Premium
Flexible link functions in a joint model of binary and longitudinal data
Author(s) -
Li Dan,
Wang Xia,
Song Seongho,
Zhang Nanhua,
Dey Dipak K.
Publication year - 2015
Publication title -
stat
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.61
H-Index - 18
ISSN - 2049-1573
DOI - 10.1002/sta4.98
Subject(s) - inference , computer science , markov chain monte carlo , joint probability distribution , posterior probability , binary number , bayesian inference , binary data , flexibility (engineering) , markov chain , reversible jump markov chain monte carlo , bayesian probability , mathematical optimization , mathematics , statistics , artificial intelligence , machine learning , arithmetic
Joint models of binary primary endpoint and a longitudinal continuous process have been proposed when their association is of interest. The dependence between these two submodels can be characterized by introducing a common set of latent random effects. An important consideration that has been less investigated is to choose appropriate link functions for the binary primary endpoint in this joint model. We introduce two families of flexible link functions based on the generalized extreme value distribution and the symmetric power logit distribution. Our work is the first to investigate the importance of an appropriate and flexible link function in improving the estimation and prediction of a Bayesian joint model. Markov chain Monte Carlo is used for the posterior computation and inference. Flexibility and gains of the proposed joint model are demonstrated through detailed studies on simulated data sets and a real data example. Copyright © 2015 John Wiley & Sons, Ltd.