of the homogeneity of the ratio estimates from different candidate IVs (Windmeijer et al., 2015). The basic idea of this approach is to report a causal estimate based on genetic variants whose ratio estimates are mutually similar. A related approach based on step‐wise selection using a heterogeneity test statistic was proposed by Johnson [2013]. Although the exclusion of certain variants to consider their potential impact on the overall estimate may be a reasonable sensitivity analysis, a potential danger of such an approach is that of post hoc or data‐driven analysis, in which genetic variants are cherry‐picked for inclusion in the analysis model, and dissenting variants are filtered out.
