st: Testing for binary mediation using logit and survey design adjustments (weights, etc.)

["G. Vazquez" <academic.demographer@gmail.com>]

Hello Statalisters,

Following work by Kenny, and information from
http://www.ats.ucla.edu/stat/stata/faq/binary_mediation.htm, I used
Michael Mitchell & Phil Ender's -binary_mediation- command to test for
mediation using a binary outcome variable (and continuous mediator,
binary independent and a slew of covariates).

However, because the survey design used multi-stage stratified
sampling, the coefficient estimates (especially for the independent
variable) are substantially different (direction even reverses) when
survey design is adjusted using the - svyset- command versus when the
regression is run without design adjustments (pweights, psu, strata
are all specified).

Would it be possible to test for mediation by hand-calculating the
effects using coefficients from models with the svy adjustments, for
example using Nathaniel Herr's spreadsheet at
http://nrherr.bol.ucla.edu/Mediation/logmed.html? It seems there would
also be some problems doing that; first, the covariance of the
independent and mediator variables is needed.

Any clarification about whether there is some way to test mediation
effects given a binary outcome variable and the need to use pweights,
psu, and strata design specifications would be greatly appreciated.

I use Stata 10.

Thank you in advance,

Georgiana
University of California, Irvine
*
*   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/