Hello,
I have noticed (what I think) is a problem in the working of the sensatt
routine.In the course of implementing sensitivity analysis I specify a command
such as that provided below:
sensatt log_wages_hrs train_combined, reps(1000) p11(0.4390) p10(0.4390)
p01(0.4010) p00(0.3010) pscore(prop_score2)
where log_wages_hrs and train_combined are the outcome and treatment variable
respectively and prop_score2 is the propensity score variable that I provide to
the program. The problem seems to be that the estimated ATT and standard error
are exactly the same irrespective of both the values of the parameters pij
provided as well as the number of iterations specified(and are equal to the
baseline ATT estimates).
If, however I modify the above command such that:
sensatt log_wages_hrs train_combined ability_score male O_level_comp ,
reps(1000) p11(0.4390) p10(0.4390) p01(0.4010) p00(0.3010)
and allow the sensatt program to calculate the propensity score for itself
(using the varlist ability_score, male and O_level_comp), I obtain very
different estimates of the ATT and the standard error after the simulation of
the confounder variable( even though the baseline ATT is the same, as in the
command for the user provided propensity score). Moreover, both the computed ATT
and the standard error vary as the pij parameters change. These results seem
very strange to me. Has anyone else faced such problems while implementing sensatt?
Murali Kuchibhotla
*
* 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/
