Recommendations for appropriate statistical practice in toxicologic experiments.

Abstract

Appropriate statistical practice in toxicologic research is reviewed. The problem centers on the antagonistic needs to discover toxic effects and avoid false indictments of harmless compounds. Specific problems which distort error rates include having many dependent variables, conducting many tests on a variable, data snooping, lack of statistical power and 5) violations of statistical assumptions. Solution strategies include top down planning, designing efficient experiments, balancing type I and type II error rates, selecting hypotheses and 5) using appropriate statistical analysis methods. Top down planning and the "leapfrog" design strategy are particularly emphasized.