Hello!

I had some questions in trying to apply CEM to mediation analysis.

SEM methods work fine with the CEM weights so that on the surface, mediation seems easy. However, I am not sure that the standard errors are correct. In mediation analyses, it is recommended to do bootstrapping instead of using the output z-values and corresponding p-values as mediated variables often do not follow normal distributions to be highly conservative.

I have thought of doing bootstrapping with the weights just on the matched sample but am not if this is the correct approach. For example,

https://www.stata.com/statalist/archive/2012-02/msg00822.html suggests that bootstrapping needs to be done more complicatedly in order to correctly incorporate variability.  

In sum, I would like to get some feedback on how to correctly do mediation (and mediated moderation) analysis using CEM. Even if complete answers are not possible, I would still truly appreciate any kind of related information.

Thank you in advance.

Best,

Yon Soo Suh