Is there any underlying assumption when we can reliably use Cohen's D? Do we assume our samples should be Gaussian? I know that Cohen's D is heavily biased by outliers and sample size. But is there any assumption on which it works?
Asked
Active
Viewed 42 times
0
-
Just some thoughts... Since Cohen's d is based on the difference in means and the pooled standard deviations, you would want to use it only in cases where these statistics make sense. If there are outliers or the distributions are very skewed, using means and standard deviation may not be the best way to express the effect size. I would think that non-Gaussian distributions that are symmetric would be fine, though. – Sal Mangiafico Oct 13 '22 at 13:18
-
How, exactly, do you contemplate "using" Cohen's d? For reporting a standardized effect size? For designing a sampling program? Something else? – whuber Oct 13 '22 at 16:46
-
1@whuber: I want to compare effect size between two populations (normal vs abnormal lab tests) where distribution is unknown. – Dexter Oct 14 '22 at 14:07