Publications & Resources
Latent Variable Modeling in the Hierarchical Modeling Framework: Exploring Initial Status x Treatment Interactions in Longitudinal Studies
Michael Seltzer, Kilchan Choi and Yeow Meng Thum
In intervention studies, it is important to assess whether one program might be more effective for individuals with extreme initial difficulties, while another might be more effective for individuals with less extreme initial difficulties. In settings where we obtain time-series data for each person, this entails examining interactions between treatment and initial status on rates of change. In this report, we illustrate a fully Bayesian approach to studying interactions of this kind in the Hierarchical Modeling (HM) framework. This approach provides data analysts with a number of important advantages, including the ability to handle situations in which the number and spacing of time-series observations varies substantially across individuals, and the ability to obtain robust estimates of parameters of interest. Various extensions of our approach are discussed in detail.
Seltzer, M., Choi, K., & Thum, Y. M. (2002). Latent variable modeling in the hierarchical modeling framework: Exploring initial status x treatment interactions in longitudinal studies (CSE Report 559). Los Angeles: University of California, Los Angeles, National Center for Research on Evaluation, Standards, and Student Testing (CRESST).