A new approach to overcome shortcomings with multiple testing of reproduction data in ecotoxicology

Abstract

In ecotoxicology species reproduction tests and multiple testing of reproduction data are wide spread. While normal approximation of the data is a minor problem often the requirement of variance homogeneity is not fulfilled. Variance homogeneity is necessary to assure the proper application of statistical procedures like pairwise t tests, Dunnett t test, and Williams t test. A Poisson model can solve this issue preserving meaningful results and rendering statistical analysis more reliable. Moreover, sequential application of pairwise statistical “control vs. treatment” tests is a drawback concerning \(\alpha \)-inflation. The closure principle (CP) for hypothesis testing is used to generate a step-wise approach for detection of the No/Lowest Observed Effect Concentration using the computational approach test (CAT). The advantages and disadvantages of the combined CPCAT approach compared to the widely used t tests are pointed out and results of real data and fictitious data analysis are compared revealing the superiority of the Poisson model and CPCAT.