About SEM Software

There are several main points to be made about the software programs that can be used for performing SEM.

(1) There are quite a few software programs currently available for performing structural equation modeling. The first of the widely used programs of this sort was LISREL, which as of this writing is still available. A number of other programs are also available including EQS, Amos, CALIS (a module of SAS), SEPATH (a module of Statistica), and Mplus. There are also two packages in R, lavaan and "sem", which are of course available for free.

(2) New users of SEM inevitably wish to know which of these programs is best. One point in this regard is that most of these programs are updated fairly frequently, making any description I might give of the limitations of a program potentially out of date. Another point to make is that different individuals prefer different features. Some want the software that will allow them to get started most quickly, others want the software with the most capabilities, still others want the software that is freely available to them.

(3) My thoughts about software is that the different packages serve different users. Currently, in this website I try to provide information for multiple types of users and, therefore, provide examples in several different languages. Here I give a brief description of some different languages and their characteristics.

(6) Amos. Amos is a popular package with those getting started with SEM. I have often recommend folks start learning SEM using the free student version of Amos simply because it is such a good teaching tool. It has the most useful manual for beginning users of SEM as well. What it lacks at present: (1) limited capacity to work with categorical response variables (e.g. logistic or probit forms) and (2) a limited capacity for multi-level modeling. Amos does have a Bayesian component now, which is helpful. That said, at the moment, it is a pretty limited Bayesian implementation and leaves the more advanced options out.

(7) Lavaan. We are now well into what can be called the "R-age" and it is, well, all the rage alright. R is transforming quantitative analysis. Its role will continue to grow at a dramatic rate for the foreseeable future. There are two R packages dedicated to second-generation SEM analyses ("classical sem", which involved the anaysis of covariance structures). At the moment, I choose the lavaan package to present at this site, which is not to say that the SEM R packages isn't just fine. As of 2015, a new R package for local estimation of models is available, aptly called "piecewiseSEM".

(8) Bayesian Implementations. At present, I am starting to present winBUGS tutorials for Bayesian implementations. winBUGS is a free software package for Markov chain Monte Carlo (MCMC) simulation using Bayes theorem. More can be found about this advanced application method under the Bayesian tab.

I am happy to receive any comments about your experiences with SEM software or about this material at jimgrace001@gmail.com.