Missing Data in Longitudinal Studies: Strategies for Bayesian Modeling and Sensitivity Analysis

摘要: PREFACE Description of Motivating Examples Overview Dose-Finding Trial an Experimental Treatment for Schizophrenia Clinical Recombinant Human Growth Hormone (rhGH) Increasing Muscle Strength in the Elderly Trials Exercise as Aid to Smoking Cessation Women: The Commit Quit Studies Natural History HIV Infection Epidemiology Research Study (HERS) Cohort among Substance Abusers: OASIS Equivalence Competing Doses AZT HIV-Infected Children: Protocol 128 AIDS Group Regression Models Preliminaries Generalized Linear Conditionally Specified Directly (Marginal) Semiparametric Interpreting Covariate Effects Further Reading Methods Bayesian Inference Likelihood and Posterior Distribution Prior Distributions Computation Model Comparisons Assessing Fit Nonparametric Bayes Analysis using Data on Completers Selection with a Multivariate Normal Model: Random Binary Longitudinal Data: CTQ I Summary Missing Mechanisms Introduction Full vs. Observed Full-Data Assumptions about Mechanism at Applied Dropout Processes Observed-Data Parameters Ignorability Assumption under MAR MNAR General Issues Specification Sampling Using Augmentation Covariance Structures Univariate Covariate-Dependent Incomplete Case Studies: Ignorable Missingness Marginalized Transition Weekly Outcomes II Auxiliary Variable HERS CD4 p-Spline handling Nonignorable Extrapolation Factorization Mixture Shared Parameter Informative Priors Sensitivity Some Principles Parameterizing Pattern-Mixture Elicitation Expert Opinion, Construction Priors, Formulation Analyses A Note Fully Parametric Literature Local Not Pediatric Varying Coefficient Appendix: distributions Bibliography Index