Path analysis using GSEM for categorical data

[Galit Andersson]

I want to run a path analysis using the GSEM command in Stata. I have data that is mainly composed of categorical variables and some dichotomous. I would like to find the indirect paths between different covariates and explanatory variables to the outcome variable sexual satisfaction (dichotomous variable). To do that, I first ran a multivariate logistic regression analysis between various independents variables and my outcome (sexual satisfaction). Then, I ran several other logistic regression analyses with the significant explanatory variables from the previous analysis as the outcome variables and the remaining explanatory variables as independent variables. In the next step, I would like to try to fit the results from these regression analyses in a path model and see whether the direct and indirect paths are significant and whether the model fit is good.

I have a few questions about my strategy:

1. I know I can use GSEM with a dichotomous outcome but can I use the GSEM command also for categorical independent variables?
2. I have a hard time determining which variables are endogenous and which ones are exogenous variables. Another problem is that I don’t know whether to include my independent categorical variables as one variable or as several (i.e. 1.var 2.var etc.)
3. Below I have two models that I have tried using the SEM Builder. In the first example, I tried the results from my regression analyses using the errors to indicate correlations and covariance. In the second example, I did the same thing only here I indicated my categorical variables by 1.var 2.var etc. and I treated the directly correlated variables to sexual satisfaction also as outcome variables.
4. Which one, if any, of these models is correct?

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. You don't tell us even what you're using to estimate the model or exactly what the estimation statement looks like. You will also increase your chances of a response by offering a shorter, more focused posting.

1. I know I can use GSEM with a dichotomous outcome but can I use the GSEM command also for categorical independent variables?

Exactly what do you mean by categorical? Have you carefully searched the examples in the GSEM documentation?

1. I have a hard time determining which variables are endogenous and which ones are exogenous variables. Another problem is that I don’t know whether to include my independent categorical variables as one variable or as several (i.e. 1.var 2.var etc.)

What is endogenous and what is exogenous are as much substantive issues about your project as statistical issues. If you have a non-ordered categorical variable, then the i.varname factor variable notation is usually helpful. You'll need to include (or let Stata include using i.varname) dummies for each level, omitting one.

1. Below I have two models that I have tried using the SEM Builder. In the first example, I tried the results from my regression analyses using the errors to indicate correlations and covariance. In the second example, I did the same thing only here I indicated my categorical variables by 1.var 2.var etc. and I treated the directly correlated variables to sexual satisfaction also as outcome variables.

I don't understand what you've done. There is no general solution to "which model is correct" - it depends on your theory, etc. In my opinion, it is usually better to include the entire analysis in the GSEM estimation rather than estimating parts using other techniques. It is often the case that the assumptions in the pre-analysis are incompatible with the GSEM assumptions.

1. Which one, if any, of these models is correct?