Chapter 8:

In this chapter, we discuss various statistical issues affecting estimates from Mendelian randomization. While it is not possible to completely separate theoretical and pragmatic considerations, the focus of this chapter is more on understanding why and how these methodological issues affect findings. In turn, we discuss weak instrument bias, allele scores, sample overlap, winner's curse, selection and collider bias, covariate adjustment, non-collapsibility, time and time-varying effects, power, choosing variants from a single gene region, binary exposures, and alternative estimation methods. We particularly focus on factors affecting the bias, coverage, and efficiency of estimates.

Bias in a statistical analysis is not a binary phenomenon, and it would be a fallacy to think of biases as being either present or absent. Hence, pragmatic choices are necessary and judgement is required to design and perform the optimal analysis (or set of analyses) given the data available.

.