Re: st: propensity score matching

[Steven Samuels <sjhsamuels@earthlink.net>]

On May 8, 2008, at 3:51 PM, Charman, Susan wrote, in part.

> 1. I was under the impression that "standard" nearest neighbour  
> matching is done without replacement (matches removed)
Morgan and Harding (2006, p. 34) suggest that, among the basic  
algorithms, with-replacement nearest neighbor matching is superior to  
without-replacement matching. So, the default choice in -psmatch2- is  
probably deliberate.


> and yet psmatch2 creates the _weight for which I get very high  
> values ( e.g n=45) . this suggests that a subject is used 45 times  
> as a match ( there are other subjects available with the same  
> propensity score but not used).
Perhaps this, from the -psmatch2- help, is relevant:

"Note that the sort order of your data could affect the results when  
using nearest-neighbor matching on a propensity score estimated with  
categorical (non-continuous) variables. Or more in general when there  
are untreated with identical propensity scores."


Reference:
Stephen L. Morgan and David J. Harding Theory and Practice Matching  
Estimators of Causal Effects: Prospects and Pitfalls in Sociological  
Methods Research 2006; 35; 3-60


-Steve
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/