0
$\begingroup$

I'm working on propensity score matching, but I'm experiencing some problems. I am using the "propensity score matching" tool in SPSS v. 24 (but we've tried the same in R and "manual" by logistic regression). After PSM, surprisingly, my two treatment groups are even more different in terms of baseline covariates (age, % male, etc.) then they were before matching. Furthermore, the mean propensity score is also different between the treatment groups. So I suppose something is not going right here...?

I am matching on about 20 covariates. All of them, except age and eGFR at baseline, are coded as 0 and 1. I have experimented with this, leaving out some of the covariates, adding some more, but nothing really seem to work.

I have about 10.000 people in treatment group A and 90.000 people in treatment group B, so I think the amount of patients should not be the problem...

Did anyone else experience these problems and may have a suggestion where this goes wrong? Do I have to use a different method?

Thank you for your help!

Improve this question
$\endgroup$
1
  • $\begingroup$ Your data may be succumbing to the issues in this paper: gking.harvard.edu/publications/… What happens when you try full matching or weighting? $\endgroup$ – Noah Nov 2 '16 at 23:13
0
$\begingroup$

You would need to provide more information to figure out the problem. And please clarify, there is a PSM extension written by IBM (me, actually), but there is also a third-party tool for the same purpose. Which one are you using?

For starters, please provide the syntax and the Viewer output for the procedure. If you want to do this offline, you can send them to jkpeck AT gmail.com.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.