Finally, lack of demonstrable heterogeneity may be perceived as a criterion of credible replication. [54] However, one should note that tests and measures of heterogeneity address whether effect sizes across different datasets vary, not whether they are consistently on the same side of the null. Dataset-specific effects could vary a lot, but they may all still point to the same direction of effect. Given the potential diversity of LD structure across populations, and differences in phenotype definitions and measurements across studies, between-study heterogeneity should not dismiss an association because the effect sizes are not consistent, if the evidence for rejection of the null hypothesis is strong.
