Key points are not available for this paper at this time.
SUMMARY A combined P-value test is described that is specifically designed to test the consensus of a set of independent tests, all of which address the same null hypothesis. It is shown that Fisher's combined probability test is unsuitable in this application. A test based on the arithmetic mean of the normaltransformed P-value is shown to properly balance component tests that support or refute a common null hypothesis. A means of determining the family-wide significance of individual tests, which are part of a collection of conceptually related tests, is also described.
William R. Rice (Fri,) studied this question.