What is the role for molecular genetic data in public policy?

Individual specific, but no longer unobserved, heterogeneity

In countless studies labor economists attempt to isolate the influence of a single explanatory variable (e.g. education) on an outcome of interest (e.g. earnings). To convince readers of their principal finding, researchers often assess whether their estimates are sensitive to the inclusion of additional observable characteristics, such as experience, age, height, and health. Until recently, genetic data at the molecular level were not available. This may represent a serious limitation to prior research, since studies on heredity suggest that genetic factors could explain up to 65–80% of the variation in height and 20–40% of the variation in educational attainment.

Without data on genetic characteristics, researchers could only use longitudinal data to create a proxy to capture all unobserved (time-invariant) individual specific heterogeneity. This approach, generally referred to as a fixed-effects analysis, would allow to control for both unmeasured individual specific factors, such as one’s genetic code, and unmeasured skills, such as innate ability, perseverance, industriousness, or motivation. This approach assumes that the unobserved factors have a constant effect for individuals as they age, which is a strong (and not always realistic) assumption. As shown in the Illustration, a variant of the well-studied FTO gene may have different impacts on the body mass index (BMI) as individuals age, implying that the effect actually changes over time.

Genetic data may be what is truly meant by unobserved individual specific heterogeneity. Genetic markers and the structure of one’s DNA are fixed at conception and do not change over the lifecycle, unlike one’s skills, which an increasing body of evidence shows can be developed via education and training investments. In other words, how genes are expressed may change, but the structure of DNA itself cannot.

Data on genetic markers may permit researchers to enter the black box of individual specific unobserved heterogeneity. If genetic markers are linked to traits that are both accounted for and explained by regression models, then failing to control for genetic differences may generate biased estimates and distort research findings. Yet one of the challenges is how to incorporate genetic diversity across individuals in an efficient and meaningful way. After all, there are millions of locations in one’s DNA where genetic differences emerge. An early literature explored links between economic outcomes and specific locations on the chromosome pair (referred to as candidate genes). However, since the findings on the strength of these associations were not robust across samples and may have been due to data snooping, the research community has proposed investigating links between outcomes and a sufficient statistic, called a polygenic risk score, which is often calculated as the cumulative weighted sum of the variation in multiple genetic locations.

However, to determine these weights and the genetic locations to be included in the calculation of a polygenic risk score, evidence from a genome-wide association study (GWAS) is needed. A GWAS can be thought of as a hypothesis-free scan for associations between a specific outcome and subsets of the millions of genetic variants. Looking at the predictive powers of polygenic scores derived from different GWASs over the last five years, a recent study summarizes how the score’s ability to explain variations in educational attainment has increased from 3% to 7% (in this study, educational attainment is taken as years of completed schooling) [2]. This research points out that very large data sets are needed to undertake GWASs since most of the statistically significant effects are very small. Further, larger sample sizes also permit the possibility of identifying rare genetic effects that may not be measured by the underlying genotyping arrays currently used in GWASs.

In sum, while the amount of variation explained by these genes is significantly smaller than estimates from heritability studies that exploited variation in families, recent research shows that the significant markers identified are also associated with neurocognitive disorders and brain function [3]. Hence, this type of research not only identifies the individual genes, but also suggests the biological pathways by pointing out the commonalities in correlations of specific genes with different outcomes (i.e. brain function and educational attainment).

Genetic factors can identify new challenges and diagnoses

When economists use genetic data in the empirical microeconomics literature, they must often overcome empirical challenges, for example when estimating the effects of poor health on outcomes such as years of schooling or wages. The challenge arises since poor health is often measured with error, while also being systematically related to unobserved determinants of the outcome variable. In these situations, instrumental variables regression can be used. Here an additional variable (an “instrument”) is used to replace the problematic independent variable with a proxy that is unrelated to both the unobserved factors that affect the outcome and the measurement error. A study from 2009 provides the first application of genetic markers in an instrumental variables application to identify causal estimates of how physical and mental health conditions in adolescence affect academic performance [4]. Thereby, the health variables are isolated from other explanatory factors such as nurture inputs, which in this case might include the neighborhood in which the families reside and the peers with which the adolescents associate. In using this approach, the researchers assume that the genetic instruments are not only correlated with the poor health measures, but that they only influence academic outcomes through these health variables. The main findings from this study are that depression and obesity both lead to an approximate one standard deviation reduction in academic performance. But this deterioration is shown to differ by student gender: young women are found to be more adversely affected by negative physical and mental health conditions. Lastly, the separate impacts of inattention and hyperactivity on academic performance differ sharply in magnitude and sign.

While this investigation and subsequent work have led to significant debate about which genetic markers are valid instruments, such work has also highlighted that measuring an individual’s health status is challenging due to comorbid health conditions, that is, situations where two or more disorders or illnesses occur in the same person (whether simultaneously or sequentially). In its empirical application, the 2009 study shows that omitting comorbid diagnoses would result in biased estimates of the causal effect of specific health diagnoses on socio-economic outcomes [4].

Many current definitions of health are based strictly on symptoms. Multiple disorders can share symptoms, but confusion in reaching a diagnosis can occasionally arise from how doctors decide to classify disorders. For example, beginning in 1994 all forms of attention-deficit disorder would be called “attention-deficit/hyperactivity disorder,” even if the person was not hyperactive. With molecular genetic data, it is likely possible to define the genetic etiologies of specific disorders. That is, conditions that are observationally difficult to distinguish may be composed of a small number of distinct conditions that differ based on their genetic makeup, thereby permitting new classifications of disorders [5]. Doing so and subsequently using genetic data would lead not just to a more accurate diagnosis, but also to benefits from more effective treatments [6]. Further, as knowledge about one’s genetic makeup increases, associated risk scores will provide individuals with an opportunity to respond to a condition more rapidly, and at lower costs than having an expert reach a diagnosis, before any symptoms manifest. In summary, the use of genetic markers has clarified the challenges researchers face in precisely measuring health in any empirical study, and may assist medical professionals in both diagnosing and treating health conditions.

Genetic lotteries and research design

Genetic markers are inherited from one’s biological parents, whose own markers are, in turn, inherited from their own biological parents. Dynastic effects relating to the line of heredity present obvious challenges in the use of genetic information as either control or instrumental variables, since, where unobserved, such effects may confound the estimates. Consider estimating the link between one’s genes and educational attainment. If there is a history of positive assortative mating in the family on the relevant genetic loci that affect educational outcomes—which would mean that these genes are positively correlated with educational achievement across generations—researchers may be led to overstate the relationship between an individual’s attainment and their own genes. While control variables could be added to capture parents’ or grandparents’ education, there remains a possibility of bias from omitting prior generations in the dynasty.

To overcome this challenge and obtain causal estimates, two researchers have introduced a new empirical strategy—coined the genetic lottery—which exploits genetic inheritance within full biological siblings in a bid to remove dynastic effects [7], [8]. While economists are familiar with natural experiments, the authors point out that within families, differences in the inheritance of specific genetic markers present the opportunity for additional experiments in “nature” [8]. The academically important contribution of the two studies is that the genetic lottery approach can be used to test whether family fixed-effect estimators on their own solve endogeneity problems.

The main finding in the first of the above studies is that inattentive symptoms in early childhood, assessed via a screening questionnaire, have large and lasting effects in reducing completed schooling [7]. The study also finds little consistent evidence that the overweight status of adolescents influences years of completed schooling. The authors argue that there are policy implications for the timing of interventions since health and productivity are often speculated to have a complex interdependence in the workplace. That is, health measures targeted at youth may affect future education and career choices, so there may be large benefits to individuals’ adult labor market outcomes from school-based programs that target childhood and adolescent health measures.

The availability of genetic data thus provides a way to test key identifying assumptions in this common research design, which has been applied in almost every branch of empirical economics as well as behavioral genetics. The genetic lottery provides a new research design for social scientists, and in many GWASs researchers now examine if the results are robust to controlling for family fixed effects.

The importance of genetic factors in social inequalities

Studies that use genetic markers as an instrumental variable are not suggesting that heredity is destiny. To illustrate this point, consider the rapid spread of obesity around the globe in the last 50 years. Researchers in the biological sciences have conducted searches for genetic variants that play a role in obesity, and have developed a rich evidence base for genetic and epigenetic mechanisms involved in the susceptibility and development of obesity. However, genetic changes remain largely stable across many generations of a population, so this is unlikely to explain the drastic rise in obesity rates over the last 50 years.

Many economists and social scientists have hence conducted a parallel program that explores whether changes in the environment that make it easier for people to overeat and harder for people to get enough physical activity can explain the rise in obesity. This research has influenced policy discussions related to taxes on sugar sweetened beverages, labeling of fast food in terms of calories, and the presence of soda machines in schools, among other proposals. Evidence underlying these policy debates often comes from studies that ignore genetic marker confounders, which may be an important consideration, particularly if identical policies or treatments have heterogeneous effects across individuals with different genetic makeups. These differential responses could be significant in cost–benefit exercises if a given policy is discovered to influence one group (defined here on genetic characteristics) substantially more than others. After all, just because genetic characteristics have historically been unobserved, this is not an excuse to engage in potentially discriminatory treatment now that genetic information is becoming available.

Obesity is a complex disorder with both genetic and environmental causes, and this requires researchers to use data and estimate models that allow environmental and genetic effects to interact in complicated ways. Further, to develop credible evidence on how genetic and environmental factors contribute to obesity requires exogenous variation in environmental conditions. A study from 2015 considers an alternative strategy which uses data from different birth cohorts in the offspring sample of the Framingham Heart Study (a US project of the National Heart, Lung, and Blood Institute and Boston University), and econometric models that can identify unknown breakpoints in relationships between a specific genetic factor and obesity (as measured by BMI) [9]. This genetic variant is of one well-studied obesity-promoting gene, commonly referred to as the FTO gene. The study tries to sort between differences in BMI within individuals over time (age effect), population-wide differences in BMI over time (period effect), and differences in the experienced period effect across individuals of varying ages (cohort effect). A robust relationship is found between birth cohort and the FTO risk allele with BMI, with an observed inflection point for those born after 1942 [9]. Figure 1 presents smoothed means of BMI for individuals sampled in the Framingham Heart Study with three unique genotypes and separated by birth cohort (before and during/after 1942). Individuals with two of the genotypes (AA and AT) in the later birth cohort have consistently higher BMIs over time, while those with the more common TT genotype do not. The results show that the environment one grows up in plays a large role in BMI, and that BMI also increases on average as one ages.

The above study also has important implications for the scientific literature focused on gene discovery since it raises the possibility that genetic associations may differ across birth cohorts due to variation in prevailing environmental contexts [9]. Many GWASs pool different data sources in order to increase the sample size to help identify small effects and increase their statistical power. However, doing so may introduce bias if information on differences across subjects’ birthdates and when and where the data were collected are not accounted for. These lessons hold for any study using genetic data to explain complex outcomes that result in social inequality and demonstrate the need for careful (econometric) modeling.

Genetics and public policy

Heritability plays a role in nearly every socio-economic and health outcome. This feature has long been ignored by social scientists and policymakers. However, heredity is not destiny and much work is needed to translate the revolutionary advances in genetics and genomics to reach both policy audiences and the broader academic community. The idea that a benevolent social planner can induce greater success by personalizing policy to individuals based on their genetic code has clear appeal to many, as does the concept of personalized medicine. Yet, as discussed earlier, issues related to privacy and discriminatory treatment based on genetic characteristics open a new set of challenges to designing effective policy.

The speed at which molecular genetic data can be effectively integrated within policy design is directly tied to improvements in understanding how genetic markers operate. For example, huge advantages may be gained if genetic screening can reliably predict complex learning disorders. That is, even if a disorder is a function of many genes, each with very small effects, researchers can calculate a single summary score from an ensemble of markers that have well-validated significant effects. The summary score provides a measure of an individual’s risk for a specific disorder or trait, which, in many situations, may take psychologists years to diagnose. Armed with knowledge of whether their child or employee is at an elevated risk for a given poor outcome, parents and employers will be able to make different investments, years prior to receiving a formal diagnosis. These investments may affect how the underlying genes are expressed and therefore reduce the risk for future poor outcomes. Moreover, as knowledge advances the predictive accuracy of these summary scores will increase. This could improve policy evaluations, since researchers could net out the contribution of genetic factors on the individual outcomes to provide cleaner evidence on the effectiveness of environmental inputs.

An important insight worth stressing is that even when a problem has its root in genetic factors, there is still a role for public policy [1]. Take the example of poor eyesight. Even if poor eyesight was strictly a result of genetic inheritance, policymakers could provide glasses to those afflicted. In other words, when discussing genetics and public policy, attention should not be focused upon the question of whether a specific outcome or trait is primarily a function of genes. Instead, policymakers must ask whether the available evidence suggests that a policy would pass a cost–benefit test. Any cost–benefit test could further consider the consequences of issuing a broad mandate versus targeting policy to those with specific characteristics. In summary, the success of integrating molecular genetics within a policy making environment does not require a wholesale transformation of how policies are developed.

Caution and care are needed when using and interpreting genetic data

Evidence from GWASs has been very influential and garnered significant press coverage in recent years. A strength of GWASs is that they investigate the full genetic code, in contrast to methods that specifically test a small number of prespecified genetic regions. Hence, GWASs represent a non-candidate-driven approach. Often, the popular press reports the results from GWASs as being causal, but they are only simple links between genetic variants at a specific location; the outcome under investigation and the mechanism by which a given genetic variant impacts a given outcome generally remains unknown. The effect size of any significant location is typically quite low, and it is rare to find a genetic variant that can account for more than 0.001% of the variation in traits of interest to labor economists. For example, evidence from a recent GWAS concludes that the effects of any single difference in genetic markers on self-employment are likely to be a very small proportion (<0.002%) of the variance in self-employment [10].

Most complex disorders are the result of many genetic variants, each with small effects. However, caution is required if one decides to change their behavior in response to this new information. In April 2017 the US Food and Drug Administration allowed the genetic testing firm 23andMe to sell reports with qualifiers showing customers whether they have an increased genetic risk of developing certain diseases and conditions. The number of conditions are limited and this reversed a decision in 2013 that forced 23andMe to stop the delivery of results on health-related traits.

An example where care is taken with reporting genetic information is the Stanford Cancer Institute tool for individuals with known mutations in the BRCA1 or BRCA2 genes, which put women at higher risk for cancer (available at http://brcatool.stanford.edu/). This calculator provides information on how the chances of survival change in response to different preventive measures at different ages. Yet this type of calculator is available only for a handful of genetic variants, suggesting that there may be unintended consequences from responding (especially over-responding) to a predisposition (that is correlational), as well as pointing to the possibility of patient-demanded medical care that may prove to be ineffective. Given the limited understanding of how genetic markers operate, the provision of this information without appropriate context and qualifiers could worsen outcomes.

The moral and ethical questions of how molecular genetic data should be utilized are heavily debated. These data may characterize an invasion of privacy and violate human rights. Concerns are emerging in multiple venues that the presence of molecular genetic data may require careful regulation on the part of governments. The principal concern is with regard to the potential for genetic data on inherited predispositions to influence decisions in either the workplace or on insurance coverage. For example, over 60 countries presently have a DNA database of convicted criminals, and there is debate about whether the benefits to society trump an individual’s right to privacy. Further, since these data can be linked to arrest records, they can also potentially be used by employers when deciding to hire, promote, or terminate workers. Similarly, these data may lead to disparate treatment or genetic discrimination by health insurers, who might refuse to give coverage to an individual who has a genetic difference that raises their odds of a specific health disorder. The rapid availability of molecular genetic data for use by others, for reasons that were not apparent at the time of data collection, may additionally constitute a violation of individual privacy, on top of the usual data security concerns. In summary, genetic data sets provide information on factors that were previously unobserved or ignored, but since this information is poorly understood, great care must be given to ensure it is used appropriately.