What is the effect of correlations among observed covariates or between observed and unobserved covariates on the quality of matching, when matching iss done using

  1. Propensity score
  2. (Euclidian/Mahalanobis) distance metric

Should one try to remove correlations by using Mahalanobis distance instead of Euclidean distance metric?

  • $\begingroup$ Effect of correlations on what? $\endgroup$
    – user318288
    Apr 20, 2021 at 22:40
  • 1
    $\begingroup$ The Mahalanobis distance does not remove correlations between observed and unobserved covariates. Either metric might work; try both and use the one that yields better balance. There is no reason to rely on theory when methods can be safely evaluated on their balancing performance. $\endgroup$
    – Noah
    Apr 21, 2021 at 3:12


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.