| Abstract
| - We are grateful to Bengt Klefsjo, University of Lulea, Sweden, for discovering and correcting an error in our formulae for the asymptotic variances of n* V* and n*K* when Ho is not true. The asymptotic variances under Ho are correct as stated, the incorrect formulae were not utilized in the paper, all other results in the paper are unaffected, and the tests can be implemented as originally described. Setting Qx = V* and Qt = K*, write where, for t = 1,2, n12[Qi−{μ(Ji,F)/μ(F)}]=Li/X¯ Li=n12[n−1∑s−1n{Ji(8/n)−a(Ji,F)}Xs]and a(JitF) = n(J{,F)/(j.(F). Using this representation, Klefsjo points out that Lt is asymptotically normal with asymptotic mean and variance given by (2-14) and (2-15) but with J(u) replaced byJ*{u) = Jt(u) −a(Jt, F).Since /i(J*, F) = 0, our asymptotic mean is correct but our asymptotic variance should be changed to(ó*). When F(x) = 1−exp (x), we have a(J1) = o(J2, F) = 0 and thus Ji* = Ji, i = 1, 2. Hence, our null asymptotic variances for n*7* and n* K* are correct as stated.
|
