Substance use is a well-known risk factor for public health and well-being in the United States. The identification of particular stages of substance use can reveal optimal opportunities for intervening in the onset process for different groups of individuals. Moreover, this information can be used to assess the relative treatment effects on different stages of substance use. Latent transition analysis (LTA) is ideally suited to this type of research, and it has been used successfully to describe stages in the area of substance use prevention and treatment. It is important that a model be assessed adequately at the outset of an analysis, as it has important ramifications for all following analysis. Drug abuse researchers currently have several methods at their disposal for evaluating the fit of LTA models, depending on the software package being used. Unfortunately, it is not clear which of the currently available methods is optimal; in fact, different available tools often suggest different models. The overall goal of this project is to evaluate and compare the performance of existing model assessment tools including new methods that are not readily available to social and behavioral scientists. The proposed project involves extensive simulations to investigate the behavior of statistical selection methods. Statistical simulations are an extremely useful tool for this type of methodological research because the correct latent class structure of the empirical data is typically unknown. We will conduct a thorough simulation study to compare ten selection methods to assess their performance under a variety of situations that occurs in practice in drug abuse intervention research. In addition, we will lay the groundwork for the development of new goodness-of-fit indices for LTA models, measuring how much better a model fits, with a 0-1 scale, as compared to a competitive alternative. As a result of this project, we will provide drug abuse scientists with methodologically sound and practical guidelines for diagnosing model fit and selecting appropriate LTA models under different conditions. A series of articles will be submitted to peer-review journals, and a project web site also will be developed in The Methodology Center web site at Penn State, where we will post information on model selection that will be accessible to drug abuse researchers. ? ? ? ?
| Chung, Hwan; Anthony, James C; Schafer, Joseph L (2011) Latent class profile analysis: an application to stage-sequential process in early-onset drinking behaviours. J R Stat Soc Ser A Stat Soc 174:689-712 |
| Chung, H; Breslau, N (2008) The latent structure of post-traumatic stress disorder: tests of invariance by gender and trauma type. Psychol Med 38:563-73 |
| Chung, Hwan; Lanza, Stephanie T; Loken, Eric (2008) Latent transition analysis: inference and estimation. Stat Med 27:1834-54 |
