Mixed-Effect Hybrid Models for Dyadic Longitudinal Data with Nonignorable Dropout

• Activity Number: 575
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Jaeil Ahn*+ and Ying Yuan
• Companies: MD Anderson Cancer Center and MD Anderson Cancer Center
• Keywords: Dyadic ; Longitudinal data ; Nonignorable dropout ; mixed-effect ; latent class

Abstract

Dyadic data consist of paired measurements and are frequently observed in marital relations and dating couples over times. The problem of non-ignorable dropouts under longitudinal dyadic data could lead to biased results. To deal with, we consider a recent class of models, called mixed-effect hybrid models (MEHMs), where the joint distribution of the outcome process and dropout process is factorized into the marginal distribution of random effects, the dropout process conditional on random effects, and the outcome process conditional on dropout patterns and random effects. MEHMs take advantages of well-known selection models and pattern-mixture models:direct modeling the missingness process as in selection models and unloading the computational burden in pattern-mixture models. Noting that a considerable number of missing patterns under dyadic pairs could add complexity to MEHMs, we propose a Bayesian latent class model to reduce dimensionality. This work is largely motivated by an example that originates from the longitudinal dyadic study on spousal relationships and pain in metastatic breast cancer. We illustrate the performance of our approach compared to alternatives.