Combining subjects from the population cohort who were obese (BMI ≥30 kg.m−2) or normal weight/underweight (BMI <25 kg.m−2) with the corresponding case-control subjects, and assuming a GSV has a dominant effect, the combined cohort was sufficient to give >98% power to detect associations (at P<0.05 for Fisher’s exact test) of GSVs present almost exclusively in obese subjects (odds ratio, OR = 50) at a frequency in cases of 0.005, or with power of 94% or 83% for odds ratios of 10 or 5 respectively; even for a GSV frequency in cases of 0.002, power was 67% (OR = 50), 53% (OR = 10) or 40% (OR = 5). On the basis of the observed GSV frequencies and ORs in the original reports, median power was 79.8% (minimum 53%) for the 11 loci for which the corresponding GSVs were detected in our cohorts; thus, we might expect to replicate ∼80% of genuine associations. Although this is likely to be something of an overestimate because of OR overestimation due to the “winner’s curse” [20], the minimum OR in the original reports was
