Stata 12 came with a module to perform Structural Equation Modeling. Like Amos, there is a SEM diagram builder and fancy dialog boxes but as always commands are directly returned on the command-line so it is not difficult to learn how to write your SEM model directly at Stata prompt or in a do file.

Recently, a book on Discovering Structural Equation Modeling Using Stata was published by Stata Press (Alan C. Acock, 2013). It is well written and introduces main concepts for factor analysis and structural equation modeling. Datasets and do-files used in the book can be downloaded from Stata Press. Some tips about the Stata 12 SEM package have been provided elsewhere. And of course, the UCLA server has already covered some good material about Stata SEM on their Stata FAQ.

An Overview of Stata’s “sem” for Structural Equation Modeling, by K. Bollen and S. Bauldry, is also a good start. Of course, there are many others presentations available via few Google clicks, e.g., Structural Equation Modeling Using the sem Command and SEM Builder.

Now, Stata has just announced the release of Stata 13, with extended capabilities for latent variable modeling and new support for generalized linear response variables (`gsem`

command). See, for example, this example with an ordered logit model. So I expect more publications related to the use of Stata for psychometric research in the future, and it would be very helpful to have a report paper showing the different capabilities of R, Mplus, and Stata on benchmark datasets. Surely, a direct comparison of gllamm vs. Stata built-in SEM package would be very insightful as I always found `gllamm`

to be much to slow to be used in routine tasks (IRT, multi-group CFA, etc.). Actually, using a combination of `polychoric`

and `factor`

to carry out a factor analysis on Likert-type items is also very slow.