Markov Transition Models for Binary Longitudinal Data with Missing Values

• Activity Number: 273
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xiaowei Yang*+ and Kun Nie and Steven Shoptaw and Qing J. Zhang
• Companies: BayesSoft, Inc. and BayesSoft, Inc. and BayesSoft, Inc. and BayesSoft, Inc.
• Keywords: missing data ; longitudinal data analysis ; Markov chain ; transition model

Abstract

In longitudinal studies, binary repeated measures are commonly met. Statistical analyses for such datasets are often plagued by problems with missing values due to nonresponses or withdrawal. For example, due to the chaotic nature of substance abuse research, many participants would miss their clinic visits or drop out of the studies prematurely. When all measures are observed, Markov transition models provide a good analytical solution where repeated measures on each subject are viewed as samples from a Markov chain with transition probabilities: P(I,j) = prob(Y(t)=j|Y(t-1)=i), where I and j equals to 1 or 0 indicating recent "use" or "no-use," Y(t) and Y(t-1) represent two measures collected on time t and t-1 (t=1, ., T). This modeling strategy brings convenience to study dynamic changes of binary variable through time, which simply involves a modified logistic regression model. By estimating transition probabilities for each treatment condition, we can compare their treatment effects. We propose a method to extend the above Markov transition models.