9. Conclusion

Guidelines and the selection of the different methods presented are, at least partly, based on experience and we cannot cover all statistical methods available, for example we have not discussed resampling methods like jackknife in detail (for further reading see Good, 2006). More details on designing specific experiments and performing statistical analyses on the ensuing data can be found in respective chapters of the COLOSS BEEBOOK (e.g. in the toxicology chapter, Medrzycki et al., 2013).

Experimenters need to use statistical tests to take (or to help take) a decision. A statistical analysis can be conducted only if its assumptions are met, which largely depends on how the experiment was designed, defined during the drafting of the study protocol. Without some effort at the a priori conception stage and input from those knowledgeable in statistics and/or experimental design, the resulting analyses are frequently poor and the conclusions can be biased or flat-out wrong. Why spend a year or more collecting data and then realise that, due to poor design, it is not suitable for its original purpose, to test the hypotheses of interest. The most important point to understand about statistics is that one should think about the statistical analysis before collecting data or conducting the experiment.