# Meta-Analysis Resources

Tools for Those Who Summarize the Evidence Base

Resources and networking for those who conduct or interpret meta-analyses related to any phenomenon that is gauged in multiple studies.

# Mann Whitney U and Kruskal Wallis values into Cohen's d AND merging multiple outcomes

Hi everyone,

I just wonder how to convert Mann Whitney U and Kruskal Wallis values into Cohen's d. Any formulas or references you may recommend?

A second question: Is there a way to merge (combine) 2+ subgroup into one group? e.g.

Variabels : Single/Married/divorced...Each has mean+std dev+n

Can I merge single+divorced to get one group ?

Let's say GROUP-1 [SINGLE+divorced] and GROUP-2 [MARRIED]

Sample:

Variable         n        X      s

single           124   1,54   0,63

divorced       65    1,24   0,49

married        132   1,87    0,8

Can it be like ?

Variable                     n          X          s

single+divorced      189       1,39    0,56

married                   132       1,87      0,8

or can I combine them using weighed average taking n into accaunt?

or first calculate Cohen's d for both groups then get average of two Cohen's ds?

Views: 794

### Replies to This Discussion

Re Question 1: One thought is finding the exact p value for the U or Kruskal-Wallis values and then converting them to a regular t or F and converting that statistic to the SMD (Cohen's d).

Re Question 2: I doubt you get very different results either way, but I usually prefer to combine the means and SDs (appropriately weighted for their sample sizes) and then calculate the SMD. Of course you would pool the relevant SDs. (The old dstat 1 had a handy feature for just that purpose.)

Mr. Johnson

Thank you so much for your assistance.

Blair T. Johnson said:

Re Question 1: One thought is finding the exact p value for the U or Kruskal-Wallis values and then converting them to a regular t or F and converting that statistic to the SMD (Cohen's d).

Re Question 2: I doubt you get very different results either way, but I usually prefer to combine the means and SDs (appropriately weighted for their sample sizes) and then calculate the SMD. Of course you would pool the relevant SDs. (The old dstat 1 had a handy feature for just that purpose.)