We are interested in a mediation model where the outcome is time to
event with censoring and wonder whether Stata/gllamm will work.
Following MacKinnon and Dwyer (Evaluation Review, 1993), we are using
the "product of coefficients" method where the coefficients are obtained
from three separate regression equations, denoted a, b, and c, where a
is the coefficient for the path from the independent variable to the
mediator, b is the path coefficient from the mediator to outcome, and c
is the path from the independent variable to the outcome. c is called
the "direct effect" and the product ab is the "indirect effect". I'm
curious to find out if our results from the three separate regressions
hold up in a path model when all coefficients are estimated together.
