Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

Non Parametric Independent Test effect size

edited January 2021 in JASP & BayesFactor

Hi everyone!

I'm performing a research using JASP, I already have the results but I'm struggling in how to interpret the effect size:

I just want to ask about the effec size metric used in the bayesian version of mann whitney's U,

is it the cohen's D? and in that case:

since mann withney's test was meant to be used with non-parametric data, why does the effect size is measured with cohens'D and no with serial rank correlation or another non-parametric effect size metric?

and how is the best way to interpret that effect size (big, small, medium)?


:)

Thanks in advance!

Comments

  • AKAIK there is no effect size for the Bayesian Mann-Whitney test - the W is the test statistic.

    In the non-Bayesian, the rank-biserial correlation is indeed given.

  • Hi tourette95,

    As a result of the approach used to enable Bayesian inference for the Mann-Whitney test, we obtain a posterior distribution for the standardized effect size cohen's d, but on the latent level. You can read more about this in the corresponding paper (https://www.tandfonline.com/doi/full/10.1080/02664763.2019.1709053), although it can be a bit technical. I am planning on writing a blog post that explains some of the rank-based Bayesian analyses that explains the underlying technique.

    I will also add the rank-biserial correlation to the Bayesian output.

    Kind regards,

    Johnny

  • Thank you so much!!! :)

Sign In or Register to comment.