Propensity Score Matching with Replacement

[derek_cruik]

Hello,

I'm working with the PSMATCH procedure and testing out different matching methods.  When using the REPLACE matching method (k=1), I was wondering if there is a way to limit the number of times a single control case can be used to be matched with treated cases?

Looking at my matched output, I have 95% of all treated cases being matched with 6 or less control cases; however, I do have some extreme observations where a single control case is used to be matched with 100 different treated cases.  Is there any way that I can limit it so that doesn't happen within the procedure itself, or is this something that I may need to look into post-match?

Any insight or help is very much appreciated.  Thanks!

[SteveDenham]

The only thing that comes quickly to mind is to loosen the matching criteria.  It sounds as if the only "possible" matches for the cases are the selected controls.

SteveDenham

[derek_cruik]

Thanks for your reply. I've already turned the caliper off when matching with replacement. I know the whole concept of matching with replacement is so that you can keep all of the treated cases in your match output.

I've seen that there are some potential macros that have been built that may be able to limit the number of times a control case can be used to match to a treated case, but it doesn't appear that it would be using the PSMATCH procedure so I would need to look into creating the variable assessment output myself.

Thanks again,
Derek

[SteveDenham]

I think you are right. A method might involve something like:

Use propensity scoring to match until a record is used a maximum number of times (say 5).

Remove that record from the controls data.

Repeat as needed until all cases are matched.

This would result in a stratified matching, with the matching getting worse as you go along.  Your subsequent analysis should take this into account somehow.

SteveDenham