Assume $x_1, x_2, \dots, x_n$ belong to group $A$ and $y_1, y_2, \dots, y_m$ belong to group $B$.

We want to test if values in group $A$ are smaller than values in group $B$ as values in B.

I design the test statistics similar to Kendall's Tau by getting the pairs of $\mathrm{sign}(x_i-y_j)$ for $i= 1, 2, \dots, n$ and $j= 1, 2, \dots, m$.

As there are in total $n\times m$ pairs, let $$T= \frac{\sum_{i,j} \mathrm{sign}(x_i-y_j)}{m\times n},$$ where $T$ is between $[-1, 1]$.

My questions is that how do I get the test significance for $T?$

  • $\begingroup$ If any of the $x_i$ equal or exceed any of the $y_j,$ wouldn't that contradict your hypothesis? What, then, would be the point of $T$? $\endgroup$
    – whuber
    Commented Jun 18, 2020 at 19:30

1 Answer 1


For continuous rvs X and Y (i.e. absent ties), your test statistic is exactly equivalent to a Wilcoxon-Mann-Whitney test - indeed, it's linear in the usual U statistic.

So the easy way to test it is to use a Wilcoxon-Mann-Whitney with the same direction of null and alternative, the p-value should be the same.

Or you can just do a permutation test of your statistic.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.