Which Wilcoxon test to use?

Written Jan 30, 2018

Wilcoxon rank sum? Wilcoxon signed rank? Wilcoxon one-sample? Two sample? Mann-Whitney U? If you are confused, read on.

All of these are basically non-parametric versions of the t-test.

That is, you want to check if two groups of numbers have a similar distribution/center (without the normal distribution assumption in a t-test).

Here is a quick guide that I made for myself (and you!):

Wilcoxon rank sum (same as “Mann–Whitney U test”)

Basically, like your standard t-test to compare two samples from two different populations.

In R:

wilcox.test(df$X, df$Y)   # both numeric
wilcox.test(y ~ A)        # A is a group/factor

Wilcoxon signed-rank

Use when comparing two related samples, matched samples, or repeated measurements on a single sample.

In R:

wilcox.test(df$Y1, df$Y2, paired=TRUE) 


A handy R reference for this: