Choose Language Hide Translation Bar

An Introduction to Structural Equation Models in JMP(R) Pro 15 (2019-US-45MP-273)

Level: Beginner


Laura Castro-Schilo, JMP Research Statistician Developer, SAS


Structural Equation Models (SEM) is a new platform in JMP Pro 15 that offers numerous modeling tools. Confirmatory factor analysis, path analysis, measurement error models and latent growth curve models are just some of the possibilities.


In this presentation, we provide a general introduction to SEM by describing what it is, the unique features it offers to analysts and researchers, how it is implemented in JMP Pro 15, and how it is applied in a variety of fields, including market and consumer research, engineering, education, health and others. We will use an empirical example — that everyone can relate to — to show how the SEM platform is used to explore relations across variables and test competing theories.


It would be great to get the JRN for this presentation - the PDF is informative, but I'm having a hard time understanding the results without seeing the actual process.



I didn't have a journal for the presentation but I just captured a brief video on my computer, in which I fit models consistent with each of the "Emotion Theories" in the presentation. I'm attaching the video to this reply. Together with the slides, I hope it's easier to understand how to use the SEM platform. Here are the takeaway points:


  1. We have 3 a-priori theories of how our variables relate to each other
  2. We fit models in SEM that map onto each of the theories
  3. We look for the most appropriate model by:
    1. Examining individual model fit. In this example we used the chi-square, a measure of misfit --we want it to be small with respect to the degrees of freedom (df) and non-significant, but the CFI and RMSEA can be used too and are better with large sample sizes. We also used the normalized residuals heatmap; we want those residuals within +/- 2 units.
    2. Comparing fit across models. The AICc weights help us with this task. When the models are nested (i.e., one is entirely contained within the other --in our example, the model for Theory #1 is nested within that for Theory #2), we can take the difference between the chi-squares and the df to obtain a delta-chi-square and delta-df that can be tested for significance with a chi-square distribution.

I hope this helps!