# Is my interpretation for Wilcoxon Signed Rank test correct?

I ran a wilcoxon signed rank test on paired samples where the outcome variable was a test score. The samples were paired by as siblings (younger and older sibling). I have a problem interpreting the result... So far I've read different kinds of interpretations online and even on stack exchange. Please correct me if my interpretations as well as my hypothesis are wrong:

Null hypothesis- The median of the math test score differences of the older sibling and the younger sibling group is 0.

P-value = 0.08 *I came up with two different interpretations from what Ive seen from other places.. (1) At a significance level of (alpha) 0.05, we fail to reject the null hypothesis and conclude that the median test score of the younger sibling group is NOT significantly different from the median test score of the Older sibling group.

OR

(2)At a significance level of (alpha) 0.05, we fail to reject the null hypothesis and conclude that the median of the score differences between the paired samples is NOT significantly different from 0.

Am I interpreting it right?

What is being tested

Your (1) indicates that you tested for 0 difference in population medians and your (2) indicates that you tested for the population median of pair-differences being 0.

Strictly, a signed rank test is not testing what either of your interpretations say.

The actual population quantity being considered is the pseudomedian of population pair-differences (across all population pairs); the test relates to the sample pseudomedian of pair-differences (1-sample Hodges-Lehmann estimator), and the corresponding estimate is based on it.

If you have symmetry of pair-differences that will be the same as the second interpretation, but the test only needs that symmetry under the null; it can be perfectly easy to interpret in a wide class of cases when you don't have symmetry under the alternative.

[For example consider pairs of positive observations with no change in distribution under the null and a scale shift under the alternative. The test is perfectly suitable under both the null and alternative but in general under the alternative both your interpretations of what is being tested are going to be wrong.]

Consequently, you cannot easily address the suitability of this assumption from observing the data because you don't know whether the null is false*.

However, if the symmetry assumption does hold under the null (and the considerations here can sometimes be addressed by a simple argument of no effect), then the (under the null) the population pseudomedian of pair-differences will correspond to the population median of pair-differences.

A rejection doesn't automatically imply that the same applies under the alternative, however (at least not that I can see).

If you can make an argument for symmetry under the alternative, it would still correspond to the second interpretation, but - while it's often fairly easy to make the argument under the null - it's much more difficult under the alternative.

It may be safer to stick with what the test actually does look at.

* for that matter you shouldn't be choosing a test based on what you find in the data