimators. They are MLE approach, SIMEX approach, and semiparametric estimation approach. While the unbiasedness of the former two approaches both depend on some unobserved information, a semiparametric approach can give consistent parameter estimators without any assumption of the distribution of the error-prone covariate. We first generalize a traditional conditional score approach to transition measurement error model, and apply the approach on both linear and logistic transition models to achieve consistent estimators. For the linear case, we then discuss the loss of efficiency in the conditional score approach. Finally, under the condition that a small set of validation data is available, we propose a one-step estimation approach to achieve semiparametric efficient estimators. Both numerical calculation and simulation studies are performed to evaluate and illustrate the proposed approaches.