Chapter 5:

In this chapter, we introduce the inverse-variance weighted method, which combines summarized data for multiple variants into a single causal estimate. We show that estimates from the inverse-variance weighted method are the same as those obtained from the two-stage method, and hence most efficiently combine information from multiple valid instrumental variables.

The inverse-variance weighted method combines summarized data on multiple genetic variants to provide an estimate equal to that which would have been obtained from a two-stage method if individual-level data were available. The ability to perform analyses using summarized data has allowed the widespread proliferation of Mendelian randomization based on publicly-available data. However, the inverse-variance weighted method assumes that all genetic variants are valid instrumental variables, an assumption that may not hold in practice.