how to evaluate goodness of fit of GSEM?

[wendy chang zhou]

Hello everyone,

I am new to SEM. My model has a count mediator and a continuous outcome, so I am using GSEM command to run the mixed model of two equations. One is to predict the count mediator, and the other is to predict the continuous outcome including the mediator and three control variables (x4, x5, x6). x1, x2, x3 are independent variables of interest. My purpose is to test indirect effects using structural equation model.

The code is like:

gsem (mediator<-x1, x2, x3, nbreg) (outcome<-x1, x2, x3, x4, x5, x6)

My questions are :

1) what is the difference between GSEM and the situation that I run two models separately. It seems that GSEM just puts two equations in one command. The results are the same as separate models of a nbreg (negative binomial regression) and a linear regression.

2) how to evaluate the results of GSEM? Can GSEM produce goodness of fit tests like sem, so that the model can be evaluated, modified and compared. I do a lot of research, but I cannot find helpful information about that. Any relevant literature recommended?

I appreciate your help in advance.

[Phil Bromiley]

You didn't get a quick answer. You will 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.

gsem will estimate a wide variety of models for which specific procedures exist. Since you do not constrain the covariance between the errors in the two equations, yes, it is just like estimating them separately. As for fit, sem and gsem produce numerous fit indices. You can also get AIC and BIC for them.