Biometrika 1962 49(1-2):283; doi:10.1093/biomet/49.1-2.283-b
© 1962 by Biometrika Trust
Corrigenda |
Corrigenda
I regret having published a test for the hypothesis (4.1, ijy) based on the inequality (4.5, ijy). When we carry out the minimization of Gijy(t ) we get
as a factor. When Tijy is different from 0 we may divide by |Tijy| on both sides of the inequality. We then find that (4.5, ijy) is equivalent to an inequality stating that a positive definite quadratic form in (aiaj) and (bibj) is less than or equal to a constant, so that a test based on this inequality is a test of the hypothesis
=
i, ßj = rather than the hypothesis (4.1, ijy).
The preceding result does not represent any argument against using a test for the hypothesis (4.1, ijy) based on (4.4). Instead of minimizing Gijy(t) with respect to t we may choose a certain value oft, or take an average of the right-hand side of (4.4) for different values of t in such a way that we make the level of significance of the test as far as possible independent of
i,
j, ßi, ßi for values of these parameters satisfying (4.1, ijy).