Dear Manuela,
Usually the propensity score is not used to assess its "contribution" in terms
of the outcome, or to assess whether it is a confounder. Instead, the propensity score is
generally used to reduce confounding due to a set of other variables that are known (or
suspected) to be confounders. You would use the confounders to estimate the propensity
score, and then do matching, subclassification, or weighting using the propensity score in
order to reduce bias in the estimated treatment effect due to those confounders. There is
additional bias reduction that can usually be obtained if you combine that matching,
weighting, or subclassification with additional adjustment for the confounders (e.g.,
through regression adjustment in matched samples).
I hope this helps, and you may find the following papers helpful in terms of thinking
about what propensity scores are used for:
http://gking.harvard.edu/files/abs/matchp-abs.shtml
http://projecteuclid.org/DPubS?verb=Display&version=1.0&service=UI&…
Liz
On 4/27/09 6:58 AM, "Manuela Expósito Ruiz" <ilonam10(a)gmail.com> wrote:
Hi, I have a basic question for the use and interpretation of propensity score (ps)
When I add de ps lika an independent variable for the multivariate regression model, when
can I translate its contribution? Must I use the treatment and the ps in the same model
and the rest coavariables using for the ps estimation? If ps is significative in all model
of multivariate regression, it´s ps a confounding variable?
Thanks
Manuela