Statistical Methods: Structural Equation Modeling (SEM)

Structural equation modeling (SEM), mostly used by social scientists, can be adopted to understand CHES processes. Specifically, SES can help estimate complex causal relationships or plausible pathways (also called path analysis) among latent variables, often psychosocial constructs that are not directly measurable. Instead, these constructs, largely abstract concepts, can be expressed by a set of concrete, measurable indicator variables. Using a unique covariance matching method, SEM allows simultaneously estimating multiple causal relationships (Bollen 1989; An et al. 2003) among the selected latent variables. Therefore, SEM provides an improved approach to detecting causality from correlation.

As a special type of SEM, latent trajectory modeling uses patterns of change (i.e., trajectory parameters such as intercept and slope) in the response variable y, instead of psychosocial constructs, as latent variables, modeling change patterns in y as well as the antecedents and consequents of these change pattern (Preacher et al. 2008). We use SAS and MPlus for our SEM and LTM modeling work. Below is a list of exemplar papers and teaching notes about structural equation modeling.

Readings and References:

An, L., A. Mertig, and J. Liu (2003). Adolescents' leaving parental home in Wolong Nature Reserve (China): psychosocial correlates and implications for panda conservation. Population and Environment: A Journal of Interdisciplinary Studies 24(5):415-444.

Bollen, K.A. (1989). Structural equations with latent variables. New York: Wiley.

Preacher, K. J., Wichman, A. L., MacCallum, R. C. & Briggs, N. E. Latent growth curve modeling (SAGE Publications, 2008).

Examples, Models, and/or Documents:

LTM Teaching Notes