A mathematical and statistical analysis of significance testing in the life sciences and in the context of Genome Wide Association Studies.


Florian Stampfer

MSc student (Adv. Philipp Mitteröcker)
Unit for Theoretical Biology, Department of Evolutionary Biology
University of Vienna

Null hypothesis statistical significance testing has become an integral part of the epistemological process in almost any empirical science, despite it being completely inappropriate in most contexts. The cult-like obsession with p-values has led to various problems and misinterpretations, despite warnings from many different scientists from various fields for almost a century. One of the latest misuses lies in its application in the context of genome wide association studies, where very small effect sizes and extremely large sample sizes make hypothesis testing completely meaningless. In this thesis this problem is analysed mathematically and an alternative method using bootstrapping and confidence intervals to give realistic estimates of genetic effect sizes is proposed. Stronger inferences can be made from these results, while needing far smaller sample sizes.