Three-stage designs for seed testing experiments

SUMMARY The proportion of infected seeds in a population can be estimated indirectly by testing samples of seed. This paper discusses the choice of sample sizes in small experiments that use only 12 samples in total. Single-stage and three-stage designs are developed, based on a prior distribution for the proportion of infected seeds. The three-stage designs have an efficiency of about 70% whereas for single-stage designs the efficiency is only about 40%.