When a large number of factors are considered in an experiment, the identification of active factors that may have a substantial impact on the outcome is needed for screening purposes. Computer simulation experiments include many factors. For screening, different designs, such as orthogonal supersaturated, definitive screening, A-optimal, or D-optimal designs are proposed in the literature. A- or D-optimal designs are generated by optimization codes to minimize the average variance or covariance of the parameter estimates of a given model, respectively. Supersaturated design can also be used along with A- or D-optimality.
In order to overcome the difficulties with classical designs related to the dependence on an assumed statistical model, the number of blocks, sample size, and multi-stratum structures, Bayesian approaches are proposed along with the supersaturated and D-optimal designs in the literature. This project aims to explore the current literature on Bayesian supersaturated D-optimal designs and develop new Bayesian A- or D-optimal designs that are more cost-effective and can be used with multi-stratum structures such as split-plot designs, control type I error around the desired level and improve the power.
This project requires intense coding to implement developed methods. The successful candidate needs to have strong programming skills. In this sense, strong knowledge of R and MATLAB is required. Also, a strong knowledge of Bayesian statistics and the design of experiments are required. In this sense, the selection committee will be looking for at least one Bayesian analysis/modelling and one design of experiments course in the previous studies of the successful candidate or one journal paper in each field is required. Candidates who are not satisfying these requirements are not encouraged to apply.
