Rank-Order Tests for the Parallelism of Several Regression Surfaces

摘要

For testing the hypothesis that several (s > or = 2) linear regression surfaces X sub ki = alpha sub k + beta sub k c sub ki + Z sub ki (k = 1,...,s) are parallel to one another, i.e., beta sub 1 = ... = Beta sub s, a class of rank-order tests are considered. The tests are shown to be asymptotically distribution-free, and their asymptotic efficiency relative to the general likelihood ratio test is derived. Asymptotic optimality in the sense of Wald is also discussed. (Author)