Bookmark and Share

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: re: st: Propensity score matching


From   "Ariel Linden, DrPH" <[email protected]>
To   <[email protected]>
Subject   re: re: st: Propensity score matching
Date   Thu, 19 Jul 2012 12:04:23 -0400

I am responding to this response that was sent to me off list:

Amir,

I am not sure what you mean by "specific indicator was calculated for a
small portion of the firms""? Do you mean that this is how the treatment
group is defined, or some other measure (baseline or outcome)?

Basically, you would match on the variables that are available for both the
"treated" and "untreated population" (except for the outcome you are looking
at). In health care research this approach is typically used for cohort
studies (where we try to replicate an RCT on observable characteristics and
assume that the unobserved characteristics are not large enough to bias the
results). It could also be used in a case-control scenario where we know who
got the outcome and who didn't, and we try to determine which factors
differed between the groups (assuming that the "factor" is what contributed
to the differential outcome).

I am not sure which scenario best fits your problem. 

I think this may be a bigger issue than what could be reasonably answered
via the statalist. My suggestion is to consult a statistician/economist to
help you with your study.

Ariel 





;*********************

Dear Sir/Madam
 
 Dear Ariel
I just saw that you answered one question regarding propensity score
matching. I appreciate if you give me some ideas regarding my matching
problem as I did not get anything from Statalist. I have a sample for which
the specific indicator was calculated for a small portion of the
observations (firms). However, this indicator is very important but cannot
be calculated for the whole sample due to the availability of data. I just
want to know that is there any matching technique that I can use to find the
matching firms in the whole sample to the ones that I calculated the
indicator for, and then use the same indicators for them as they are
matched.
 
Thanks In advance for your help.
 
Best Regards,
 
Amir Khorasgani     

The simple answer is that matching is intended to create balance between
treated and controls on pre-intervention covariates. 

For your situation (as I understand it), the companies who issue bonds
should be comparable to those firms who do not issue bonds. Here
comparability is determined by ensuring that they balance on covariates that
are measured before they start issuing bonds, since the "issuing of bonds"
appears to be the intervention. 

Your additional comments "both sample varies between 2007 and 2011" and "how
should the dataset look" gives us no meaningful information to help you. You
should think carefully about how you pose a question if you would like to
get a meaningful response.

Ariel 

Date: Wed, 18 Jul 2012 19:57:53 +0100
From: "KASHEFIPOUR E." <[email protected]>
Subject: st: Propensity score matching 

Hi,

I have a question regarding the PSM estimation. I want to investigate the
companies who issue bonds between 2007 and 2011, so this group would be as a
treatment sample. Then I want to match the treatment group with a set of
companies who do not issue bonds during the observation period (2007-2011).
I was wondering how I could match them as both sample varies between 2007
and 2011? How should the dataset look like?

Best,
Eln



*
*   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/


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