I am using the package MatchIt in R to perform propensity score matching. I have chosen to use nearest neighbor matching with a caliper of 0.2 and since in my case i have more cases than controls i have to use the replacement=TRUE option, so that a control can be used more than once.

The graphical histogram check is satisfying and the stand.mean differences are all small with a max of 0.03 (btw any other suggestions for testing the matching?) I want to use the matched dataset to check the treatment effect after all the matching(perform logistic regression with mortality as outcome and treatment as explanatory variable now) and i am wondering if i should take into consideration the weights that were resulted from the matching. Since i used the replacement option not all observations have a weight of 1 anymore. Shall i use this somehow or can i just perform an unweighted final logistic regression on the matched data to estimate the effect of treatment.

  • $\begingroup$ standardized mean differences should be used to check balance, along with plots of the distribution of the propensity score in the two grops (eg density plots). stand mean diff less than 10% is considered good match! p values should not be used. u dont need to use the weights. If you have performed a matched analysis and thei appear to be comparable at baseline (i.e no differences) then compare them directly... $\endgroup$ – Adam Robinsson Feb 16 '15 at 15:05
  • 1
    $\begingroup$ Hey Adam, i did a bit more research on this topic and based on this paper Elisabeth Stuart pages 9, 13 , when analysis is performed with replacements the weights should be taken into account, although it is still unclear to me how.Any suggestions would be appreciated $\endgroup$ – Barrett Mar 2 '15 at 16:01

Your Answer

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

Browse other questions tagged or ask your own question.