Workweek Optimization of Experimental Designs: Exact Designs for Variable Sampling Costs

摘要

An optimal experimental design combines high-quality parameter estimation with efficient use of resources. This paper proposes a new method for heuristic optimization of experimental designs in the presence of variable sampling costs. The method finds the inexpensive designs with desirable statistical qualities and provides substantial insight regarding the relative importance (in monetary terms) of sampling at specific design points. The method is illustrated within the context of a start-stop exposure study in aquatic toxicology. Fast heuristics enable the analyses of a large range of sensitivity issues and examination of trade-offs between information and cost. The method is described for designs in which replicate sampling is prohibited; a generalization allowing for replicate sampling is provided as an appendix in the online supplemental materials.