Structural Equation Modeling

Structural equation modeling (SEM) is one of the most widely used method for analyzing multivariate data in the social and behavioral sciences.

It has its roots in factor analysis which began with an article by the British psychologist Charles Spearman (1904). In the first half of the 20th century, factor analysis was mainly developed by psychologists using ad hoc procedures. In the second half of the 20th century several statisticians became interested in the challenging statistical estimation problems involved, notably D.N. Lawley, T.W. Anderson, K.G. Jöreskog, M.W. Browne, A. Satorra, D.Sörbom, and B. Muthén.

Jöreskog extended the classical exploratory factor analysis to confirmatory factor analysis, second-order factor analysis, multiple-group factor analysis and to the general structural equation model and he developed methods for estimation and testing of such models in cross-sectional, longitudinal, multi-group, and multilevel data.

Former PhD students in our division of statistics have made important contributions to this methodology: Sörbom (1976) extended the multi-group model to include means of latent variables, Muthén (1977) developed methods for categorical observed variables, Ulf Olsson (1978) studied maximum likelihood estimation of polychoric correlations, Hägglund (1985) developed two-stage least-square methods, Quiroga (1992) studied the robustness of polychoric correlations to departures from underlying normality, and Yang-Wallentin (1997) developed methods for estimating non-linear latent relationships.

Recent developments in structural equation modeling include extensions to estimation from complex survey data, generalized linear models and time series data. Ongoing projects in the department of statistics focus on SEM with ordinal variables, SEM with ranking data, and various applications within SEM frame work.