Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

re: st: psmatch2--coefficients on covariates? interactions?

From   "Ariel Linden, DrPH" <>
To   <>
Subject   re: st: psmatch2--coefficients on covariates? interactions?
Date   Sun, 17 Apr 2011 12:40:18 -0700

Hi Joanne,

Once psmatch provides you with the matches (I am assuming you are running a
1:1 match), you can run a regression Y = treatment variable if _weight !=.
and derive your estimates. If you have a 1:k matching strategy you'd add the
weight in the equation to account for multiple controls. Then you'd run the
same models without the "if statement" above to provide the stats on the
unmatched population for your table.

As for the second question about interactions: you can certainly include
interactions if they make sense (as with any statistical approach, it helps
to think about the purpose of the strategy and what underlying assumptions
you expect to explore). If you are finding that the model perfectly predicts
treatment, I would question if indeed there is an overlap in covariates (as
one would expect in the propensity score approach)? You need to explore the
data and find out what is going on? Are the treated subjects drastically
different then the non-participants?

If you really feel that a more complex propensity score model is required,
you should consider boosted regression using the logistic model option
(findit - boost -).

I hope this helps


Date: Sat, 16 Apr 2011 13:09:35 -0700 (PDT)
From: gradstud <>
Subject: st: psmatch2--coefficients on covariates? interactions?

Dear Statalist,

Is there a way to obtain the coefficients on the other covariates included
in the propensity matching model (using psmatch2)? I would like to display
these on a table to compare with my OLS coefficients.

Is there a way to include interactions with my treatment variable in the
matching model? The interactions drop because they perfectly predict the
outcome in the logit model (as they are interactions with the outcome). I
could center them but they are dummy variables so that seems odd.

Any help is greatly appreciated!


*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index