PhD position in Statistics – Bayesian sample size calculations for trials with multilevel data (# of pos: 2)

Are you interested in working with a dynamic team of researchers? Do you have the ambition to contribute to scientific progress? Then a PhD position at the department of Methodology and Statistics of Utrecht University might be just the right job for you. You will work on the project “Bayesian sample size calculations for trials with multilevel data”, which is funded by the Dutch Research Council (Open Competition scheme).

Funding agencies and (medical) ethical committees often demand a sample size calculation to be included in research proposals. The common approach to sample size calculation is statistical power analysis. This approach relies on null hypothesis significance testing, where a null hypothesis of no effect is tested. This approach has received severe criticism and Bayesian informative hypothesis testing has been developed as an alternative. With this approach one or more informative hypotheses are formulated, based on findings in the literature, expert knowledge or expectations. These are then evaluated on the basis of the Bayes Factor, which quantifies the amount of support for an informative hypothesis, given the data.

Statisticians from the department of Methodology and Statistics are leading experts in the field of Bayesian informative hypothesis testing. They are involved in the development of theory and its implementation in the R package bain and the free software jasp. Over the past years Bayesian informative hypothesis testing has become increasingly used by applied researchers in the social and behavioural sciences. It is therefore necessary to provide such researchers with tools for a priori Bayesian sample size calculation.

The specific focus of this project is on trials with multilevel data. Such data occur when subjects are nested within groups, or when repeated measures over time are nested within subjects. The multilevel (mixed) model is an appropriate model for the analysis of such data and a priori sample size calculations for trials with such data based on statistical power analysis are widely available. A priori sample size calculations for Bayesian informative hypothesis testing in trials with multilevel data are still lacking.

Two PhD candidates will be involved in this project. The first focuses on cluster randomized trials, where existing groups of subjects (e.g. schools, general practices) are randomized to treatment conditions. The second focuses on longitudinal intervention studies, where various treatment conditions are compared to each other over time on some outcome variable (e.g. stress, smoking status).

Both PhD candidates will be supervised by dr. Mirjam Moerbeek, who is an expert in the field of sample size calculation, especially for trials with multilevel data, and prof. dr. Herbert Hoijtink, who is an expert in the field of Bayesian informative hypothesis testing. The specific focus of the sub-projects will be determined in collaboration with an advisory board, consisting of statisticians and applied researchers, that has been set up for the purpose of the project. Both PhD theses will be supplemented with R packages, that allow applied researcher to calculate their sample sizes a priori.

The project builds upon the PhD thesis by dr. Fu .

The responsibilities of the PhD candidate are:

• conducting the research (literature research/review, conducting simulations in R, presenting the results, preparing and storing R packages for sharing);
• writing international scientific publications and a dissertation that combines the theoretical and statistical aspects of the project;
• giving presentations at (inter)national scientific conferences;
• active participation in the research team of the UU department of Methodology and Statistics (M&S);
• knowledge utilization: collaborating and sharing findings with practitioners (e.g., professional publications and presentations for scholars/teachers/practitioners);
• following courses/trainings (e.g., statistical courses).

The work also includes 10% teaching tasks and statistical consultancy. Of course, you will be well guided and supported by your two supervisors.

When applying, please specify if you have a preference for the project on cluster randomized trials, the project on longitudinal intervention or if you prefer both projects equally.