Many research groups cannot afford to genotype the large sample sizes needed to reliably detect genetic markers that are weakly associated with a trait using a genome-wide platform. This has sparked interest in multi-stage designs, where a subset of available samples are genotyped using the genome-wide platform, and then a subset of the “most promising” markers (typically those with lowest p-values) are genotyped using a custom platform. These designs are reviewed in more detail elsewhere in this issue.[16] We should note that the primary motivation of multi-stage designs is not to increase power by testing fewer hypotheses in the second stage samples, and hence paying a smaller penalty for multiple testing at the second stage. Rather the primary goal of multi-stage designs is to save genotyping costs or maximize power given a fixed genotyping budget. If genotyping costs were not an issue, then the multistage approach is less powerful than simply testing all markers in the entire available sample. [16–18] As genotyping costs decrease, and as more samples have been genotyped as part of previous GWA analyses, single-stage analyses become more common. [17]
