PSY-8006: Introduction to Structural Equation Modeling at NTNU, Trondheim.

In Structural Equation Modeling (SEM), one aims to combine a ‘measurement model’ of a number of observed indicator variables measuring an unobserved latent construct, with a structural path model measuring associations between such constructs and a number of other covariates. This is useful in instances where we wish to combine several variables measuring what we believe to be different aspects of a single phenomenon (or ‘factor’), e.g. happiness or depression, and measure how such factors relate to other factors or other observed variables. SEM can be combined with other types of modeling, such as time-to-event-, multilevel-, and mediation analysis, amongst other.

The course was hosted by the Department of Psychology, at NTNU campus Dragvoll. It set out to give a comprehensive introduction to the various building blocks used in a complete SEM, as well as the software packages available for use in analysis alongside possible advantages and drawbacks of each of the options. This included specification and estimation of confirmatory factor analysis (CFA) used to identify latent constructs, as well as mediation analysis within the SEM software framework. Lectures were held by prof. Mehmet Mehmetoglu, author of statistical manuals regarding CFA and SEM, as well as developer of several user-written STATA packages in various fields of statistics. Course days were divided between lectures and Q&A sessions before lunch, with labs and practical, interactive problem-solving during the second half of the day. A great benefit of the latter was the companion syntax files for both Mplus and STATA, which gave participants a lot of useful code for their own projects – a boon for the busy PhD-student. The course is worth 10 SPs and evaluates based on a written individual paper of 4-5000 words where the acquired skills are to be applied in practical problem solving.

I found the course to be at an appropriate level for PhD-students lacking experience with factor models; it gave a comprehensive introduction to the rationale of SEM analysis, its intrinsic building blocks, and the various manners such models can be expanded and adapted to other types of data. If you have a functional understanding of ordinary regression models and wish to expand this with latent variable modelling, I would recommend this course.

By Anders Årnes, PhD-fellow at the Pain department, University Hospital North-Norway, and the Faculty of health sciences, UiT-The Arctic University of Norway.