Several articles report the power analysis of GWAS for given sample sizes and effect sizes. For example, Wang and colleagues showed that a sample size of 1,000 cases and 1,000 controls allows for estimation of an allelic odds ratio of 1.5 with 80% power when the disease allele is common, with a frequency between 0.4 and 0.5 [86]. The sample size necessary to detect the same effect when the disease allele is less common, for example, a frequency of about 10%, is 2,000 cases and 2,000 controls and increases almost exponentially with smaller disease allele frequencies [8]. These calculations make two assumptions: (1) they assume stringent conditions on the P-value to control the overall probability of type I error, and (2) they assume that the analysis is conducted using standard logistic regression. The two assumptions are not independent, because both the P-value and power are relative to the statistical method used for the analysis and not only to the sample size. Imposing stringent conditions on the significance of individual tests has become the popular approach to control the global significance
