The main limitation of the present study was the relatively small sample size, particularly for the measurement of anxiety. Using the more conservative alpha level, power calculations for our varying sample sizes ranged 45%−99% to detect an effect size of 1%. This resulted in insufficient power to attain genome-wide significance for some associations that were suggestive. However, it has been shown that most borderline GWAS significant results (i.e.,P>5×10−8 and P≤10−7) are potentially genuine associations (Panagiotou and Ioannidis 2011). Thus, we can place confidence in several of our SNP and gene-based test findings because of the replication support we found. The use of different psychological scales across cohorts is considered advantageous because the meta-analysis results will invariably detect associations that relate to reliable trait variance (i.e., variance that is common across tests that purportedly measure the same underlying trait).
