The correlations of the composite phenotype with each individual phenotype clearly show that the association is dominated by the average number of drinks reported. In post-hoc testing to examine the origin of the composite associations, average drinks was tested as a single phenotype for genetic association. Table 5 shows that while the signal was consistent, more information is gained by including all four alcohol-related phenotypes in a composite compared to average drinks alone, as evidenced by the less strong p-values (rs13280604 (p=0.04), rs2304297 (p=0.054), rs892413 (p=0.011) and rs1072003 (p = 0.035)). This highlights the advantage of using the FBAT-PC approach, since testing each trait individually would have failed to yield sufficiently significant p-values to overcome multiple testing.
