Structural equation modeling

Structural equation modeling (SEM) is a multivariate statistical analysis technique that simultaneously unites Factor Analysis and Multiple Regression Analysis. It analyses the causal relationships among observed variables and latent constructs, including linear and nonlinear effects. SEM includes two basic types of models. The measurement model represents the theory that specifies how a set of observed variables measure the latent constructs. The structural model represents the theory that shows how latent constructs are causally related to each other.

SEM can apply to various data types, cross-sectional data, longitudinal data, time-series data, or multilevel data. For example, the cross-sectional models help us to assess causal and mediation hypotheses; Latent Growth Curve Models often apply to analyse potential changes in the latent construct of interest over time; Item Response Theory Models usually analyse the patterns of individual behaviours and questionnaire responses, and Multilevel Models can assess the cause of variations between different data levels. SEM has been widely applied in social sciences and spread to other natural sciences in recent decades, e.g., information science and medical research.

The Department of Statistics in Uppsala has a long tradition of structural equation modeling, and it is known as the birthplace of SEM. Professor Emeritus Karl G. Jöreskog is the pioneer in SEM, and the LISREL (linear structural relations) program (Jöreskog and Sörbom) was the first software for analysis of Structural Equation Models. Nowadays, Professor Fan Wallentin and Associate Professor Shaobo Jin with colleagues carry on this rich tradition and actively contribute to the field.

Responsible researcher: Fan Yang Wallentin

Last modified: 2021-08-31