• The Korean Journal of Applied Statistics
• /
• v.37 no.4
• /
• pp.467-484
• /
• 2024

Longitudinal data are measured repeatedly over time from the same subject, so there is a correlation from the repeated outcomes. Therefore, when analyzing this correlation, both serial correlation and between-subject variation must be considered in longitudinal data analysis. In this paper, we will focus on the marginalized models to estimate the population average effect of covariates among models for analyzing longitudinal binary data. Marginalized models for longitudinal binary data include marginalized random effects models, marginalized transition models, and marginalized transition random effect models, and in this paper, these models are first reviewed, and simulations are conducted using complete data and missing data to compare the performance of the models. When there were missing values in the data, there is a difference in performance depending on the model in which the data was generated. We analyze Korea Health Panel data using marginalized models. The Korean Medical Panel data considers subjective unhealthy responses as response variables as binary variables, compares models with several explanatory variables, and presents the most suitable model.