Mediation analysis with STATA and control variables?

[Santiago Valdivieso]

I've seen a lot of forms to do a mediation analysis, being Baron and Kenny (1986) steps the most popular. However, I see that they do the regressions just with the three variables of interest (reg DV IV; reg Mediator IV; reg DV IV Mediator). My first question is: is necessary to include, also, the control variables to do the analysis? And, if so, how it can be done with STATA. I've read that SEM is a good way, but it is normally done withouth control variables.

[Clyde Schechter]

Perhaps the examples of Baron and Kenny mediation analysis that you have seen are simplified examples that use no other covariates ("control variables".) There is no reason you cannot include them. You should include them in the same way in all of the regressions. You can do them with ordinary regression, you do not have to use SEM for this. Where you do need SEM is if you use the Sobel test approach--which is rather different.

[Santiago Valdivieso]

Thank you Clyde! My problem, in this case, is that I don't have the theory to include regressors for the mediator variable when it is used as a dependent variable (so, following Baron and Kenny, I can see if the independent variable predicts the mediator). As I'm analyzing many possible mediators (>8) my study can become very long. Do you see a simpler way out to carry out this analysis?

[Marcos Almeida]

You may also take a look at the erm (extended regression models) machinery in Stata.

[Mike Lacy]

See also -ssc describe khb-. It's well-supported in various publications, and relatively easy to use.