Collings B J, Margolin B H, Oehlert G W
Biometrics. 1981 Dec;37(4):775-94.
The fluctuation test proposed by Green, Muriel and Bridges (1976, Mutation Research 38, 33-42), a short-term microbial test for mutagenicity, yields binomial observations for which the probability of success varies with the background mutation rate, the total number of microbial growth cycles for which a microbe is at risk of mutation, and the rate of induced mutation for the compound being tested. A standard one-tailed two-sample binomial test is preferable to the two-tailed test adopted by Green et al. for analyzing data from a control versus single positive dose fluctuation test. Based on exact power computations, recommendations are offered for the design of such a fluctuation test. A simple method of guarding against the impact of a small number of aberrant observations in a multisample binomial problem is studied; it is applicable to the fluctuation test when the protocol involves replicate measurements . Finally, the case of more than one positive dose of the test compound is investigated. Two statistical tests for this situation, both extensions of one-tailed two-sample test, are extensively compared. A departure from monotonicity at high doses has a more serious effect on the power of the 'regression' test than on that of the 'isotonic' test. A variant of the isotonic test, based on the angular transformation, should be avoided.
