Mann-Whitney U and Bonferroni Correction

I recently presented some work in my department in which my colleagues and I had applied a Mann-Whitney U test for multiple pairwise comparisons. We used a Bonferroni-adjusted alpha for the test.

After my presentation, a guy generally considered to be a statistics whiz told me that the Mann-Whitney U test already accounts for multiple pairwise comparisons, so the Bonferroni correction was unnecessary. I cannot find validation of his statement anywhere; can someone please tell me whether this is true?

Also, would a Bonferroni correction be necessary when comparing multiple correlation R values?

Thank you!


Less is more. Stay pure. Stay poor.
I have not heard of what the whiz mentioned. I have seen and use a macro that adjusts p-values for pairwise comparisons. Leaving me to believe, beyond myself, others correct for pairwise comparisons. When thinking about how the M-W works, I don't see how it controls for pairwise. I typically use the Wilcoxon rank sum (synonym pretty much) and it ranks values and examines them, no correction?

Could he have been talking about the program automatically doing it??

For your later question, I typically don't see people correcting these. Not to say you are not suppose to, but if you were, I would think this would be a built in feature when generating correlation matrices or at least an option.


Less is more. Stay pure. Stay poor.
I have not read these, but they may help:

Dunn, O.J. “Multiple comparisons using rank sums”. Technometrics 6 (1964) pp. 241-252.

Steele, R.G.D. “A rank sum test for comparing all pairs of treatments”. Technometrics 2 (1960) pp. 197-207.


Thank you for your reply! For a moment, I too thought he might be suggesting that the program automatically corrects, but then I realized he never asked what program I used (nor did I provide that information). I will take a look at the articles you posted. I may also give the Wilcoxon test a try and compare my results.