Introduction to Structural Equation Modeling

In person in Calgary | July 18-19, 2023, during Rocky Mountain Methodology Academy

$849 (Canadian dollars) + 5% GST (10% discount for graduate students)

Structural equation modeling (SEM) is widely used in many disciplines including psychology, education, communication, biology, medicine, and others. It is often used to analyze and test models derived from theory and existing literature linking variables together in a structural system. It is also used to manage and account for some of the effects of measurement error that many statistical approaches assume is absent.  This course introduces the fundamentals of SEM as a general analytical tool, including how to set up measurement and structural models, latent variables, path analysis, definitions and quantification of model fit, and the implementation of SEM in statistical software.


In this course, you will learn about the basic principles underlying structural equation models and then learn how to apply these to practical problems. The topics covered include:

  • Introduction to latent variable models, measurement error, path diagrams
  • Estimation, identification, interpretation of model parameters
  • Model fit and model improvement
  • Mediation models in the structural equation framework
  • Multiple-group models
  • The FIML approach to analysis with missing data
  • A brief introduction to alternative estimators for non-normal data
  • A brief introduction to models for means and intercepts

This course is delivered in person and will take place over two days at the University of Calgary. Class will begin at 9am and end at 5pm each day.

We will make primary use of the lavaan package in R but will also briefly demonstrate the sem procedure in STATA and provide instructional videos for use after class for those who wish to work with STATA instead of R.   Very little prior experience with R is required.  It is recommended that participants bring a laptop to class with a recent installation of R with the lavaan package installed (installation of the haven and semTools packages, while not required, will be helpful).  R softweare is open-source (free).

This course will be helpful for researchers in many fields that employ regression models, factor analysis models, psychometric scale construction, or a combination of these.   Structural equation models appear in fields as diverse as psychology, education, business, human development, public health, communication, social work and political science.   Participants will learn how to estimate models to adjust for biases created by measurement error, which is often left undealt with, and to learn new approaches to deal with missing data.   A reasonably strong familiarity with multiple regression analysis (including multiple regression with dummy variables) is essential.   While a bit of prior experience with R would be helpful, no prior experience with any software is assumed.

Upon completing this course, you will

  • have mastered the basic conceptual ideas behind latent variable modeling.
  • have learned how to combine “scale construction” or factor analysis with multiple regression modelling in an integrated framework.
  • understand and appreciate the bias created by both random and systematic measurement error and how this can be diagnosed and dealt with.
  • understand the translation of diagrams to equations and parameter specifications and vice versa.
  • be able to use diagnostic tools to improve the “fit” of models.
  • understand situations in which models might not converge (software fails to yield valid estimates) and how to deal with these situations.
  • be able to make informed choices about appropriate fit criteria (e.g., when is a model good enough?)
  • be able to test for measurement equivalence across groups in comparative research

A certificate of completion from the Canadian Centre for Research Analysis and Methods is provided at the end of the course.

In this course, we do not cover some of the more advanced topics associated with structural equation modeling.  Most notably, we do not discuss (other than parenthetically) how SEM models can be used for longitudinal (panel) data and how the SEM framework can be valuable when constructing growth curve models.  Beyond a brief introduction to estimators for non-normal data, we do not discuss models involving categorical (dichotomous or ordered) indicators or models for categorical (ordinal, nominal or count) outcomes.   We do not discuss latent variable interaction models in any detail.   And we do not discuss multilevel SEM models.  The “foundations” provided in this course will help if, in the future, participants wish to move into these advanced topic areas.