I am reading an unpublished paper that employs diff-in-diff to firm data containing a total of 2598 observations. Of these observations, 2474 are in the treatment group and 124 in the control group.

I read here that the control group should not contain less than 20% of the total sample. But clearly, in this published paper the control group contains less than 5% of the sample. Also, the absolute size of the control group also looks small to me.

I would like to understand if the results of the diff-in-diff in this paper can be reliable given that the control group contains less than 5% of the sample.


  • 1
    $\begingroup$ Their effective sample size is pretty much that of the control group, so not a lot. It is also possible that there are all sorts of positivity violations in the propensity score with such little control data (depending on the reasons why they have so little data). It also depends on what parameter they go for. If they estimate something like the effect among the treated then the lack of data for the control group is less of a problem, although still an issue. $\endgroup$
    – user327671
    Aug 9, 2021 at 4:05
  • $\begingroup$ @LarsvanderLaan they compare the reaction of the treated firms' stock price with the control firms' in response to some news announcement. $\endgroup$
    – Anup
    Aug 9, 2021 at 5:20


Your Answer

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