The validity of genes related to neurotransmitters as instrumental variables

Health economics has a tradition of evaluating the models of instrumental variables (IV) used in our field. For example, French and Popovici (2011) assess the instruments used in 60 recent studies measuring the effect of substance use (alcohol, illicit drugs, and tobacco) on various outcomes. Rashad and Kaestner (2004) evaluate the instruments used by Sen (2002) and Rees and colleagues (2001) to identify the effect of alcohol on risky sex among adolescents. Terza et al. (2008) review the health economics literature that applies linear IV methods in inherently nonlinear regression settings. In this editorial we discuss the IVs used in Fletcher and Lehrer (2009a), hereafter FL, to estimate the effect of health conditions on educational attainment. We focus on FL because it is novel, well-written, and careful in many ways; it won the 2009 Victor R. Fuchs award for ‘the best research paper with the potential to spawn new research in an underdeveloped area of health economics or health policy’. The use of genes as IVs (and in other empirical capacities) is likely to become more widespread as large social science datasets continue to add data from genetic assays. Therefore, consideration of this method is timely and likely to be useful in the future. Our comments should not be interpreted as exclusively about FL, but also apply to other papers in this emerging area of empirical health economics (e.g. Ding et al., 2009; Fletcher and Lehrer, 2009b), including our own (Norton and Han, 2008). FL estimate IV models using data from the third wave of the National Longitudinal Survey of Adolescent Health (Add Health) to test whether overweight (based on self-reported weight and height), depression (based on the 20-item Center for Epidemiologic Studies-Depression Scale), and attention deficit and hyperactivity disorder or ADHD (based on 18 retrospective questions) affect years of schooling completed when respondents were on average 22 years old. FL argue persuasively that these measures of physical and mental health are endogenous to educational attainment. For example, poor grades in school may contribute to depression (reverse causality) and unobserved factors, such as rate of time preference, may contribute to both poor health and worse educational outcomes. As a result, simple regression would result in biased estimates of the effect of health on educational attainment. FL address the endogeneity of health by using genes as IVs. The specific IVs they use, listed in column 1 of Table I, are genes related to the neurotransmitters dopamine and serotonin. Fletcher and Lehrer (2009b) note that geneticists have linked these specific genes to overweight, depression, and ADHD, and that these correlations make sense in light of the fact that neurotransmitters affect mood, attention, and are associated with feelings of pleasure and reward, and thus serve as positive reinforcement for certain decisions. For example, people who are genetically predisposed to low levels