An evaluation of some statistical methods for analysing numbers of abnormalities found amongst litters in teratology studies.

Skeletal abnormalities in rabbit, rat and mouse foetuses are found to follow closely beta-binomial distributions rather than simple binomial distributions, i.e., foetuses from the same litter are found to have closer chances of being abnormal than foetuses from different litters. Control and treated groups of 12 litters were simulated using beta-binomial distributions and were used to compare the performances of three types of analysis on two-group teratological studies: Student's t test on the litter proportions, transformed or weighted where necessary; the Wilcoxon distribution-free on ranked litter proportions; and likelihood ratio tests assuming a beta-binomial distribution of abnormalities. It was found that the likelihood ratio tests do not have Type I errors equal to the nominal level and are not more powerful than the Student's t test or the Wilcoxon test. The latter two tests produce similar results but the Student's t test is marginally more powerful.