Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: -doseresponse-
daniel klein <email@example.com>
Re: st: -doseresponse-
Fri, 6 Jan 2012 15:54:41 +0100
-doseresponse- is a user-writen ado file and you are asked to explain
where it comes from
(http://www.stata.com/support/faqs/res/statalist.html#stata). In this
case either from SJ-10-4 or from SSC (I did not check which is the
You are also discuraged to send exactly the same question twice
you did (http://www.stata.com/statalist/archive/2012-01/msg00178.html).
You migth be better of writing directly to the authors, but that is
not guaranteed either, as (at least some of) your questions are beyond
the scope of "technical support" and your best option migth therefore
be to do some literature research.
I cannot say much on the topic, but I would like to comment on your
third question. As far as I have seen -doseresponse- implements some
kind of propensity score matching (PSM) -- if not stop reading here.
As I understand it, the very purpose of using PSM is to make sure we
only compare what we have actually observed. If this cannot be done,
i.e. if the variables are unbalanced after matching, it cannot be
done. Including those variables in a regression framework after
matching does not solve the problem of extrapolating (i.e. comparing
what we did not observe), which is why we have chosen to use PSM over
the regression framework in the first place. So I do not really see
how it might be usefull to include unbalanced variables in the outcome
I am using -doseresponse-. I have the following questions:
1. It does not allow for the assessment of common support. Do you have
any suggestion on how to exam common support when the treatment is
2. It only conducts balance check after applying the GPS. How can we
know if the balance is improved or not? Or how to check balance before
applying GPS accordingly?
3. It is an ideal situation when balance is acheived on all
covariates. However, it is often the case that some covarites may
remain unbalanced. One strategy is to add the unbalanced covariate
into the outcome model together with the GPS. Is it easy to modify the
program so it can allow adding unbalanced covarites to the outcome
model? The original model only includes T and GPS and their
higher-order terms and interactions.
Your help is so greatly appreicated!
* For searches and help try: