|Home | About | Journals | Submit | Contact Us | Français|
Reducing health disparities in the United States has been targeted by numerous policies. We examine educational disparities in mortality and life expectancy among non-Hispanic blacks and whites in the 1980s and 1990s. Despite increased attention and substantial dollars directed to groups with low socioeconomic status, within race and sex groups, the educational gap in life expectancy is rising, mainly due to rising differentials among the elderly. With the exception of black males all recent gains in life expectancy at age 25 occurred among better educated groups, raising educational differentials in life expectancy by 30 percent. Differential trends in smoking-related diseases explain at least 20 percent of this trend.
Disparities in mortality and morbidity across educational groups in the United States have been prevalent for decades. Attention to these disparities intensified with Kitagawa and Hauser's comprehensive national account of educational mortality differentials in 1973,1 and a wave of related disparities research followed.2 Heartened by early successes in the Healthy People 2000 initiative to reduce disparities in health care and health outcomes, the Healthy People 2010 initiative aims to eliminate health disparities entirely by the end of this decade. Groups such as the Institute of Medicine and many private organizations have also promoted this agenda.3 In 2003, the Agency for Health Care Research and Quality produced its first annual National Healthcare Disparities Report to track racial and socioeconomic disparities in health and health care.4
Several health interventions have been designed to address racial and socioeconomic disparities in health. Well-known examples include the Disproportionate Share Program ($12.9 billion annually)5 and the State Children's Health Insurance Programs and related Medicaid expansions ($7 billion annually). In addition, much of the $6.6 billion annual budget for the Health Resources and Services Administration (HRSA) targets disadvantaged populations via funding for community health centers and the Ryan White Care program for HIV/AIDS. The current wave of state initiatives to expand health insurance coverage explicitly targets low-SES children and adults. Finally, efforts to improve public health promotion and disease prevention activities have expanded over time, with a particular focus on risk factors that are prevalent among populations with low socioeconomic status: tobacco use, obesity, and under-use of prevention and screening services.6
Recent research on racial disparities demonstrates mortality gains for blacks relative to whites during the 1990s7, but data on educational disparities in health are not comprehensive. Most research measuring educational health disparities ends in the 1980s. These studies show widening differences in mortality by education.8 Many of the programs noted above post-date that era, however. More recent evidence focuses on area-level disparities.9 But, significant migration and demographic change means that area-level analyses may understate or overstate the full extent of health disparities at the individual-level.10 Another large study after 1990 pools data from 1959 through 1996, making it difficult to attribute trends to a single decade.11 Furthermore, analyses of death by cause can augment our understanding of educational disparities in mortality because some causes of death are more amenable to intervention than others, suggesting different policy approaches to ameliorate disparities.12
We analyzed life expectancy and mortality trends by education group, and then estimated which diseases account for differential trends in mortality.
We used two different sources of data to estimate mortality trends. We matched Census population estimates to death certificate data in the Mortality, Multiple-Cause-of-Death (MCD) files for 1990 and 2000. 13 The nearly universal sample of deaths makes the MCD attractive, but there are questions about the accuracy of education reporting on death certificates.14 To validate these data and to assess an earlier time period, we used the National Longitudinal Mortality Study (NLMS), which followed members of the March Current Population Survey (CPS) through subsequent deaths. The CPS data span years 1981−85 and 1991−95, and deaths span 1981−1988 and 1991−1998.15 The initial CPS sample is non-institutionalized. For this reason, mortality rates in the NLMS are below those in the MCD, which includes all deaths. We restricted analyses to non-Hispanic blacks and whites to limit the impact of immigration on our estimates. The Appendix summarizes the data and sample characteristics.
Survey respondents in the NLMS and Census respondents report education, while informants, usually next of kin, report education on death certificates. Previous studies document some inconsistencies between these methods, in particular overstatement of high school graduation relative to incomplete high school experience on death certificates.16 As a result, we chose two broad categories of education where agreement is high. Low-education refers to “12 or fewer years of education” and high education indicates at least 13 years of schooling. This cutoff yields the following shares of high-education individuals in our study: 47.3% in 1990 and 54.5% in 2000 using Census data, and 43.9% in 1981−1988 and 56.1% in 1991−1998 in the NLMS.
Because rising educational attainment over time could affect our results, we re-computed estimates after equalizing the share of individuals in the high and low education groups. We did this by randomly reassigning some individuals with 12 years of education to the high education group in the early period to match education shares in the latter period. The patterns we document are robust to the changing composition within education groups (Appendix Table A6).
For a given age-race-sex-education group (e.g. white low-education women aged 65−74 in 1990), we divided the number of deaths by the group's population to compute age-specific mortality rates. These were calculated for ages 25−84 since most education is completed by 25, and few NLMS participants survived beyond 84. To incorporate mortality after age 84, we assigned published death rates within age-sex-race groups.17 Period life expectancy is then constructed in the usual fashion. Details regarding the mortality and life expectancy estimates are available in the appendix.
For 1990 and 2000 we estimated cause-specific mortality rates based on the five leading causes of death: diseases of the heart, cancers, cerebrovascular diseases, chronic obstructive pulmonary disease (COPD), and unintentional injuries.18 Because ample evidence implicates tobacco use as the underlying cause for a substantial portion of deaths,19 we distinguished lung from other cancers, yielding six cause-of-death categories.20
We computed life expectancy by education both within race-sex groups and for the entire study population, holding the race and sex composition constant at year 2000 shares. We also calculated age-standardized mortality rates using 10-year age strata with the year 2000 population standard. The changes in these rates from 1990 to 2000 were decomposed by cause of death and age group, as described in the Appendix. We computed standard errors for 1990, 2000, and changes over time using the Delta Method, and NLMS variance estimates accounted for the complex survey design.21 Lastly, given the importance of causes of death for which smoking is a risk factor, we computed the sex-specific, age-adjusted share of the adult population (25 and older) reporting they were “current smokers” for selected years between 1966 and 2002, in various annual National Health Interview Surveys.22
Between the 1980s and 2000, life expectancy increases occurred nearly exclusively among high-education groups (Exhibit 1). Comparing 1981−88 with 1991−98, life expectancy at age 25 grew 1.4 years for high education individuals, but only 0.5 years for less educated individuals, a 0.9 year difference (p = 0.014). Between 1990 and 2000, life expectancy grew 1.6 years for the high education group, but remained unchanged for the low education group (p<.001 for the difference). As expected, life expectancy in the MCD data was lower than in the NLMS because of the inclusion of the institutionalized population. In both datasets, educational gaps in life expectancy increased by about 30%. This similarity is striking given the differences in sampling frame and education measures between the sources.
Further, the levels of the education differences are large. In 2000, life expectancy for a 25 year-old with a high school degree or less was 50 years. For a person with some college, life expectancy was nearly 57 years.
Widening educational disparities in life expectancy do not arise from increased racial or gender differences in mortality trends. Our population estimates hold demographics constant at their 2000 level. More fundamentally, Exhibit 2 shows similar trends within each race and sex group. The growing educational gap in life expectancy is most pronounced among women, regardless of race. Life expectancy at age 25 declined among less-educated black and white women, while rising 1 year or more for more educated women. By the year 2000, highly educated women could expect to live over five years longer than their lesser educated counterparts, in both races. For white men the gap was 7.8 years in 2000, up 1.6 years from 1990, and for black men it was 8.4 years, up by 1.3 years.
Overall, and even within-race trends in mortality by education mask another important trend. Although SES differences in mortality were rising, mortality differences across genders and races were falling. From 1990 to 2000, the life expectancy differences between blacks and whites decreased by 1.8 years for low-education males, by 1.5 years for high-education males, and by 0.7 and 0.6 years for low- and high-education females. Life expectancy for low-education white males increased in both the NLMS and MCD data while it decreased for females. Within the high-education group, life expectancy increased for both men and women, but the gains for men were almost double that for women. Racial and gender gaps in mortality persist, but they are smaller now than they were in the 1980s.
Diseases of the heart, cancers, and COPD contributed over 60% of deaths in the sample in 1990 and 2000 in all race and sex groups. Exhibit 3 shows the contribution of each cause to the increase in educational mortality gaps. Diseases of the heart and cancers excluding lung cancer contributed most to rising education differentials, 32%. Lung cancer and COPD, two diseases largely attributable to tobacco use, accounted for another 21%. These diseases were particularly salient among less educated white women over the age of 45, among whom lung cancer and COPD contributed about 25% of rising educational gaps in mortality (Appendix Table A7). Exhibit 4 shows that by the year 2000, lung cancer and COPD death rates were twice as high among low education white men and women and black men, compared to the more educated in these groups.
Increased education differentials among the elderly account for much of the growing gaps in mortality and life expectancy. Within each race-gender group, at least half of the growth in life expectancy gaps and over 60% of the growth in mortality gaps come from individuals aged 65 or older (Appendix Tables A8-A9). In contrast, trends among adults aged 25−44 contributed little to growing gaps in mortality, and among young black men aged 25−44, educational mortality gaps narrowed from 1990−2000. Among the causes of death we examined, the narrowing educational differential among black men aged 25−44 was driven by unintentional injuries and heart disease deaths (Appendix Table A7).23
Adult smoking rates have declined significantly since the mid 1960s (exhibit 5). Among men, the reduction in smoking prevalence was relatively even by education, though somewhat greater for the better educated. Between 1966 and 1995, rates dropped 22 percentage points among the better educated versus 18 percentage points among less educated men, despite the fact that better educated men already had lower smoking rates in 1966.
Smoking among women increased from 1965 through the late 1970s, and then began to decline. The decline from 1979 through 1995 was 12 percentage points among the better educated and 5 percentage points among the less educated. Interestingly, the sharp divergence in smoking rates by education occurred between 1979 and 1983, coincident with the first Surgeon General's Report highlighting smoking risks for women. Over the entire time period, smoking among female college attendees declined by over 16 percentage points, compared with only 7.5 percentage points among less educated women.
The 1980s and 1990s were periods of rapidly rising life expectancy, but the mortality declines that yielded these gains did not occur evenly by educational group. On average, we find very little change in life expectancy among less educated black and white non-Hispanics, and very substantial increases in life expectancy among the more educated. These patterns mirror similar widening of education differentials in disability and self-reported health status over the same period.24 The growing gap in life expectancy by education occurred during a period of increasing attention to health disparities and increased public spending designed to improve the health of less advantaged populations.
One important exception to this pattern is that educational mortality disparities narrowed among young black men, a finding consistent with recent evidence that racial mortality gaps narrowed in the 1990s. Nevertheless, a 5 year gap in life expectancy between blacks and whites remains.25 Across sex groups, men made faster gains in mortality than women over this period, narrowing mortality and life expectancy gaps by sex.
Our data span a period of rapidly rising income inequality, providing one potential explanation for widening educational disparities. However, data do not support this explanation because health disparities narrowed across race and gender as inequality increased.26
Our results suggest that differential trends in smoking may explain a significant part of widening gaps in mortality and life expectancy. The diseases contributing most to the growing educational gap in mortality include diseases of the heart, lung and other cancers, and COPD, all of which share tobacco use as a major risk factor. Lung cancer and COPD alone account for one quarter of the increasing gap in life expectancy for women over 45, consistent with their sharp divergence in smoking rates during the 1980s. For men, the divergence in smoking was more moderate, as was the increase in the mortality gap attributable to tobacco-related causes of death.
Public policy designed to reduce health consequences related to smoking may have indirectly contributed to this disparity. In the half century since the harms of smoking became widely known, tobacco control measures have proliferated. Cigarette labels warning of the health hazards of smoking have been required since 1966. Cigarette advertising was banned from television and radio in 1971.27 During the 1980s and 1990s, many states and localities instituted smoking bans in the workplace. By 1993, 70% of indoor workers had smoking bans in work,28 and by 2007, every state had some smoke free air provision.29 In addition, cigarette taxes have increased rapidly in recent years, after falling in real terms in the 1970s. On net, the real price of cigarettes has nearly tripled since the 1960s.30
The proliferation of tobacco control policies brought remarkable reductions in tobacco use. In the four decades following the 1964 Surgeon General's report, per capita annual consumption of cigarettes among adults fell by half. However, declines were greatest among the most educated groups. The growing gap in mortality by education for smoking-related causes supports the longstanding paradox that attention to prevention can widen disparities in health across education and income groups.31 In this case, the advances related to knowledge about risk factor control; we cannot say whether the same is true about medical technologies.32
The focus on tobacco does not imply that other efforts to reduce disparities in health were not successful. Indeed, we confirm recent work highlighting relative gains in life expectancy overall for blacks compared with whites during the 1990s.33 Further, other studies have shown that Medicaid expansions targeting low-income pregnant women and children improved health outcomes among these populations.34 Our study does not argue that these policies were unsuccessful. Rather, it suggests that these efforts were swimming against a strong tide, one which overwhelmed billions of dollars spent annually on additional medical care. On the more positive note, our results suggest that one place to look for real progress is tobacco control efforts for low SES groups because mortality trends mimic trends in smoking that occurred decades earlier. These long-run consequences of health behaviors bolster the argument for early childhood intervention.35
The explanation for differential smoking trends is complex. Basic knowledge does not appear to be the major issue. In 1986, 90% of Americans surveyed across the board reported that smoking causes lung cancer and emphysema, and 80% believed it contributed to heart disease and bronchitis.36 Translating knowledge into action has proven more complex, however. Innovations that target less-advantaged groups might offset this unintended consequence of medical progress. Some caution about this conclusion is needed, however. Without addressing the underlying factors that lead less educated individuals to be less able or willing to invest in better health, measures to reduce smoking may simply lead to a shift from tobacco-related deaths to other causes.37
Beyond the differential change in smoking, there is the national trend towards increased obesity. As with smoking, obesity is more common among the less educated than among the better educated. Further, recent research suggests that obesity might contribute to nearly as many deaths as tobacco.38 Although the population health consequences of obesity remain controversial, the obesity trends in recent years and into the future could further widen socioeconomic gaps in health.
In summary, during a period of increased focus on disparities in health, virtually all gains in life expectancy occurred among highly educated groups. Causes of death related to differential trends in cigarette smoking by education have contributed substantially to rising mortality differentials. Larger and better targeted efforts to push successful health interventions into less educated groups may be needed to achieve the goal of reducing disparities in health by socioeconomic group.
We gratefully acknowledge funding from the Russell Sage Social Inequality and Health Project, National Institute of Aging grants P30-AG012810 and P01-AG005842, and National Institute on Drug Abuse grant 5K01DA19485-2. Drs. Nicholas Christakis, Christopher Jencks, and Kirsten Smith provided helpful comments on the manuscript. We thank Norman Johnson, Eric Backlund and U.S. Census Bureau staff for for producing estimates from the National Longitudinal Mortality Study.