3
$\begingroup$

I've got a problem with some count data of mine.

I am using a cohen's D/ Hedges G to compare the effect size between two different treatment groups in my experiment, and then repeating that process for a different strain of fly with the same treatment, and comparing effect sizes for the two different groups to see if effect of treatment varies by strain.

However the actual data I get out at the end is count data in a poisson distribution, not a normal distribution.

Is there an effect size coefficient that is designed for poisson distribution, or should I just stick to these ones and take it with a pinch of salt?

$\endgroup$

1 Answer 1

1
$\begingroup$

Effect sizes only compare means (or whatever, like $\eta^2$). They make no assumptions about distributions. So they are applicable regardless of the conditional distribution.

Of course, their interpretation hangs on whether the means are the only thing we are interested in. Some events may have a tiny probability but a huge impact if they do occur, so the mean is very small, but in such a case, we should also be concerned about the variance.

$\endgroup$
3
  • 1
    $\begingroup$ A bit of counterpoint: because an effect size is traditionally defined using a standard deviation as the reference unit, the concept implicitly does make a (mild, nonparametric) distributional assumption: namely, that the SD is suited to this purpose. As the distribution grows further from Normal--particularly as it becomes more skewed--that assumption is worth questioning, as the OP indicates. $\endgroup$
    – whuber
    Nov 9, 2022 at 15:02
  • $\begingroup$ @whuber: yes, I agree. I tried to convey this in my second paragraph. $\endgroup$ Nov 9, 2022 at 15:04
  • $\begingroup$ I didn't get that message from your paragraph because "concerned about the variance" could mean many things. $\endgroup$
    – whuber
    Nov 9, 2022 at 15:07

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.