Evidence On The Mechanisms Linking Education And Health
To be convincing, studies of the effect of education on health will need to understand the pathways that link the two. Because there are a large number of potential mechanisms, this is a difficult task. In addition, the evidence on mechanisms is somewhat weaker than the evidence on causality, because often assumptions about what constitutes a mechanism have to be made.
Some studies have attempted to look at why education matters for health. Consider the evidence on the effect of education on sexual behaviors and fertility. An important reason why education improves outcomes for girls is that it delays marriage and fertility, because the common practice is for girls to marry soon after finishing school. This, in turn, means girls will have fewer years of ‘exposure’ to get pregnant, and thus fewer children over their lifetime. Also girls in school have children later, which is beneficial because reproduction during the early teenage years is riskier for the health of the mother and the infant compared with reproduction in prime adult years.
The results from the randomized trial in the Dominican Republic also seem to be driven in part by the incarceration effect: most boys who are not in school start working or are idle – the set of people whom they interact with when they are not in school is different from their peers in school, and ‘treated boys’ (those given the message about the value of education) report that their peers are significantly less likely to drink and smoke. Note further that early exposure to a different set of peers could have important long-term consequences, as smoking and drinking are addictive behaviors that affect youth’s physical and mental development.
Consider now the natural experiments that use compulsory schooling as an instrument for education. In the US in the 1910s, children who were not in school were either idle or working. The main occupation for children of ages 10–15 years at the time was agriculture. Agricultural work is substantially more hazardous to health than school work. Thus, it is possible that the health effects of forcing children to stay in school during this period are driven by the difference in health hazards across environments. However, by the 1940s the types of jobs adolescents engaged in when they were not in school were substantially different, and perhaps not as hazardous. This may explain why the returns to post-WWII compulsory education in the UK were smaller.
However, the evidence suggests that the effect of education is not limited to this incarceration effect alone. Uniform provision in Kenya delayed marriage well beyond the increase in years of schooling generated by the intervention, so at least in this case, incarceration alone cannot explain the observed effects.
Another possibility is that education matters (sometimes) for health because schools directly provide information on how to improve health, and it is the health information itself, rather than being in school that affects behaviors. More educated individuals are indeed better informed about health risks in developed countries. And when information first becomes available, it seems to first become known to the more educated, who, in turn, seem to be the first to respond. Educated mothers stopped smoking at higher rates after the 1964
Surgeon General Report first widely publicized the harms of smoking, and their babies’ health increased more as a result. Smoking rates started declining for the best educated in the 1950s, before the Surgeon General’s report, as the dangers of smoking were increasingly discovered. Similarly in Uganda in 1990, there was no relationship between education and HIV, but one emerged by 2000 after a decade of information campaigns on prevention. In the UK, when information was first (incorrectly) reported about possible autism risks associated with the mumps, measles, and rubella (MMR) vaccines, vaccination rates fell more in areas with more educated individuals. In fact, in some studies it appears as if all of the effect of education is explained by information, for example, studies find that most of the effect of maternal education on child height can be explained by differences in information.
But information cannot be the whole explanation; differences are observed in health behaviors by education even when there are no differences in information by education. For example, in the experiment in the Dominican Republic that informed children on the returns to school, there were no differences in the extent to which smoking and drinking were perceived as harmful by the treated and the control boys, and yet the treated boys stayed in school longer, smoked less, and drank less. Similarly, in developed countries today, knowledge of the harms of smoking is nearly universal, and although there are some small differences by education in knowledge, these differences are very small compared with the differences in smoking rates by education. Curriculum interventions alone had little impact on behavior in the Kenyan intervention. Finally, observational studies suggest that a small portion of the effect of education on behaviors is due to differences in knowledge. It appears that when knowledge first becomes available on how to improve health, it substantially increases education disparities. But in the long run, information diffuses and other factors are more important in explaining the associations between education and health.
In this sense, information may be like other innovations in health. For example, more educated individuals are more likely to use recently approved drugs than the less educated, and this appears to be driven by those with chronic conditions who use drugs repeatedly, suggesting that learning is an important component of the education effect. Similarly, in developing countries, more educated individuals are generally more likely to adopt new innovations. Whether the initial advantage of the educated fades away or gets stronger with time, might, in turn, depend on the type of health technology. For example, some medication regimes are difficult to adhere to, and the educated might have a permanent advantage at using them – this is the case for diabetes type 1. Other innovations instead are ‘deskilling,’ such as the birth control pill, in which case eventually the less educated catch up. The results from malaria interventions provide some interesting evidence on this point: when access to malaria treatment improves, the gap in access between the educated and the uneducated falls. However the educated still behave quite differently from the uneducated in their treatment-seeking behaviors: they appear to be more likely to know the likelihood that they have malaria and they are more likely to visit a health-care center and less likely to use other treatments when their symptoms are worse. This is not true among illiterate individuals.
The evidence from randomized interventions suggests that some mechanisms are important, whereas others are not – but certainly as this paper discussion suggests the extent to which any findings are generalizable is not clear. Some of the effects of schooling might operate through the incarceration effect as already discussed. Another important mechanism is income, as the Malawi conditional cash transfer intervention suggests. Finally peers are also important. In the Dominican Republic intervention discount rates, risk aversion and health information were not affected by the intervention, even when schooling increased. However, treated boys had lower incomes and reported that their peers drank and smoked less – these two channels most likely explain the observed decreases in smoking and drinking among the treated.
Interestingly, this evidence is consistent with the exploratory and descriptive studies. Rough calculations from these suggest that observed factors can account for approximately 70% of the effect of education (in a statistical sense), through resources (30%), family and friends (10%), and information (10%) and cognition (20%). However, risk aversion, discounting, stress, and other personality traits did not appear to mediate the relationship between education and behaviors – although the noise in these measures gives one some pause.
Summary
On balance, the literature reviewed highlights a wealth of interactions between education and health. Education appears to be causally related to health in many settings, but not always, and the reverse is true as well.
Equally important, this review highlights some unanswered issues. The most important issue is to understand in more detail when and how education translates into health. To what extent is education associated with specific knowledge, with cognitive ability in general, or with different social settings, either during school or after? Some evidence on this may come from looking at the quality of education individuals receive. Most of the literature has looked at the impact of additional years of schooling. Yet many of the theories say that the quality of the years should matter as well. This has not been explored in any great detail.
Simple experimental designs that randomly encourage individuals to obtain schooling can be useful in providing further evidence of causality on health and health behaviors, but they cannot conclusively answer the question of whether education alone is responsible for the observed effects because in general it is difficult to satisfy the exclusion restriction that is needed to reach such conclusions. However, more sophisticated designs could be implemented to help identify mechanisms and causality both. For example, one could design an experiment with three treated groups, where individuals are given unconditional cash transfers (cash-only group), conditional cash transfers if they attend school (attendance group), and conditional cash transfers for both going to school and obtaining good grades (performance group). Under the assumption that all treatments induce changes in education, income, and grades, the separate effects of education, income, and health can be learned. By comparing the controls with the cash-only group one can estimate the effects of income on health and health behaviors. By comparing the outcomes of the cash-only group and the attendance group one can obtain an estimate of the effects of attendance. Finally, by comparing the performance group and the attendance group one can learn about the effects of education content.
Furthermore, it is vitally important to understand the translation from intention into action. In developed countries, everyone knows the behaviors that are good for health and (as suspected) many would like to improve their health. Yet people systematically fail at this task, that is, they struggle to change their behaviors. How are these failures understood, and what types of interventions would reduce them? In a way, this is asking for a benchmark by which to compare education. Improving health by inducing more education is costly; many people do not enjoy schooling, and forcing additional years of schooling comes at a price. If the impact of education on health can be replicated using other methods, this would be very attractive.
In sum, the burgeoning literature on education and health is just the beginning. A review written a decade from now will ideally have many more specific conclusions to draw.