in the discovery phase and the assignment of nearly all subjects to genetically matched groups based on analysis of many markers distributed across the genome lessened the potential for stratification. Furthermore, all of the approximately 4000 subjects included in this study were evaluated with a standardized instrument using a rigorous quality-control procedure that reliably diagnoses substance dependence and other psychiatric disorders.21,26 In any event, it is possible that some subjects were misclassified as controls because they were not sufficiently exposed to cocaine to become dependent on the drug. This, however, would bias the results toward the null hypothesis. Because our replication samples had sufficient power to detect allele frequency differences of 7% to 10% for CIP and 8% to 15% for CD in either population, lack of significant findings with individual SNPs in the EA sample could be attributed to an inadequate sample size. However, significant haplotypic associations in this population suggest that genetic background rather than sample size was the limiting factor. Third, only one of the results from analyses of individual SNPs in the replication samples would be considered significant after adjustment for multiple comparisons using a conservative Bonferroni correction (threshold, P=.004 in EAs and P=.006 in AAs
