3.6  Physical health

Of all the wider benefits of education, the association with physical health has probably been the most widely studied, particularly within an extensive medical literature on the socio-economic determinants of health. There have been, for example, a large number of studies which look at the individual and combined effects of different socio-economic variables on adult mortality. By far the largest of these is the United States National Longitudinal Mortality Study (NLMS). This study takes its sample from nine Current Population Surveys conducted between 1979 and 1985.[12] Respondents aged 25 years and over were matched to the United States National Death Index to determine deaths occurring from the date of the initial survey up to 1989. A total of 530,507 men and women were included in the study, of whom 54,304 died during the follow-up period.

Using the NLMS, Sorlie, Backlund and Keller (1995) estimate the relative risk of mortality for men and women in different age groups. They find, after adjusting for individual year of age, that a longer education (measured by years completed) is associated with lower mortality in men and for women, with the strongest relationships in the 25 to 44 year age group and the weakest relationships in the 65 and over age group. Adjusting for socio-economic variables (race, employment status, family income, marital status and household size) reduces this association, showing that a good part of the relationship between education and mortality is due to socio-economic factors such as income. There still remains, however, a general pattern of declining mortality risk with increasing years of education.[13] This suggests that education has an effect on mortality which is independent of other socio-economic factors.

Other, smaller studies also show a relationship between mortality and years of education, after adjusting for demographic and social factors, although not all agree that this relationship is independent of income. Lantz, House, Lepkowski, Williams, Mero and Chen (1998) and Behrman, Sickles, Taubman and Yazbeck (1991), for example, find in their studies that the apparent effect of education on mortality is fully explained by the association between education and income.

Backlund, Sorlie and Johnson (1999) build on the work of Sorlie et al by describing the functional form of the relationship between education and mortality in the NLMS, using data on people aged 25 to 64. In particular, Backlund et al test whether there is a monotonic relationship in which more years of education leads to progressively lower risks of mortality. In fact they find that the best-fitting model for both men and women is a step function, where the steps correspond to the attainment of a high school diploma, at 12 years of education, and a college degree, at 16 years of education. This pattern of results suggests a credentialing or signalling effect: that the year spent getting a high school diploma (year 12), for example, is more valuable in the United States than completing year 11 or year 13, since at the end of year 12 the student leaves with a significant and status-defining qualification.

Persistent relationships between increased length of education and health have also been found when health is defined in terms of morbidity, risk factors for diseases, or self-assessed health status. Most of these studies are from the United States. For example, in a six-year follow-up study of almost 10,000 adults aged 51-61 years, Wilson (2001) finds that years of education is significantly associated with the incidence of four out of eight common chronic diseases which he monitors, after controlling for age, gender, race, household wealth, and some early life variables. In a cross-sectional study, Winkleby, Jatulis, Frank and Fortmann (1992) find that years of education is significantly associated with a set of cardiovascular disease risk factors (blood pressure, cholesterol, and cigarette smoking), after controlling for age, income and occupation. In another cross-sectional study, Ross and Mirowsky (1999) find that years of education is significantly associated with self-reported physical functioning and perceived health status, after controlling for age, gender, race, marital status, parents’ education and a range of employment and income variables.

Studies also try to uncover mediating factors – other than employment and income – which could explain a causal link between education and health. Amongst these, for example, are: better knowledge of, and practice of, health behaviours like exercise, smoking and diet (Schrijvers, Stronks, van de Mheen and Mackenbach 1999; Leigh 1983; Sander 1995a; Kenkel 1991); job sorting leading to less exposure to occupational hazards (Kemna 1987); a greater sense of personal control and social support (Ross and Mirowsky 1999); and more appropriate use of medical care. There is, however, no consensus in the literature on the relative contribution of these mediating factors. Studies look at different mediating factors using different datasets and, not surprisingly, get different results.

Two recent NBER papers use IV methods to investigate the relationship between education and health. Lleras-Muney (2002) tries to account for unobserved characteristics by using state compulsory schooling laws as an instrument for years of education. She applies this instrument to a sample from the 1960 United States Census of white people who were 14 years of age between 1915 and 1939, and observes a similarly-defined group in the censuses of 1970 and 1980. This allows her to calculate death rates for synthetic cohorts defined by gender, year of birth and state of birth. Lleras-Muney presents several IV estimations, each of which shows a statistically significant association between years of education and a 10-year probability of dying. This association suggests a reasonably large causal effect of education on mortality: an additional year of education is estimated to lower the probability of dying in the next 10 years by at least 3.6 percentage points. Further adjusting for income or occupation reduces this association but does not remove it, implying that education has an impact on mortality which is independent of any association with socio-economic status in adulthood.

Oreopoulos (2003) also uses state compulsory schooling laws as an instrument for years of education, and applies this instrument to data on people aged 25 to 84 in the 1990 and 2000 United States Censuses. These two Censuses asked questions about physical and mental health limitations. Oreopoulos finds that an additional year of compulsory schooling lowers by 1.7 percentage points the likelihood of reporting a disability that limits personal care, and lowers by 2.5 percentage points the likelihood of reporting a disability that limits daily activity. Oreopoulos also uses changes to the school leaving age in the United Kingdom as an instrument, and applies this to data from General Household Surveys in the 1980s and 1990s. This survey asked respondents to say whether they were in good, fair or poor health. Oreopoulos finds that a one-year increase in schooling lowers the probability of reporting being in poor health and raises the chances of reporting being in good health. He does not, however, adjust for income, so it is not clear whether education’s impact on mortality is independent of the effects of being better-paid.

In summary, the evidence from a wide range of longitudinal and cross-sectional studies in a number of countries, using different methods, different measures of health, and different control variables, indicates that better-educated people experience better health. This finding generally holds when the greater earnings of better-educated people are taken into account. Even the best of these studies, though, are plagued by a lack of control variables for early life factors, which may potentially influence both educational decisions and outcomes later in adulthood. Studies showing an association between education and health are therefore likely to have overstated the strength of this relationship, but it is almost impossible to know by how much, and whether this is significant. The sheer weight of studies showing a positive association makes a persuasive argument for saying that education has a causal effect on health, although it is not at all clear how great this effect is. It is also difficult to say what mediates education’s effect on health, and many different explanations have been proffered. It may be that there are many different channels operating.

Other pieces of evidence also point to a causal relationship between education and health. We might expect to see a relationship, for example, since education seems to be related to cigarette smoking (section 3.1) and smoking is the leading cause of preventable death in New Zealand (Ministry of Health 2002). IV studies also point to a causal relationship, at least for students compelled to stay in school through compulsory schooling changes. Because of the way they are designed, IV studies such as that of Lleras-Muney (2002) avoid the problems of omitted variables which plague longitudinal and cross-sectional studies. However, as with Lochner and Moretti's (2001) study on imprisonment (section 3.3), there is a question over how generalisable they are to contemporary New Zealand. For example, Lleras-Muney’s sample was of people who were affected by compulsory schooling policies in the United States between 1915 and 1939, and it is not clear that her results can be generalised to New Zealand in 2004, with a completely different education system and where students already receive a minimum of ten years schooling.

3.7  Social connectedness and political participation

Social capital refers to the stock of active connections between people, as constituted by participation in, and knowledge of, civic affairs; trust in other people; and reciprocal help and support in the community. Interest in social capital has been motivated by apparent associations between levels of social capital and social and economic outcomes (Putnam 2000). Bynner and Egerton (2001) use the National Child Development Study, a longitudinal study of all children born in Britain in one week in 1958, to investigate associations between educational attainment, measured by qualifications, and later life outcomes, measured at age 33. After controlling for some parental and socio-economic variables, they find a persistent association between qualifications and membership of organisations, and also between qualifications and the likelihood of voting in local and general elections. These results suggest that education may help to build social capital.

Plausible mechanisms can be thought of to explain the connection between education and social capital. Without a basic level of literacy, for example, it would be difficult to know about public events or public resources, to read the newspaper, or to know what is going on in the community. Schools are also places where students are inculcated with similar knowledge and values, and where they learn to interact with children of different backgrounds (Heyneman 2001).

Brehm and Rahn (1997) develop and test a model in which individuals contribute to social capital by having a degree of trust in other people and by being members of groups and organisations. Using data from the annual United States General Social Surveys between 1972 and 1994, Brehm and Rahn find that duration of education is strongly related to membership of groups, after accounting for measures of psychological engagement, resources and social conditions. Education was also related to feeling that other people could be trusted, after accounting for measures of demographics, family background, resources, being a victim of crime, and life satisfaction. Better-educated respondents belonged to more groups and were more trusting of other people; increased education may therefore contribute to social capital.

Voluntary work is a manifestation of help and support in the community and increased volunteering may therefore contribute to social capital. Wolfe and Haveman (2001) include increased volunteer work as one of the wider benefits of education, but the studies they refer to may have been biased by failing to account for account for unobserved family characteristics which might affect both school achievement and an individual’s willingness or ability to do voluntary work. Gibson (2001) uses data from a New Zealand sample of identical twins to test whether this is the case. Using identical twins brought up together accounts at one stroke for both inherited ability (since identical twins have the same genetic structure) and shared family background. Gibson finds that increased education in fact significantly reduces the probability of volunteering. In addition, for those who do volunteer, increased education significantly decreases the number of hours of volunteer work that are performed.

Karp and Banducci (1999) look at voter turnout in New Zealand using the New Zealand Election Study, which is held after each general election. Controlling for age, ethnicity, political leanings and party preferences, they find that increased educational qualifications were significantly associated with a greater probability of voting, in both the 1990 election (under first past the post) and the 1996 election (under proportional representation).

Two recent NBER papers on voter participation use instruments for the quantity of education, in an attempt to control for unobserved characteristics. Using different datasets and different instruments, Milligan, Moretti and Oreopoulos (2003) and Dee (2003) both find that educational attainment has a large and statistically significant effect on subsequent voter participation in the United States.[14] Voter turnout in the United States is poor, however, so the results of these studies may not be generalisable. With regard to voting, New Zealand is more like the United Kingdom, where people are legally required, and actively assisted, to register as voters and where there is a reasonably high turnout in general elections. Milligan et al (2003) investigate voting behaviour in the United Kingdom, using data from British Election Studies between 1964 and 1997. Normal regression estimates, controlling for gender and age, show that an extra year of education has only a small effect on the probability of voting, increasing this by about one percentage point. Instrumental variables estimates – again using compulsory schooling laws as an instrument for years of education – are similar in magnitude but statistically insignificant.

In summary, the evidence for a causal connection between education and social capital seems relatively weak, mainly because studies tend to lack information on early life factors. Apparent associations between education and later behaviour may therefore be confounded by unobserved family or other background characteristics. For example, parents who encourage their children to pursue more education might also tend to nurture strong civic virtues. The best study in this regard is the twin study of Gibson (2001), which in fact shows that increased education decreases a person’s likelihood of doing volunteer work. The results of IV studies showing a strong effect of education on voting in the United States may not be transferable to the United Kingdom or to New Zealand. In general, it is not apparent that one more year of education significantly adds to the social skills, civic values, collective responsibility, etc., inculcated during 10 or 11 years of compulsory schooling.