Significance tests for differences between ranked variates drawn from normal populations

David B. Duncan · 1947

Consider n variates drawn from n normal populations with the same variance for which an independent estimate based on p degree of freedom is available. A multiple choice significance test is required for testing the n(n - 1)/2 differences between the variates taken in all possible pairs. The main requirement of the test is that it has, as closely as possible, the same properties with respect to each difference as an level t test has to the single difference in the simple case, n = 2. This type of problem is commonly met for example, in varietal yield trials and similar agricultural experiments;A test is proposed which places upper bounds on errors comparable to type 1 errors in the case n = 2 and at the same time compares favorably with other alternatives in minimizing a weighted average risk function relative to the remaining errors. The test is given in detail for the cases n = 3, alpha = .05, p = infinity, n = 3, alpha = .05, p = 10; and n = 4, alpha = .05, p = infinity.

Cited by 12 works. doi:10.31274/rtd-180813-14125

Authors