Cohen's d (or Hedges' g) are often used to compute effect size. They rely on the assumption of homogeneity of variance across samples however. Because of the pooling of variance that they do, I'm also in doubt that they can be useful in the situation when one want to compare groups in generalized linear models with non-Gaussian distribution of errors (Poisson for instance), possibly overdispersed for some groups.

Am I right, and if yes, what should be an adequate measure of effect size in this situation ?

Ideally I'm looking for a metric which could be used for both binomial and Poisson or negative binomial error distribution models.

  • $\begingroup$ This book may be useful. I read the first edition many years ago. $\endgroup$ – Peter Flom - Reinstate Monica Jul 19 '13 at 10:31
  • $\begingroup$ @PeterFlom. Thanks. I'll try to get a copy of this. $\endgroup$ – Jehol Jul 22 '13 at 9:59
  • $\begingroup$ did you ever figure this out? I am trying to calculate effect size for a GEE with negative binomial distribution $\endgroup$ – PTSDresearcher Mar 21 '15 at 19:05
  • $\begingroup$ Nope, unfortunately :-( $\endgroup$ – Jehol Mar 23 '15 at 8:17

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.