Methodology & Statistics
Faculty of Social Sciences
Prof. Herbert Hoijtink
On June 3rd 2016, Xin Gu defended his thesis entitled
Bayesian Evaluation of Informative Hypothesis
(Thesis and summary not available yet)
Bayesian evaluation of informative hypotheses in general statistical models
Null hypothesis significance testing is by far the dominant research tool for the evaluation of empirical data collected by experiments and observational studies in areas such as the behavioral and social sciences, biology, epidemiology and medicine. This is surprising because null hypothesis significance testing has strongly been criticized (see, for example, Cohen (1994), Royall (1997) and Wagenmakers (2007)). One of the reasons is probably that researchers tend to stick to the methods they have always used. However, another reason may very well be that there are no attractive alternatives.
Bayesian evaluation of informative hypotheses provides an attractive alternative. This approach no longer requires researchers to focus on the null hypothesis. It allows them to focus on the theory or expectation they are interested in and to answer the question: “is my theory/expectation supported by the data or not”. Applied researchers start to discover the existence of informative hypotheses and the first publications in which they are used have appeared. The PhD project proposed will substantially increase the class of statistical models for which informative hypotheses can be evaluated. It will therefore contribute to the construction of a toolkit that will enable researchers to straightforwardly evaluate their theories/expectations.
Furthermore, this project will address statistical issues related to the evaluation of informative hypotheses that are in need of further research: how to evaluate informative hypotheses formulated using equality constraints; and, how to move beyond the multivariate normal linear model. It will therefore also contribute to the further development of statistical theory.
CSC (China Scholarship Council)