Far East Journal of Theoretical Statistics
Volume 16, Issue 2, Pages 205 - 234
(July 2005)
|
|
ROBUST METHODS FOR INCOMPLETE LONGITUDINAL DATA WITH MISMEASURED COVARIATES
Grace Y. Yi (Canada)
|
Abstract: In
longitudinal studies, a response variable,
often with a set of covariates, is observed on
individuals repeatedly over time. Missing
observations often arise in such studies.
Moreover, some covariates may be subject to
measurement error. Standard inference methods
for longitudinal data are no longer valid
under this scenario. In this paper we propose
robust methods to estimate parameters of
interest in settings of longitudinal data with
missing observations as well as mismeasured
covariates. The response process is
characterized by means of generalized
linear models where only the mean and variance
structures are specified. Measurement error is
modeled by the functional method for which the
distribution of the true covariates is not
required. Only the moment structures of
measurement error are assumed. Consistent
estimators of the mean response parameters can
be derived from unbiased estimating equations,
provided the missing data process is correctly
accommodated. To provide flexibility of
modeling strategies, we explore both
conditional and marginal approaches to
facilitate the missingness. The proposed
estimation approaches are specifically
addressed to various regression models that
are often utilized for practical problems.
Asymptotic properties are established. The
implementation of the developed methods is
easy and straightforward. The performance of
the proposed methods is assessed through
simulation studies. |
Keywords and phrases: drop-out, estimating functions, longitudinal data, measurement error, missing at random, regression models. |
|
Number of Downloads: 249 | Number of Views: 764 |
|