Abstract and poster for SETAC Europe, May 2005, Lille
Statistical issues in fish life-cycle tests with many endpoints.
N. van der Hoeven1 and D.R. Dietrich2
In fish life-cycle tests many variables are measured. Testing each variable separately will lead to a high probability of randomly finding "significant" effects. Therefore, the significance level should be split over all comparisons, lowering the power of the test. To increase power, effects on combined variables can be tested. Interdependent cell type frequencies, e.g. gonadal cell types, always add to 100%. Therefore, absolute frequencies cannot be tested independently and should be translated to relative frequencies within continual smaller cell type groups. Histopathological lesion scores are ordinal and should be tested with non-parametric statistics. Prior estimates of the expected variance, allows calculation of the observable effect with the experimental design chosen and thus the adequacy of the design. Last but not least, finding no significant effect may never be interpreted as prove of the absence of an effect. Before designing the experiment and doing the statistics, the biological expert should indicate which effects are relevant and should be observable.
Abstract and poster for SETAC Europe, May 2005, Lille