Hello Temidayo,

Since the Kruskal-Wallis test operates on rank (ordinal) values, the Tukey 5-number summary (minimum, lower hinge, median, upper hinge, maximum) for each batch would be a good tabular summary. For a pictorial summary, the grouped boxplot contains virtually identical information.

Good luck with your work!

@David Morse gave you the optimal and unique answer. I agree with him

@ David Morse, Thanks for the precise answer. So we should use mean ranks in the discussion or the medians?

Hello Temidayo,

If you have the original scores (not ranks), those should be used for presenting the summary statistics overall and by group/batch. If not, and all you're discussing is the K-W test, it won't make a lot of difference whether you use mean or median rank (they tend to track very closely after a rank transformation). But, in keeping in the spirit of ordinal values, median is better than mean.

I agree with @David. The original scores could be used to produce the box plots, which quickly enable you to compare the medians. Box plots also provide information on the spread or dispersion. You may interpret whether the data from different groups have similar distributions.

Then, you can proceed with the K-W test and any post-hoc analysis.