If I have two groups (15 subjects in each group) and the data is not normally distributed what test do you advise me to use? Mann-Whitney U or Kolmogorov-Smirnov Z? I have seen that in almost all studies that used non-parametric tests they used Mann-Whitney U. Any opinions? I want to compare the results of the two groups (the results show scores from a quality of life questionnaire - the scores are from 0 to 100).
Dear Holubiac Iulian Ștefan
Both KS and MW are appropriate for your goal, unless it seems that MW is more common. Maybe in some fields there is a preference to use one over the other. So as you said, follow the mainstream.
Best,
If you're also curious to learn where/how the two distributions differ (beyond simply *if* they differ), then you could use the recent multiple testing methodology from the Journal of Econometrics article
Article Comparing distributions by multiple testing across quantiles...
for which there is corresponding R and Stata code at
https://faculty.missouri.edu/~kaplandm
as well as a corresponding Stata Journal article
Article distcomp: Comparing distributions
The error rate control is exact even in very small samples (like you have).
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude