1
$\begingroup$

I have recently read some pieces suggesting that regression discontinuity designs could be the best statistical approach for causal inference stemming from non-randomized studies (eg 1 and 2).

However, as far as I am aware, the strongest claim for accuracy and validity has been made so far for propensity score matching (and possibly inverse probability of treatment weighting) (eg 3)

Thus, what is the comparative accuracy and validity of regression discontinuity vs propensity score matching?

Improve this question
$\endgroup$
2
  • 2
    $\begingroup$ If their assumptions are met, either is valid. The questions are: 1) how robust are they to violations of their assumptions; 2) a-priori, how likely is it their assumptions would hold in a given situation; & 3) how efficiently do they use the data you have? $\endgroup$
    – gung - Reinstate Monica
    Commented Apr 15, 2022 at 18:23
  • 4
    $\begingroup$ Those two approaches are for much different settings and neither of them "does" causal inference. They both do conditional causal inference, conditional on adjusting for the right variables. And there are many reasons not to do propensity score analysis, covered here. $\endgroup$
    – Frank Harrell
    Commented Apr 15, 2022 at 18:24

0

Reset to default

Your Answer

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