Nonparametric Tests

Jonckheere-Terpstra Test Table: Exact Critical Values of J

Exact critical values for the Jonckheere-Terpstra test for an ordered trend across groups, such as increasing dose.

  • k = 3 to 6 groups
  • Exact values
  • Unequal sizes in calculator

Jonckheere-Terpstra Table

Jump to a value of k below. Rows: subjects per group · Columns: one-tailed α · Cells: critical J. Reject H₀ if J ≥ the cell.

k = 3 groups

k = 3 groups
One-tailed α →
n per group ↓
0.10.050.0250.010.0050.001
2101112———
3202223252527
4343638404245
5515457606266
6717579838692
795100105110114121
8121128134140145154
9152160166174180190
10185194202212218230

k = 4 groups

k = 4 groups
One-tailed α →
n per group ↓
0.10.050.0250.010.0050.001
2181921222324
3374042444548
4636770737681
595100105110114120
6134141147154158167
7179188196204210222
8231242251262269283
9290302313326334351
10354369382397407427

k = 5 groups

k = 5 groups
One-tailed α →
n per group ↓
0.10.050.0250.010.0050.001
2283032333537
3596265697175
4100106110116119126
5153160167174179189
6216226235244251264
7290303313325334350
8375390403418428449
9470488504522534558
10576597615636650679

k = 6 groups

k = 6 groups
One-tailed α →
n per group ↓
0.10.050.0250.010.0050.001
2404345474952
385909498101107
4147154160167171181
5224234242252259272
6317330342354363381
7427443457474484507
8552572589609623650
9693717738761777810
10850878902930949987

Tip: click any value to highlight its row and column.

How to read this table

  1. Order the groups by the hypothesised trend (for example placebo, low, middle, high dose).
  2. For every pair of groups i < j, count the pairs of observations where the value in the later group is larger (ties count ½). J is the sum of these counts.
  3. Reject H₀ in favour of an increasing trend if J is greater than or equal to the table value. For unequal group sizes, use the calculator.

Worked example

A study gives placebo, low dose and high dose to 5 patients each, and the counts of increasing pairs sum to J = 56. In the k = 3 table, row 5, α = 0.05 gives 54, so there is a significant increasing trend at the 5% level.

Click the underlined links to highlight the value in the table.

Calculator

For values between the rows and columns of the table. Results update as you type.

Formula and how the values were computed

J = Σi<j Uij  ·  E(J) = (N² − Σnᵢ²)/4  ·  Var(J) = (N²(2N + 3) − Σnᵢ²(2nᵢ + 3))/72

Under H₀ the J statistic is distributed as a sum of independent Mann-Whitney statistics, and its exact distribution is the normalised q-multinomial coefficient, computed exactly. Each entry is the smallest J with P(J ≥ j) ≤ α.

Frequently asked questions

Why not just use Kruskal-Wallis?

Kruskal-Wallis tests for any difference and ignores the order of the groups. When the alternative is a monotone trend, Jonckheere-Terpstra is considerably more powerful.

Is this a test of dose-response?

Yes, it is a common nonparametric test of a monotone dose-response. For binary outcomes the Cochran-Armitage trend test plays the same role.

Related calculators and tutorials