Statistics Glossary → Hodges-Lehmann estimator (pseudomedian)
Hodges-Lehmann estimator (pseudomedian)concept
A robust, rank-based estimate of location or location shift that matches the Wilcoxon and Mann-Whitney tests.
For one sample (or for paired differences), the Hodges-Lehmann estimate is the median of all pairwise averages of the observations (the Walsh averages, (xᵢ + xⱼ)/2, i ≤ j), also called the pseudomedian. For two independent samples, it is the median of all differences between each value in one group and each value in the other, and it estimates the location shift between the two distributions. Its confidence interval follows from the same rank logic.
It resists extreme values yet is nearly as efficient as the mean for normal data. The two-sample estimate is not the same as the difference between the two group medians; it estimates the shift when the distributions have the same shape and are merely displaced. In one sample, the pseudomedian equals the median when the distribution is symmetric, but the two can differ for skewed distributions.
An everyday example
A new sleeping pill is tried on 9 people, and the gains in sleep time on pill nights are −5, 10, 12, 15, 18, 20, 22, 25 and 180 minutes. Because of the 180-minute extreme value, the mean gain is 33 minutes. The Hodges-Lehmann estimate, the median of the Walsh averages, is 17.5 minutes and better reflects the gain for a typical person.
In MerQur
In the Mann-Whitney U test as the location shift, with an exact confidence interval from the 'Extra outputs: Exact HL CI' checkbox; in the Wilcoxon Signed-Rank test, the 'HL pseudomedian' checkbox gives the median of the Walsh averages of the non-zero differences.
Analyses where this concept appears
Mann-Whitney UWilcoxon Signed-Rank
Related concepts
Rank-biserial correlationDistribution-free (order-statistic) confidence interval