Statistics Glossary → Jonckheere-Terpstra test (ordered alternative)
Jonckheere-Terpstra test (ordered alternative)concept
A rank-based test of whether the values of three or more independent groups with a natural order increase or decrease monotonically along that order.
The Kruskal-Wallis test asks whether the groups differ 'in any way'. When the groups are ordered, such as low, medium and high dose, and a trend in that order is expected, the Jonckheere-Terpstra test is more powerful. The alternative hypothesis is θ₁ ≤ θ₂ ≤ … ≤ θ_k with at least one strict inequality. The statistic J is the sum, over every ordered pair of groups (i < j), of the number of pairs in which a value from group j exceeds a value from group i (the Mann-Whitney U counts). In large samples the p value comes from a tie-corrected normal approximation.
The order of the groups must be fixed on theoretical grounds before looking at the data; if it is chosen afterwards, the test loses its meaning. A significant result indicates an overall ordered shift, not that the trend holds at every single step. The test is sensitive to differences in location; if the groups differ only in spread, interpretation becomes difficult. For repeated measures with an ordered alternative, Page's test is used.
An everyday example
A farmer gives tomato plants 0, 5, 10 and 15 g of fertiliser and measures the yield of 8 plants at each dose. He expects 'more fertiliser, more yield'. Instead of looking for any difference between groups, the Jonckheere-Terpstra test checks this ordered increase directly and can reach significance more easily than Kruskal-Wallis with the same data.
In MerQur
In the Kruskal-Wallis test, the 'Extra outputs: Jonckheere-Terpstra' checkbox gives the J = Σ U_ij test along the natural order of the groups (ascending if numeric, otherwise alphabetical), with a tie-corrected normal approximation and a two-sided p.