Advances and Applications in Statistics
Volume 18, Issue 1, Pages 73 - 87
(September 2010)
|
|
A BAYESIAN APPROACH TO NONLINEAR MIXED-EFFECTS MODELS WITH MEASUREMENT ERRORS AND MISSINGNESS IN COVARIATES
Wei Liu
|
Abstract: There are inter-individual and intra-individual variations in longitudinal studies. The inter-individual variation often receives great attention. Nonlinear mixed-effects (NLME) models have been widely applied in modeling longitudinal data, in which time-dependent covariates may be introduced to partially explain the inter-individual variation. However, some covariates may be measured with substantial errors and may contain missing values. The likelihood method for NLME models with measurement errors and missing data in covariates may lead to poor parameter estimation or identifiability problems, especially if the observed data are not rich. To overcome these difficulties, we propose a Bayesian approach, implemented by combining the Gibbs sampler with the Metropolis-Hasting algorithms, and illustrate it by analyzing a real dataset. |
Keywords and phrases: Bayesian approach, longitudinal data, measurement error, Metropolis-Hasting algorithm, missing data. |
|
Number of Downloads: 365 | Number of Views: 1062 |
|