Search Results
Author: Demirtas, Hakan
Resulting in 1 citation.
| 1. |
Demirtas, Hakan |
|
Multiple Imputation for Nonignorable Dropout Using Bayesian Pattern-Mixture Models Ph.D. Dissertation, The Pennsylvania State University, 2003. DAI-B 64/09, p. 4442, Mar 2004 Cohort(s): NLSY79 Publisher: UMI - University Microfilms, Bell and Howell Information and Learning Keyword(s): Bayesian; Modeling, Mixed Effects This dissertation examines conventional pattern-mixture models and a new Bayesian pattern-mixture model for nonignorable dropout. Many current procedures for incomplete data ignore the probabilistic mechanism producing the missing data, which is appropriate when the missing values are missing at random (MAR). However, in many longitudinal studies, subjects could be dropping out for reasons strongly related to unobserved data. Erroneous assumptions of MAR may lead to severe biases when the missingness mechanism is actually nonignorable. In this thesis, I review the present state of methods for nonignorable dropout, and examine the performance of one popular class of pattern-mixture models when the form of the population has been misspecified. Then I develop a new class of Bayesian random-coefficient pattern-mixture models that can be applied routinely to impute missing values when the ignorability assumption is doubtful. I develop procedures for model fitting and Bayesian multiple imputation under this linear mixed-effects model. I then apply this methodology to data from a randomized psychiatric trial and a national longitudinal survey (Bibliography editor: NLSY79). Evaluating the performance of this approach through simulations, I also make comparisons with selection models and semiparametric marginal models based on conventional and weighted estimating equations. |
|
|
Bibliography Citation
Demirtas, Hakan. Multiple Imputation for Nonignorable Dropout Using Bayesian Pattern-Mixture Models. Ph.D. Dissertation, The Pennsylvania State University, 2003. DAI-B 64/09, p. 4442, Mar 2004. |