Bayesian Sample Size Calculation for Multilevel trials
Funding agencies of empirical research in the social, behavioral and biomedical sciences want trials to be designed in an efficient way, hence the motivation of sample size is an important requirement in grant proposals. Sample sizes are often calculated based on null hypothesis significance testing. This approach to hypothesis testing has received severe criticism over the past decades and Bayesian evaluation of informative hypotheses has been developed as an alternative. Informative hypotheses are formulated based on researchers’ expectations or findings in the literature and are therefore more realistic hypotheses than the common null hypothesis of “no effect”. For that reason, they are being increasingly used.
The aim of the proposed project is to develop methodology to perform a priori sample size calculations for Bayesian evaluation of hypotheses testing. The project focusses on a research design with multilevel data that are common in social, behavioral and biomedical science: longitudinal intervention studies. Currently available Bayesian a priori sample size guidelines cannot be used for such trial designs, since they ignore the multilevel structure of the data, and therefore result in too low sample sizes. The methodology that will be developed in this project will be implemented in freeware software. The results of this project enable applied researchers to cost-efficiently design their trials.
In the past four years researchers focused on such sample size calculations for studies for t-tests, anova and regression models, in which they assumed the data to be independent. This assumption is very likely to be violated in the case of multilevel structure, as is often the case in social, behavioral and biomedical science. For that reason, it is important to extend our previous research to studies with multilevel longitudinal data.
Dr. ir. M.M. Moerbeek
Prof. Dr. H.J.A.H.Hoijtink
NOW open competition