Understanding The Relationship Between Education And Health
Education and health may be related for three reasons: poor health in early life may lead to less educational attainment; lower educational attainment may adversely affect subsequent health; or some third factor such as differences in discount rates may affect education and health-seeking behavior. Each of these pathways are briefly discussed.
The next section starts by describing the most important commonly unobserved determinants of both education and health. The first is parental resources: parents with more resources (broadly construed to include wealth, social networks, knowledge, etc.) will devote part of them to improving the survival of their children (by investing in their health) and also to improving their future outcomes, which, in turn, means they will invest perhaps more on their children’s education. Second, there are some important individual characteristics that theoretically are expected to increase both education and health. Ceteris paribus, more patient individuals are more likely to invest more in both education and health. Also, smarter individuals might be more likely to obtain more schooling and also have better health.
Effect Of Early-Life Health On Education
As the previous results indicate, there is a very strong correlation between early life indicators of health (such as height) and educational attainment – and this is true across all countries of the world. These correlations have been documented many times before, particularly in developing countries. As education is largely determined at young ages, this suggests that at least part of the correlation between education and health among adults is due to the fact that unhealthy children obtain few years of schooling and become unhealthy adults.
Recent studies show that the relationship observed – shorter (and sicker) children obtain less education – is a causal one. Two types of studies investigate the causal effect of health shocks on human capital accumulation: some take advantage of the so-called ‘natural experiments,’ whereas others use randomized controlled trials to investigate the question.
Most studies that investigate this causal chain find support for it: there are several examples of how disease and nutrition affect human capital formation. For instance, individuals affected in utero by the 1918 influenza pandemic obtained fewer years of schooling than those not affected. Individuals born during the Great Famine in China had lower educational achievement than those not born during the Great Famine. Malaria eradication in the US, various Latin American countries, and Sri Lanka resulted in greater education, although malaria eradication in India did not. Deworming campaigns had substantial effects on schooling in early-twentieth century American South and in Kenya today.
A related literature explores the consequences of birth weight on adult outcomes and finds similar results suggesting that those born with lower birth weights have lower levels of education, income, and health as adults. Although these studies do not directly look at nutrition, but rather at extreme events that influence birth weight, in many cases nutrition and disease are the most likely intervening mechanisms.
Direct evidence on the effect of nutrition and disease on education is available from several randomized experiments. Nutritional supplements, iron supplementation, and iodine supplementation trials in utero or during early childhood have resulted in higher educational attainment and increased cognitive ability.
Whether early life health affects education through morbidity at younger ages or expectation of life extension at older ages is unknown. Indeed, there is scant evidence on the extent to which expectations of longer life affects schooling but some evidence suggest that this channel also matters: when maternal mortality fell in Sri Lanka, girls’ education increased (but not that of boys). But it is not clear as to what extent the education–life expectancy relationship is accounted by this channel.
Overall, the evidence is consistent in showing that nutrition and disease shocks early in life are quite detrimental for human capital formation. Interestingly, the reduction in educational attainment associated with early-life health insults is not the only theoretical possibility. Sickness increases the cost of going to school in terms of effort and might also lower the returns to school if it lowers life expectancy. Thus, parents of sick children might optimally choose lower levels of schooling for those children. However, illness also increases the cost of work and might increase the returns to school (in terms of avoiding more physically demanding jobs). Thus, it could be that the return to schooling increases as people become less healthy. However, there is no empirical evidence of this alternative, although perhaps it explains why education and height are negatively related in two very rich countries: Finland and Luxembourg (Figure 9). This discussion also underscores the fact that the observed relationship between height and education reflects not only the physical effects of disease in childhood but also the behavioral responses of parents which might attenuate or exacerbate the effects of the health shock itself.
Given that health is an important determinant of schooling and the fact that education and health could simply be determined by common factors such as parental resources, it is extremely challenging to document whether in addition to these well-documented relationships, education itself affects health – this question is considered next.
The Effect Of Education On Health: Theory
Theoretical foundations for a causal effect of education on health were first provided by the seminal work of Grossman (1972), based on the human capital model of Becker (1964). One key insight of Grossman’s model of health capital is that individuals derive utility from health directly (they do not like being sick) and indirectly by affecting labor market outcomes (sick individuals work less and earn less). The other essential feature of the model is the recognition that there is a ‘health production function’ – that there are known factors that individuals (or institutions) can manipulate in order to affect health in predictable ways. These two features give rise to a behavioral model in which individuals demand medical care, food, and other goods and services because they are aware these factors will improve their health and ultimately increase their utility. (See Strauss and Thomas (2008) for an excellent exposition of the theoretical production of health over the life course and its determinants.)
In this type of model, education can affect health in a variety of ways. Most obviously, education affects the type of jobs that individuals get and the income they earn. A year of education raises income by at least 7%, and this is true in both developed and developing countries. Higher incomes increase the demand for better health, but they affect health in other ways as well. Richer people can afford gyms and healthier foods; they can also afford more cigarettes. Furthermore, when an individual’s wage increases, it raises the opportunity cost of time: because many health inputs require time (such as exercise or doctor visits or cooking), in the short run, wage increases might reduce health. Thus, the income associated with higher education may or may not improve health.
Higher educated individuals are also more likely to take jobs that provide health insurance and other benefits such as retirement accounts. Although one expects these benefits to have a positive effect on health, it is theoretically possible that they do not. For example, individuals with insurance could be less likely to care for themselves because they face lower financial costs in the event of a disease. However, because the uncompensated costs of disease are large (morbidity and premature mortality), it is not expected that these indirect effects would dominate the access associated with better insurance.
Finally, more educated people work in different industries and occupations than less educated people. To the extent that job characteristics affect health, sorting into jobs may affect health as well. At the dawn of the industrial era, this relationship was undoubtedly positive. Early in the twentieth century, the more educated were more likely to work in white collar occupations, which were substantially safer than working in agriculture or manufacturing (fewer accidents, exposure to chemicals, physical strain, etc.). Today, most individuals work in the service sector and the better educated may have jobs that are worse for their health – for example, they spend more time sitting in front of computers, which could turn out to be bad: sitting (independently of exercise) has been recently shown to detrimental to health.
Thus, the effect of education on health, through its effect on the labor market, is ambiguous. Moreover, a positive association between education and disease can arise through the conscious choices of individuals: individuals may well know that exercise is needed to remain in good physical shape, but they may optimally trade off some of their health for increased incomes when wages are high. At the extreme, when individuals have no other resources than their bodies to earn a living, they will optimally ‘use up’ their bodies to earn a living: trading off higher lifetime earnings for shorter, sicker lives. The theory of compensating differentials predicts just that: individuals can be ‘paid off’ to accept risky occupations.
The second mechanism explored by Grossman is that education can affect the production function of health directly, acting as a ‘technology’ parameter. This is the so-called ‘productive efficiency’ mechanism, in contrast to the ‘allocative efficiency’ mechanism which has already been described (the more educated optimally chose different levels of health inputs because they face different prices and budget constraints). In its simplest formulation, productive efficiency posits that the better educated will have better health outcomes, even conditional on access to the same health inputs at the same prices. Better use of information is the classic example. More educated individuals might be better at following doctor’s instructions (because they may have better self-control for instance) or they might be more likely to believe the information produced by the scientific establishment and follow its recommendations perhaps (because they took science courses in school or know scientists directly).
Car safety knowledge provides another interesting case. Both more and less educated people strongly agree that one should wear a seatbelt while driving a car. But when the survey question is asked a different way, the pattern changes: the less educated are much more likely to agree with the statement that seatbelts are just as likely to harm as help you in an accident. It may be that better educated people have a deeper understanding of the risks of not wearing a seatbelt and the probabilities that go into a calculation of optimal seatbelt use. Another example concerns how successful individuals are at using certain health technologies such as devices to help quit smoking. Conditional on making an attempt to quit smoking, the better educated are more likely to be successful quitters.
There is a third theoretical reason why education could be related to health: education could change the ‘taste’ for a longer, healthier life. For example, education may lower individuals’ discount rates, making them more ‘patient.’ There are two reasons for this. First, attending school per se is an exercise in delaying gratification, and school may teach patience; this may carry over into other aspects of life. Second, to the extent that individuals can ‘choose’ or learn what to like (in other words if discount rates can be chosen), then those with more education have a greater incentive to choose patience, because they face steeper income profiles over their lifetimes. The same argument might hold for risk aversion.
Finally, education affects the peers that individuals spend time with, and different peer sets may encourage different health behaviors. This is particularly important in the context of health, given that many health behaviors have an important social component. For example, individuals generally drink together and often smoke together. More generally, peers are thought to be essential in determining risky behaviors. Also, peers and social networks are an important source of information, and of financial, physical, and emotional support and hence can affect whether individuals get sick and how well individuals fare when they do. If on average more educated individuals have more educated peers, they will have access to a greater set of resources. If more educated individuals are more likely to be better informed (because they learned so in school or because they remain better informed later), then peers will help individuals reinforce their knowledge, in a ‘multiplier’ setting.
Note, however, that peers can influence behavior in a positive or negative manner. A peer group that focuses on sedentary lifestyles and lack long-term investment may encourage that same behavior among all members of the peer group, but one that focuses on exercise and fitness would promote the opposite.
Beyond the Grossman model, there are other theories that predict associations between education and health. The most prominent is that education predicts rank in society, and those with higher rank are in better health than those with lower rank. In small hierarchical groups such as apes and (perhaps) humans, those at the top will have access to more resources and greater control over their lives in general, whereas those at the bottom will have both fewer resources and control. As a consequence, those at the bottom will suffer more ‘stress’ and this, in turn, lowers immune responses and increases the likelihood of short-term illness and long-term chronic disease.
This theory has been shown to be accurate among mammals and other species (Sapolsky, 2004) and has been tested experimentally with animals to rule out genetic factors as the main explanation (e.g., the top of one hierarchy will suffer in health if they are transferred into a different group where they have a lower place in the hierarchy). Although it is not entirely clear whether and how this theory applies to humans in large modern societies – where Bibliography: groups are multiple and they are chosen endogenously – it provides another rationale by which education may affect health. It is to be noted that this theory has an interesting prediction: if all that matters is relative rank in society, a society with higher average levels of education may have no better outcomes than a society with lower average levels of education.
Education may also affect health because the things that kids do while in school are different than what they do outside of school. Although this is a trivial observation, this so-called ‘incarceration effect’ is extremely important to consider. For example, children in school may have less exposure to criminal activity or poor role models.
Finally, there are other possibilities. The more educated could inadvertently be better or worse off because of biological processes that are not well understood. For example, more educated women have higher mortality rates of cancers of the reproductive system. It has been hypothesized that this ‘wrong’ gradient emerges because more educated women have fewer children, and having children turns out to be protective from certain cancers. Overall, education appears to lower mortality even after all health behaviors are accounted for, which suggests that some of these nonbehavioral mechanisms might be important – although it is not obvious that all important health behaviors can be observed.
Certainly it is very likely that many of these mechanisms are at play at any one time and place and in combination they will yield complex patterns. The complex relationship between education and HIV in Africa is an interesting case in point – de Walque reports that ‘‘education predicts protective behaviors like condom use, use of counseling and testing, discussion among spouses, and knowledge, but it also predicts a higher level of infidelity and a lower level of abstinence.’’ In this example it would appear as if the educated not only seek out information at higher rates, know more, and use their information and resources to purchase protection but also have some higher risky behaviors, perhaps because of their higher incomes or lower risk (they can ‘afford’ it).
Evidence On The Causal Effect Of Education
A large number of early studies found supporting evidence for the Grossman model using largely descriptive tools. The usual prediction tested was that education and health were positively correlated. Clearly they are; the literature struggled with instruments for education to determine causality. However, these studies were not entirely convincing about whether education had a causal effect on health, because descriptive methods and imperfect instruments are not well suited to establishing causality.
A second generation of studies attempted to provide clearer evidence of a causal link between education and health again using ‘natural experiments.’ Many of these studies make use of compulsory schooling as a source of plausibly exogenous variation of education to investigate whether more school improves adult health. The intuition for this approach is simple: some individuals are forced to attend school longer because of compulsory school legislation, and researchers can examine whether the health of those who are forced to obtain more schooling improves compared with the health of those who are not required to stay in school. Studies in the US, Denmark, Sweden, the UK, and Germany using changes in compulsory schooling find that indeed these laws ultimately improved the health of the affected populations. However, recent work finds no effect of the same compulsory schooling laws on health in England and Sweden, and a study focusing on France also finds no effect of education on mortality.
The literature that has estimated the effect of education on health behaviors using natural experiments is also mixed. For example, some find that schooling lowers smoking rates but other studies find no evidence that schooling affects smoking behavior.
It is difficult to interpret this conflicting evidence. All of the papers that find positive effects of education on health use natural experiments to construct instrumental variables (IV) estimates of the impact of education. They tend to find effects that are larger than OLS. Although this has generally been interpreted as reflecting heterogeneity of treatment effects (those that are affected by the legislation have larger returns), the alternative interpretation is that the ‘natural experiment’ did not in fact work well as an experiment, and there is still substantial bias in the education estimate. For example, the results for using compulsory schooling reforms in the US are not robust to the inclusion of state-specific trends. However, there is very little variation left once these controls are added, so it is not clear whether the effects are truly overestimated or whether the variation in the laws is not sufficient to estimate an effect of education. This discussion underscores the limitations of IV studies in general. From a methodological point of view, the regression discontinuity studies make the fewest assumptions, and they find no effects of education on health.
Also interesting to note is that available studies report impacts along different margins, not only because of the obvious reason that they study different times and places but also because the ‘experiments’ themselves are different. In the UK, the changes in compulsory schooling were strictly followed and an entire cohort of individuals was forced to obtain almost 1 more year of schooling as a result. In contrast, in the US, the laws that are typically studied increased educational attainment by 0.05 of a year – that is only 1 in 20 individuals obtained one more year of schooling. There are two important differences here. First, the affected population in the US is a small sample among those that were potentially affected – it is indeed possible that returns are different for this subset. Second, in the US, only a few individuals in a given cohort and place were affected, but entire cohorts were affected in the UK. If, for example, education matters because it affects a person’s rank in society, then in the US, those who stayed in school had their rank increased relative to the counterfactual of no compulsory schooling law. This would not necessarily have been the case in the UK: an entire cohort increased their education by approximately 1 year, so an individual’s rank within their cohort was unaffected by the policy.
It is also theoretically possible that the effect of education varies over time and place, and that the results from the previous studies correctly document this variation. Indeed, the international evidence suggests that the returns to education do vary across countries. It is notable that the two studies that find no effects of education in the UK and France, study cohorts during and shortly after World War II (WWII), a time when the income returns to education were falling and generally low.
The fact that the effect of education on labor market earnings itself is causal also suggests a positive effect of schooling: if schooling is rewarded in the labor market because it raises productivity, how does it do so? Whatever general human capital is learnt in school and rewarded in the labor market might also be useful in the production of health, because it is useful in the production of goods. If education makes workers better by making them better decision makers or better able to deal with complexity or uncertainty, then these abilities can be used in other domains, in particular for health.
One central conclusion of this discussion is that investigating the specific mechanisms by which education affects health would improve the understanding of education–health link substantially. The following paragraphs discuss what is known about this next, after describing the latest attempts to infer causality in the literature.
In addition to natural experiments described at the beginning of this section, there are a variety of experimental interventions that have been carried out, mostly in developing countries, that can be used to infer the effect of education on health. In Kenya, random distribution of school uniforms – a significant cost associated with school – among upper primary-level students increased levels of schooling for both genders by a substantial amount (the dropout rate fell by 18%). Seven years later, treated girls had significantly lower rate of marriage and pregnancy, but the treatment had no effect on sexually transmitted diseases. However, random provision of HIV information to the curriculum of some students had no effect on sexually transmitted diseases, but the rate of unwed teenage pregnancies fell.
Many countries have implemented conditional cash transfers programs to help the poor. Conditional cash transfers are transfer programs where the receipt of income is conditional on certain behaviors, generally related to health or schooling. Unconditional cash transfers do not have any strings attached. Studies find that the conditional cash transfer programs have resulted in lower levels of sexual activity, teen pregnancy, and marriage rates among young girls in the short term, in addition to increasing schooling in Africa.
Although curriculum information on HIV in Africa had little effect on schooling, other information campaigns have worked. For example, a small intervention in the Dominican Republic informed 14-year-old boys about the labor market returns to school. The intervention successfully increased schooling by 0.2 years, and significantly decreased work in the formal labor market. As a consequence of this, treated boys delayed debut of heavy drinking and were less likely to smoke than untreated boys.
These studies suggest that education affects specific health behaviors, but not all behaviors. However, even here, it is not clear that one can infer that education is the ultimate cause of the changes in the observed health behavior. The gold standard for establishing causality would call for randomly assigning individuals to various levels of education. Clearly, this approach is unethical and unfeasible. Instead, these studies look at an ‘intent-to-treat’ intervention, where individuals are randomly ‘incentivized’ to obtain different levels of education. With this design, it is possible to estimate the effect of education on health, if (1) the intervention successfully raises education levels and (2) the random incentives that are provided to increase schooling affect health only through education (the exclusion restriction assumption).
In this light, consider whether randomized interventions that potentially raise schooling can be used to estimate the causal effect of schooling. Typically, interventions are designed so that reasonably sized effects on education can be detected with the chosen sample. But even if this requirement is met and the intervention increases education levels, the intervention must induce students to attend school but not directly or indirectly impact any other determinant of health. It is difficult to design an intervention that meets this assumption. Providing scholarships to those that are credit constrained is equivalent to increasing income in the short run, which directly or indirectly is likely to affect health. Providing uniforms is not quite like providing income, but it increases incomes indirectly by substituting for household spending. The more constrained individuals are in their consumption, and the higher the effect of the intervention on schooling is, the more likely it is that the income effects of the intervention are large. Finally, informing misinformed students of the returns to school affects the present discounted value of earnings of all participants, regardless of whether they are induced to attend school or not. Because health (and its determinants) is likely to depend on permanent rather than temporary (current) income, this intervention also fails the exclusion restriction assumption.
Another important limitation of randomized interventions is that in the short run schooling is not expected to affect health because the young are generally in excellent health and because health is a stock – instead it is expected that the health effects emerge slowly and cumulate. But it is difficult and expensive to follow individuals for many years; the interventions above follow individuals for several years but on average the participants are still quite young at the last follow-up (e.g., in the Dominican Republic study, the intervention takes place when boys are 14 years old and they are 18 years old when they are last interviewed). The interventions then look at health behaviors, but it is not clear how these effects will eventually translate into, for example, mortality.
There are only two studies of randomized education interventions that follow individuals over a long period of time. One looks at the participants Perry Preschool School program (PPP) 37 years later and the other looks at the participants of the Carolina Abecederian (ABC) Project at the age of 21 years. Both of these interventions occurred early in childhood, and they have been shown to have had persistent effects on wages and other outcomes. The results from these two studies are again in conflict: the treated students in the ABC program had significantly better health than the controls, but that was not true in the PPP program, although in both cases the treated appear to have better health behaviors. These results are to be taken with caution as in both cases the number of observations consists of only approximately 100 individuals.
Thus, simple randomized trials cannot conclusively answer the question of whether education affects health. But it is possible to make progress on this question by investigating mechanisms through which interventions affect education or designing more complex randomized interventions. The authors discuss the evidence on mechanisms next and conclude with a series of observations on what questions could be explored in future research.