I was wondering if there's a proven/known relationship between RI p-values and OLS p-values in RCT settings, where treatment is completely random. Alwyn Young's paper1 shows many examples of OLS p-values that were smaller than RI p-values, but is this a coincidence or broadly more true?

And if RI p-values can be smaller than OLS p-values, what would be the reason for that? Young argues that high leverage points cause OLS standard errors/p-values to become smaller than they should be, so is there a similar explanation for why OLS errors/p-values would be too large compared to RI?

Thank you!

1 Young, Alwyn. "Channeling fisher: Randomization tests and the statistical insignificance of seemingly significant experimental results." The quarterly journal of economics 134, no. 2 (2019): 557-598.

  • 3
    $\begingroup$ It would be helpful if you provided a link or a more informative citation of the paper you are referring to. $\endgroup$
    – Kitsune Cavalry
    Feb 6 at 22:46
  • $\begingroup$ Adding that now, thank you for flagging! $\endgroup$
    – msc
    Feb 7 at 15:09
  • $\begingroup$ Even better to have both, so that one may learn what it is without having to download a pdf. $\endgroup$ Feb 7 at 15:16


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.