McNemar and rank-biserial correlation

PhilJ

Hi guys, I notice that your non-parametric paired samples test (Wilcoxon signed-rank) uses the rank-biserial correlation as a measure of effect size. Could you point me to some literature to indicate this is a defensible measure? I know there is a way to calculate a Cohen's d if you know the standardised test statistic and N (https://www.sheffield.ac.uk/…/f…/stcp-marshall-WilcoxonS.pdf) but that this is also not an ideal way to proceed (http://core.ecu.edu/…/w…/docs30/Nonparametric-EffectSize.pdf), so it would be good to get something to support the R-BC instead.

Also, does JASP offer the McNemar test, or an equivalent?

JohnnyB

Hi Phil,
The choice for effect sizes for the two wilcoxon tests is based on the following paper:

• Kerby, D. S. (2014). The simple difference formula: an approach to teaching nonparametric correlation.
  journals.sagepub.com/doi/full/10.2466/11.IT.3.1

So the effect size for the rank sum test is the rank biserial, whereas for the signed rank test it's the matched rank biserial correlation. You're right that most literature omits Wilcoxon effect sizes, but these coefficients seem to be well-researched, standardized and intuitive.

As for McNeymar test, we are planning on incorporating this in JASP, but unfortunately Im not sure on the time frame.

Kind regards,
Johnny