Bridging Disciplinary Boundaries (January 11 - 14, 2007)
|Friday, January 12, 2007: 8:00 AM-9:45 AM|
|Seacliff C (Hyatt Regency San Francisco)|
|Advanced Topics in Structural Equation Modeling (SEM): Identification and Beyond|
|Speaker/Presenter:||Shenyang Guo, University of North Carolina at Chapel Hill|
Purpose: The structural equation modeling (SEM) approach has been increasingly employed by social work researchers. A recent review of eight social-work or social-work-related journals from January 1, 1999 to December 31, 2004 has found a total of 139 articles published by these journals that employed SEM. Recent advances in statistical modeling and computer software packages have made SEM a more effective and efficient tool for addressing a broad range of research questions. The rapid development in this field is both a good news and bad news to researchers. Because running and estimating a model is no longer cumbersome, testing the tenability of statistical assumptions embedded in a model (such as assumptions about independent observations and normality), and developing a meaningful and parsimonious model with accurate interpretation become crucial. Because some software packages (e.g., Mplus) were designed to estimate many kinds of statistical models, a user may be confused by many options and get lost on choosing a right type of model that is best suited to his or her research objectives. Under such context, understanding connections and differences among different types of models is important.
This workshop aims to focus on a few advanced topics in SEM to provide participants an overview of key steps leading to an acceptable model that meets statistical criteria and at the same time conveys substantively important messages.
Contents: The workshop begins with an overview of the general SEM procedure, particularly the fundamental assumption about the observed data and model-implied parameters, and its implications to identification, estimation, and model-fit testing.
It then reviews the issue of identification, a key to a successful estimation of a complex model. It reviews identification rules and demonstrates empirical means to check the identification status for a particular model. Strategies to make an underidentified model identified are illustrated.
It then takes the classical study of Bagozzi (1980) on the relationship between job performance and satisfaction and Joreskog and Sorbom's (1996) rerun of the model as an example to provide a step-by-step illustration of a modeling procedure testing non-recursive hypothesis. Participants will learn from this example empirical tips for developing a statistically rigorous but substantively meaningful model; specifically, understand the importance of developing a good measurement model first when running the general SEM.
It then reviews a few seemingly related but conceptually different models to show which models are useful to solving what types of problems. Emphases will be given to the differences between the conventional SEM and the newly developed mixture modeling; the connections between SEM (i.e., the latent growth curve modeling) and HLM; and differences between Mplus and other SEM packages.
The workshop concludes by offering a list of common pitfalls in running SEM.
Bagozzi, R.P. (1980). “Performance and satisfaction in an industrial sales force: An examination of their antecedents and simultaneity”, Journal of Marketing 44: 65-77.
Joreskog, K., & Sorbom, D. (1996). LISREL 8: User's Reference Guide, Chicago, IL: Scientific Software International, Inc.
See more of Workshop
See more of Bridging Disciplinary Boundaries (January 11 - 14, 2007)