st: moderated mediation model with panel data

[Christine Scheef <christine.scheef@unisg.ch>]

Dear STATA users, 

I am  trying to fit a moderated mediation model with panel data where the 
mediator is at the same time the moderator. 
I tried to model it as a two-stage model where A influences B in the first 
step. In a second-step, I took the predicted values of B to estimate the 
model with the moderation (AxB) on C where A and B also have a direct 
influence on C. At the moment, I am using the command xtpcse for 
autocorrelated panel data to estimate both steps.
Unfortunately, in the second-step the R-squared decreases when including 
the moderation effect (AxB) even though the interaction is highly 
significant. I checked all models for multicollinearity, however, 
multicollinearity is not an issue.

I have two questions, is it appropriate to model a moderated mediation as 
I described above? And how may the decrease in R-squared be explained? Is 
R-squared a valid measure when calculating a two-step model?

I would appreciate any advice on the model and how to estimate it 
accurately.

Best,
Christine

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/