Statistical Methods
Distinguish a Walsh Pseudomedian from a Median Difference
Define the task
A one-sample Hodges–Lehmann location estimate is the median of Walsh averages, formed from all pairs of observed differences including a value paired with itself. For paired data, create one difference per independent unit before forming averages. Do not confuse this estimate with the median of the original differences. The associated population pseudomedian equals the median under symmetry, but the two can differ for asymmetric distributions. State which location quantity answers the research question.
Carry out the check
Form (d[i]+d[j])/2 for every i≤j and take their median. There are n(n+1)/2 Walsh averages, but this does not create that many independent observations. The local tool reports the estimate, ordinary median and number of averages only. It supplies neither a signed-rank p-value nor a confidence interval. Preserve ties, zeros and the difference direction. An interval borrowed from an unrelated median procedure is not automatically appropriate. Check symmetry assumptions before interpreting a signed-rank analysis as inference about a median change.
Worked hypothetical example
For hypothetical differences (0,1,9), Walsh averages in sorted order are 0,0.5,1,4.5,5,9. Their median is (1+4.5)/2=2.75, while the ordinary median difference is 1. This deliberate asymmetric example exposes the estimand distinction. Reversing the subtraction direction negates both summaries. Report the original sample size of three rather than the six dependent averages, and keep the result descriptive unless an appropriate inferential procedure has also been specified.
Method checked 18 November 2026 against Primary method documentation: R: Wilcoxon Rank Sum and Signed Rank Tests.