STATISTICAL TESTS OF SIGNIFICANCE IN TRANSGENIC MUTATION ASSAYS - CONSIDERATIONS ON THE EXPERIMENTAL UNIT

When significant animal-to-animal variability is present in binary res ponse date, the usual statistical tests applied to such data do not al ways operate correctly. In transgenic mouse mutation data, some eviden ce of significant animal-to-animal variability already exists, suggest ing that conventional statistical methods may not be appropriate. Here , we describe an alternative statistical method that treats the animal as the experimental (or statistically independent) unit, and contrast results of its application with those from methods that take the tran sgene as the experimental unit. Using data from two publications that report experimental results for individual animals, the transgene-base d and animal-based analyses can yield very different interpretations o f the experimental data. The performance of animal-based statistical m ethods should be improved by conducting future experiments with enough animals to adequately address animal-to-animal variability. (C) 1994 Wiley-Liss, Inc.