topic Re: Kruskal-wallis test significant at p = 0.003 and Dunn´s test with no significant results in Statistical Procedures

Kruskal-wallis test significant at p = 0.003 and Dunn´s test with no significant results

mdsaraiv — Fri, 17 Jun 2016 20:23:29 GMT

A friend of mine asked me if I knew why a kruskall-wallis test of scores had significant p value of 0.003 and the Dunns test showed no statistical significant results.

One person told us to look at quartiles to see the difference. Is it ok to do that? An why Dunns test would not show where were the differences?

Thanks so much.

Maria

Re: Kruskal-wallis test significant at p = 0.003 and Dunn´s test with no significant results

SteveDenham — Tue, 28 Jun 2016 13:02:52 GMT

In Dunn's test, one group is designated as a 'Control' group, and each of the other groups are compared nonparametrically, with an adjustment for the number of comparisons. The K-W is an overall test, that at least one group differs from the others in location.

So you have two things that could cause what you see. The first is that the differences in location may not involve the control group. The second is that the adjustment for multiple comparisons is such that a 'raw' p value may be significant, but after controlling for the number of comparisons, the adjusted p value is not significant.

Steve Denham

Re: Kruskal-wallis test significant at p = 0.003 and Dunn´s test with no significant results

mdsaraiv — Wed, 31 Aug 2016 00:09:54 GMT

i just notice I forgot to thank you Steve, sorry. Thanks so much!