Mendelian randomization refers to studies that exploit the random assignment of individuals' genotypes (Davey Smith and Ebrahim, 2003). Economists are increasingly interested in the use of genetic variants as instrumental variables (IV) to identify the effect of a modifiable (non-genetic) risk factor (sometimes referred to as the phenotype) on the outcome of interest. For example, Ding et al. (2009) and Fletcher and Lehrer (2009) examine the effects of several health conditions on adolescents' academic achievement, Norton and Han (2008) examine the effects of BMI on labour market outcomes, and von Hinke Kessler Scholder et al. (2010a,b) explore if children's fat mass and height affects their human capital outcomes.
The editorial by Cawley et al. in this volume of Health Economics discusses genes related to neurotransmitters as used in some of the above studies, such as the dopamine transporter (DAT1), the dopamine D2 and D4 receptors (DRD2 and DRD4), the serotonin transporter (5HTT), and monoamine oxidase (MAOA). They argue for caution in using these in the IV context, especially when studying outcomes such as educational attainment, as 'they may be associated with too many things to satisfy the exclusion restriction'.
We fully agree with these arguments and have in a previous publication (Lawlor et al., 2008) also highlighted some other limitations of the studies cited by Cawley et al. Neurotransmitters are implicated in many neurological processes; it is difficult to argue that they can be used as valid instruments for one specific risk factor without being associated with others that could plausibly influence the outcome of interest. Nonetheless, there are now many examples of genetic variants being used to make -what is likely to be, based on the scientific literature (as we discuss further below) -correct inferences about the effects of modifiable risk factors on outcomes. This, however, can only be achieved through a careful selection of variants and sufficiently large samples. In this editorial, we reiterate the importance of the selection of variants, referring to the assumptions made in IV: the relevance assumption and exclusion restriction. We briefly discuss what we consider to be some of the key conditions that have to be met for genetic variants to be used as instruments. We do not aim to provide an exhaustive list; we emphasize key issues that are relevant to all Mendelian randomization studies. Depending on the research question of interest, other potential sources of bias may need to be considered in addition to those discussed below.
THE RELEVANCE ASSUMPTION
Mendelian randomization can only be used when there are established genetic variants that are robustly associated with the phenotype; also known as the relevance assumption in IV. The choice of instruments should rely on prior knowledge about the association between genotype and phenotype, as shown in a