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: Re: propensity score

From   "Ariel Linden. DrPH" <>
To   <>
Subject   Re: st: Re: propensity score
Date   Wed, 1 Jun 2011 08:01:50 -0700

The questions you are asking are basic to the design and structure of
propensity score matching. As a Stata list-serve, we mostly deal with issues
related to implementation of statistics in Stata and mostly assume the
person querying the list-serve has a grasp of the subject matter.

There are numerous papers and books on the topic of conducting a study using
propensity scored matching techniques. If you can't find them on your own,
we can certainly point you in that direction.


Date: Tue, 31 May 2011 06:29:50 -0400
From: zeynal karaca <>
Subject: Re: st: Re: propensity score

While digesting different component of propensity ADO codes, here is
another question on the same subject for which any help is

The research question I am working on is to tease out the differences
in treatment costs between two different insurance holders, say
Insurance A and Insurance B. I am using an administrative data for
this purpose. The process I am running is to select a set of Insurance
A beneficiaries for whom I can get a match from Insurance B
beneficiaries based on observed characteristics, call X1 and X2.

In the first stage, I am running a logit regression to determine
probability of choosing insurance A using X1 and X2 and get the
predicted probability, call Phat. In the second stage, GLS regression,
I am regressing the cost per beneficiaries on X2, X3 and Phat assuming
that the Phat will capture the differences in costs controlling for
other factors. So the naive question I have here is about X2, which
affects both the choice of insurance type during enrollment period as
well as the treatment cost associated with that beneficiary when
comparing two cohorts.

Any suggestions regarding how I should treat X2? Should it be included
in the second stage, GLS regression, of analysis, if so how to
overcome the case where right hand side variables are function of each
other? Are there other issues should I worry about while first running
pscore and then psmathc2?

As always, thanks in advance for any help provided.

Zeynal Karaca

*   For searches and help try:

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