I have data that represents treatments applied to a given number of subjects in two or more groups. After each treatment, the subjects in each group are appraised to be of quality 1-5 and the number in each group having each ranking are counted.
Some mock data for this process:
| Treatment | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| a | 1 | 1 | 20 | 0 | 10 |
| b | 0 | 5 | 13 | 3 | 0 |
| c | 1 | 4 | 12 | 4 | 0 |
I use the Kruskal-Wallis test to determine if the medians of any of these groups are significantly different from the others. <
However, if I were able to use something like a Chi-square or t-test, I could then calculate an effect size for each treatment and then determine what sample size would be necessary reliably pick up such an effect size.
Is there a method that would allow me to do this (effect size calculation and statistical power/sample size determination) for the KW test?
