I am trying to figure out how the standard error of a indirect effect for a mediation model is calculated with WLS for categorical data and the PARAMETERIZATION=THETA.
Would you please tell me is it calculated using the following way?
If the indirect path is S-->M-->T and the estimated coefficients between S and M is a, with it's standard error SEa, the estimated coefficients between M and T is b, with it's standard error SEb, the standard error of the indirect path will be square root of (a^2*SEa^2+b^2*SEb^2).
In the general case you have to also involve the covariance between a and b. You get this from TECH3. We have a FAQ posted on this with reference to the MacKinnon book on mediation for the exact expression:
9/23/11: Delta method and Sobel
Cathe posted on Saturday, March 30, 2013 - 1:33 pm
I see, so here I need to include 2ab*cov(a,b), right? I am wondering can we directly get cov(a,b) from the output using Mplus code?