Nonparametric tests of tumor prevalence data

This note discusses the statistical analysis of tumor prevalence data arising from tumorgenicity experiments with focus on the comparison of different treatments. In this situation, the commonly used tests can be classified into two types: interval-based tests and model-based tests (Hoel, D. G. and Walburg, H. E., 1972, Journal of the National Cancer Institute 49, 361-372; Dinse, G. E. and Lagakos, S. W. 1983, Applied Statistics 32, 236-248). It is known that the results obtained from the interval-based tests may vary according to the choice of intervals and, for the model-based tests, it may be difficult to justify the assumed model. A computationally simple alternative to these tests is proposed; this alternative is not interval-based and makes no strong model assumption. The results of a simulation study comparing the proposed test with other tests are presented and suggest that the proposed approach is quite satisfactory.