Is Wilcoxon test appropiate for the comparison of large, independent, nonnormal datasets?

Hi all!

I am using 'ranksum' function (Wilcoxon test) to compare two independent, nonnormal, large data sets. However, p-value is 0 I think because of the effect of large data size. Any other test statistics that may handle such a large population analysis?

Thanks in advance :)

Answer 1

[Star Strider]

Since we’re likely discussing the lognormally distributed data you previously described, and since this is an unpaired comparision, ranksum is appropriate. A p value of 0 is an excellent result if you want to demonstrate that any two results ar different, since that indicates that they are and that the calculated probability is below the ability of floating-point arithmetic to calculate any other value ().

You can also use friedman to do multiple comparisons.

[Sara Woods]

Thanks again @Star Strider, maybe I should learn about multiple comparisons (e.g., Friedman test) and corrections (e.g., Bonferroni). I am using 'ranksum' because it's simpler to code! But maybe it's appropiate to strengthen my statistical analysis

[Star Strider]

As always, my pleasure!

Using ranksum is correct for a comparison of two vectors. I would use multcompare specifying 'friedman' in the ‘stats’ structure to do multiple comparisons. For 'CriticalValueType', I was always taught to use 'scheffe', however that may reflect the comparisons I was doing.