Indirect Effects in GSEM

[adson]

Dear all,

I am working on a path model with a dichotomous outcome variable and three continuous mediators. Because of the dichotomous outcome, I am using GSEM.
Yet, I am not sure about estimating the indirect effects.

According to the Stata Manual, nlcom is the way to go. Yet, in the manual's gsem examples on indirect effects all equations have continuous mediators and outcomes.
Here, I see that the nlcom's standard procedure of multiplying the coefficients and using the delta method to calculate the other parameters is valid (its just the same as in sem).
But does this also work with a continuous mediator and a binary outcome? Multiplying the coefficient from the linear response equation with the coefficient from the logit/probit equation just seems wrong.

How would you estimate the indirect effect in such an gsem model?

Thanks for any tips!
adson

[Maarten Buis]

That is an interesting problem for which several solutions have been proposed:

M.L. Buis (2010) "Direct and indirect effects in a logit model", The Stata Journal, 10(1), pp. 11-29. http://www.stata-journal.com/article...article=st0182

R. Hicks and D. Tingley (2011) "Causal mediation analysis", The Stata Journal, 11(4), pp. 605-619. http://www.stata-journal.com/article...article=st0243

U. Kohler, K. B. Karlson, and A. Holm. (2011) "Comparing coefficients of nested nonlinear probability models", The Stata Journal, 11(3), pp.420-438. http://www.stata-journal.com/article...article=st0236

M. Sinning, M. Hahn, and T. K. Bauer (2008) "The Blinder–Oaxaca decomposition for nonlinear regression models", The Stata Journal, 8(4), pp. 480–492. http://www.stata-journal.com/article...article=st0152

[sarry]

Hi, I´m using gsem, I need to get the indirect effect with bootstrapped standard errors. Anyone knows the commond to get the indirect effect?

[sarry]

Hi, I´m using gsem, I need to get the indirect effect with bootstrapped standard errors. Anyone knows the commond to get the indirect effect?

[tyk]

I wish I had seen this sooner! First, I'd get the coeflegend (gsem, coeflegend). Then I use the nlcom command (similar to lincom).

[tyk]

Page 5 of this explains it better than I do.