Logit mediation

[Johannes Muller]

Dear all,

I am looking for the best way to conduct a mediation analysis when the outcome is dichotomous. Specifically, the DV is binary, MV continuous, and IV continuous.

As far as I can tell, SEM cannot handle this, and neither can most of the user-written programs it seems (e.g. khb or sgmediation2).

GSEM is well suited for the task, yet once I try to obtain the bias-corrected CIs the bootstrapping does take a very long time (i.e. several days).

Is there any alternative that would allow me to obtain bias-corrected CIs in a quicker way?

Would be grateful for any help and suggestions!

[Mike Lacy]

Can you elaborate on the problem in using -khb-? Here's a (contrived) example using it to estimate how much vehicle weight mediates between vehicle length and a binary version of mpg.

sysuse auto
gen byte lowmpg = (mpg < 20)  // just for illustration
khb logit lowmpg length || weight

I'd also mention that while -khb- is not slow for bootstrapping, a Stata Journal article I did with some colleagues reported that for the more difficult case of ordinal logit models, the conventional (formula-based) confidence intervals that -khb-gives for the mediation estimate worked quite well, so bootstrapping is probably not needed.

[Mike Lacy]

Whoops, I should have mentioned that -khb- is a user-written program, see -ssc describe khb-.

[Johannes Muller]

Dear Mike - you are absolutely right of course! Many thanks, this has been helpful.