these studies in which the point estimate for a follow-up study in a separate population fell outside the 95% confidence interval for the odds ratio of the original study. This includes 13 variants and odds ratios with confidence intervals for the population in which association was first discovered (12 European and one Japanese) and 20 odds ratios for subsequent tests in separate populations, consisting of eight tests in African Americans (seven not significant and one significant in the opposite direction), four tests in Japanese (one not significant and three significant in the same direction), four tests in Koreans (one not significant and three significant in the same direction), two tests in the Indian subcontinent (two not significant), one test in Europeans (not significant), and one in Chinese (not significant). Although it is possible that many of these differences are related to differences in LD (association) between markers and causal sites, genetic or environmental interactions, or simply genetic heterogeneity, it appears likely that many of these differences are due to multiple underlying rare variants that create different synthetic effects in the populations. There are also likely to be other diagnostics of synthetic associations observable in GWAS data. For example, one would
