We conducted a study in which the main outcome was a continuous variable and we had four independent study groups. During analysis, the distribution of outcome variable was skewed and we decided to use Kruskal Wallis test for the analysis. Now we would want to control for a binary potential confounder and do not know the most appropriate test. We think Two-way anova would have worked if the distribution was normally distributed and if there was equivalence of variants. Can some one help to advise us?
You will have to resample / bootstrap the distribution of the test statistic. Rank-based tests (like Kruskal-Wallis etc.) do not test the same hypothesis, and interactions in rank-based models cannot be interpreted in any meaningful way.
However, often a skewed distribution is an indication that asking for a mean difference might not be very wise. Often, it makes more sense to ask for the relative changes instead. This can be done using the very same tools with the logarithms of the data.
Hi Martin.
the equivalent may be , a distribution free test for general alternatives ina a randomized complete block, FRIEDMAN, KENDALL-BABIBGTON SMITH. or the based Friedman rank sums,, WILCOXON, NEMENYI, McDONALD-THOMPSON.
but, I advise you to take into account what @Jochen says, the procedure is distorted, not making the experimental design correctly.
Hi Jochen and Carlos,
Thank you both very much. I will take into account your advice
Thanks again
Martin
There is a test called Scheirer–Ray–Hare, which does not seem to be discussed much, nor well-respected, but it is included in Sokal and Rohlf. 1995. Biometry, 3rd ed.
There is also the aligned ranks approach. Unfortunately I haven't investigated this sufficiently, but it appears to be very promising. See: http://depts.washington.edu/madlab/proj/art/ .
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Molecular Methods
- Cloning