We have chosen in this paper to focus on the two‐sample situation using summary data as this is most relevant to Mendelian randomization; all the other methods have been developed for use with individual‐level data in a one‐sample setting, and hence are not considered in this manuscript. We look forward to further methodological developments in this area. One particular area that may be fruitful for such developments is bias‐correction methods in the meta‐analysis literature; two particular examples are the trim‐and‐fill method (Duval and Tweedie, 2000) and the use of pseudodata (Bowden et al., 2006). Equivalently, there may be application of the median‐based methods proposed in this paper in the field of meta‐analysis, to provide a more robust pooled estimate.
