st: BLUPs in a mediation model

[Jordan Silberman <silberman.stata@gmail.com>]

Hi,

I'd like to use BLUPs as a mediator and dependent variable in a mediation model.

Specifically, I'd like to test a mediation model in which experimental
condition (treatment vs. control) predicts rate of change in a
hypothesized mediator, and change in the mediator in turn predicts
change in the dependent variable. For each subj, the mediator and DV
were measured on multiple occasions (not always at the same times). To
get subject-level estimates of rates of change in the mediator, I used
xtmixed to estimate a model in which time predicts the mediator, and
with a random slope for time allowing each subject to have a unique
rate of change. Then I computed BLUPs as subject-specific measures of
rate of change in the mediator (fixed effect of time + random effect
of time for a given subject). I used the same procedure to compute
BLUPs as a subject-level measure of rate of change for the DV.

My question is this: is there anything that's not "kosher" about using
BLUPs as a mediator and DV in a mediation model? I haven't seen this
done before, but I don't see any reason why I can't do it. Obviously
there's a risk of non-normality of BLUPs, but I can of course
transform and bootstrap as necessary.

Other options would be a multilevel mediation or latent growth curve
mediation model, but neither of these will work, for somewhat
complicated reasons.

Any objections to the BLUPs approach?

Many thanks,
Jordan
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/