Statistical tools to improve the quality of experiments and data analysis for assessing non-target effects.

Abstract

AbstractWhen testing non-target effects of biological control agents, it is essential that conclusions can be drawn with high precision and confidence. However, testing non-target effects confronts the experimenter with a number of difficulties. First of all, biologically positive cases of not finding any non-target effect are more difficult to substantiate, since in standard statistical hypothesis testing, we can only associate a precise probability to err with rejecting the null hypothesis that assumes no effect, but not with accepting it. The main problem here is the effect size, i.e. the difference from the null hypothesis that is considered biologically meaningful. Secondly, there will usually be a trade-off between the costs associated with increased sample sizes and the confidence of the results of non-target effects testing. Often, sample size will be a limiting factor due to a shortage of animals, space for testing arenas, research funding, etc. Thus, it becomes especially important to optimize the experimental design and to use the most powerful statistical tools to obtain maximum confidence in the test results. Here, we will briefly (i) introduce the reader to common pitfalls of experimental design, (ii) explain the nature of errors in statistical testing, (iii) point towards methods that determine the power of statistical tests, (iv) explain the distribution of the most commonly encountered types of data, and (v) provide an introduction to powerful statistical tests for such data.