0
$\begingroup$

What effect size measure can be used for a goodness of fit test (given data and theoretical proportions)?

I read about the Cohen's w (https://en.wikipedia.org/wiki/Effect_size#Cohen's_w) but I also read (for example here https://rcompanion.org/handbook/H_03.html) that Cohen's w can be used only where the theoretical proportions are equally distributed among categories. Is it true, is it a strict bond or am I missing something? What other measure can be used?

Thanks

Improve this question
$\endgroup$
5
  • 1
    $\begingroup$ The goodness of fit of what are you measuring, and how? The term "effect size" comes from a context different from goodness of fit testing, and not all tests are associated to effect sizes, certainly not in a standard manner. That said, certain goodness of fit tests may involve something that can be interpreted as effect size. $\endgroup$
    – Christian Hennig
    Oct 26, 2020 at 10:47
  • $\begingroup$ The goodness of fit of a set of given frequencies, eg. c(25,34,73), with a set of theoretical relative frequencies, eg. (0.19,0.27,0.54). I'm using the chi-squard test. $\endgroup$
    – Luke
    Oct 26, 2020 at 11:08
  • 1
    $\begingroup$ "but I also read (for example here (...)) that Cohen's w can be used only where the theoretical proportions are equally distributed among categories" - this is not what is written there. Cohen's w can be interpreted as effect size, alright, although I wouldn't necessarily trust any fixed "small/medium/large" categorisation for any effect size. This is always context dependent. $\endgroup$
    – Christian Hennig
    Oct 26, 2020 at 11:34
  • $\begingroup$ So the equality of proportions only affects the categorization? $\endgroup$
    – Luke
    Oct 26, 2020 at 11:47
  • $\begingroup$ This is a hard question because I don't know what "effect size" means for you. Cohen's w does what it does. It involves a standardisation by the assumed proportions $p_{0i}$ whatever these are, but I can't know what exactly you think you have learnt when you call this "effect size". If the $p_{0i}$ are different, smaller $p_{0i}$-values have a stronger influence on Cohen's w, which may or may not be a problem. $\endgroup$
    – Christian Hennig
    Oct 26, 2020 at 11:54

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.

Browse other questions tagged or ask your own question.