Path Model vs SEM

[Denila Jinny]

What are the differences between SEM and Path Model? As far as I know, SEM overcomes two of the issues with path model, with latent variables and non-recursive models. But is it necessary that SEM should always have a latent variable in the model? Please clarify. Thank You.

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

SEM is a very general maximum likelihood estimator - if you look at the documentation, you'll see it can estimate a wide variety of models with and without latent variables and with or without simultaneity. GSEM is even more general. But, often there are model-specific estimators that are easier to use than SEM. For example, you can do regression in SEM, but regress is easier and comes with a great many post-estimation tools.

Path analysis, as I understand it, is a way to specify and estimate models often with latent variables and usually with more than one dependent variable. When maximum likelihood was not easy, path analysis offered a computationally viable approach to these models.

While many use diagrams to represent path analysis models, these are just as feasible in SEM/GSEM models. Indeed, Stata's SEM facility comes with a graphic user interface so you can specify a model using a diagram and then Stata will translate it into an SEM statement.

[Weiwen Ng]

... But is it necessary that SEM should always have a latent variable in the model? Please clarify. Thank You.

The only thing I can add to what Phil said is that it is not necessary for SEM to have a latent variable in the model. Browse a few of the SEM examples, and you will see that a number of them don't have latent variables. In fact, you can do the equivalent of ordinary least squares regression (aka linear regression) in SEM.

[Richard Williams]

Alan Acock's book is a good intro to SEM:

https://www.stata.com/bookstore/disc...g-using-stata/

For a briefer/free intro, you can see

https://www3.nd.edu/~rwilliam/stats2/l95.pdf

[Richard Williams]

In fact, you can do the equivalent of ordinary least squares regression (aka linear regression) in SEM.

As a sidelight, you can use Full Information Maximum Likelihood to handle missing data when you use SEM this way, e.g.

sem (x1 x2 x3 -> y), method(mlmv)

A FIML option for the regress command is one of my wish list items.