SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

Both body weight and educational attainment have risen in the United States. Empirical evidence regarding educational differences in obesity (BMI ≥30) is inconsistent. According to some widely cited claims, these differences have declined since the 1970s, and the most educated have experienced the greatest gain in obesity. Prior research was limited in grouping college graduates with nongraduates, combining men and women in the same analysis, and using self-reported rather than measured anthropometric information. Using the National Health and Nutrition Examination Surveys (NHANES), we address these issues and examine changing educational differences in obesity from 1971–1980 to 1999–2006 for non-Hispanic whites and blacks in two separate age groups (25–44 vs. 45–64 years). We find that (i) obesity differentials by education have remained largely stable, (ii) compared with college graduates, less educated whites and younger black women continue to be more likely to be obese, (iii) but the differentials are larger for women than men, and weak or nonexistent among black men and older black women. There are exceptions to the overall trend. The obesity gap has widened between the two groups of college-educated younger women, but disappeared between the least and most educated younger white men. Thus, the increase in obesity was similar for most educational groups, but significantly greater for younger women with some college and smaller for younger white men without a high-school degree. Lumping together the two distinct college groups has biased previous estimates of educational differences in obesity.

Since the 1970s, obesity (BMI ≥30) has more than doubled (from 15 to 33%) among American adults (1). In the meantime, the proportion of the population with at least some college education has increased from 12 to 56% (2). Empirical evidence regarding education-related obesity differentials is inconsistent. Some widely cited claims based on analyses of the National Health and Nutrition Examination Surveys (NHANES) are: (i) educational differences in obesity have declined, for women in particular since the 1970s; (ii) although obesity continues to be inversely related to education among women, it no longer differs by education among men; and (iii) the greatest increase in obesity has occurred with the most educated men and women (3,4). Several analyses of the National Health Interview Surveys (NHIS) found that compared with college graduates, people with a high-school degree or some college education have become increasingly more likely to be obese, but among themselves, the less educated groups have become harder to distinguish from each other (5,6,7). Two of the NHIS analyses, however, largely ignored the divergence results, concluding that educational differences have declined over time (5,7).

The NHANES data are noted for their comparable survey designs and physically measured anthropometric information, whereas the weight and height data in the NHIS are self-reported, which tends to underestimate excess weight (8). Among women in developed countries, studies using self reports are more likely to show an inverse relationship between socioeconomic status and obesity (9). None of the NHIS studies were separate by gender, although the NHANES results suggest gender differences in the education—obesity association. Incongruous classifications of education could be another issue. The NHIS results suggest significant and increasing distinctions between college graduates and nongraduates (5,6,7), but this distinction was not made in all previous studies. In the NHANES analyses (3,4), education was classified into low (<9 years of schooling), medium (9–12 years of schooling) and high levels (at least some college). We examine the time trends in educational differences in obesity for non-Hispanic Americans separate by gender and race, using the NHANES data with measured weight and height and a classification of education that distinguishes college graduates.

METHODS AND PROCEDURES

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

Three NHANES samples are analyzed with respective baselines in 1971–1974 (NHANES I), 1976–1980 (NHANES II), and 1999–2006 (continuous NHANES, released in four 2-year cycles) (10). Conducted by the National Center for Health Statistics, the NHANES are multistage stratified and clustered probability samples of the US noninstitutionalized population. Oversampled subgroups are those at high risk of malnutrition in the NHANES I and II (low-income persons, preschool children, women of childbearing age and the elderly), and low-income persons, adolescents aged 12–19 years, persons aged 60 years and above, African Americans, and Mexican Americans in the continuous NHANES. Sociodemographic and health information is collected through an in-home interview and a physical exam at a mobile examination center.

The BMI, calculated as weight in kilograms divided by squared height in meters is classified into three categories: normal-weight (BMI 18.5– <25 kg/m2), overweight (BMI 25– <30 kg/m2), and obese (BMI ≥30 kg/m2) (11). We exclude underweight cases (BMI <18.5 kg/m2), representing <2% and 5% of the male and female sample, respectively. Education has four levels based on highest completed grade or degree: less than high school (grade <12), high-school degree (grade 12 or GED), some college (13–15 years of school, or associates degree), and college graduates (at least 16 years of school or bachelor degree, as reference). One key difference from previous studies (3,4,12) is in distinguishing college graduates from those with some college.

The analysis is restricted to non-Hispanic whites and blacks aged 25–64 at baseline. For the first two NHANES, Hispanics are identified based on national origin (13). After deleting 282 cases missing for education or anthropometric measures, 612 women pregnant at time of survey and 584 underweight cases, the analysis sample has 21,504 subjects. Table 1 provides the sample size and BMI distribution by levels of education.

Table 1.  Change over time in BMI distributionsa by education, 1971–1974, 1976–1980, and 1999–2006 NHANES
inline image

Educational differences in obesity are analyzed by the prevalence model, and interaction terms between time (1971–1980 vs. 1999–2006) and education indicate the change over time in the education-obesity association (for detailed model specifications, see the Supplementary Appendix online). The analysis is done separately for each of the four gender-race groups, separate by age (25–44 vs. 45–64). Results are similar when age is adjusted for within each age strata. The NHANES I and II data are pooled to represent the 1971–1980 period, as the education-specific BMI distribution appears stable across the first two NHANES.

The analysis is implemented by the SAS (version 9.2; SAS Institute, Cary, NC) PROC GENMOD procedure with a binomial distribution and identity link function (14), and survey design effects are accounted for by the Generalized Estimating Equations framework (15). Sample weights are normalized within each of the six surveys/cycles. Results do not differ substantially when the sample weights for the continuous NHANES are constructed from a weighting procedure developed for analyzing multiple 2-year cycles (16), or when the analysis is unweighted, with no adjustment for survey clustering and stratification. As the analysis does not adjust for additional covariates, model estimates are identical with cross-tabulated results, except for rounding error. The modeling, however, provides estimates of sampling error and direct statistical tests.

RESULTS

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

Table 1 compares the obese prevalence across educational categories and time periods, based on model estimates shown in the Supplementary Appendix online. For white women, educational differences in the obese percentage were all positive, indicating an overall elevation in the obese tendency of noncollege graduates relative to graduates. Changes over time in these differences were also positive, and statistically significant for women with some college. Thus, differences in obesity widened over time for the two college groups, and there was no evidence of a decline for other comparisons. In 1999–2006, the obese prevalence was lowest among college-graduated women (20% at ages 25–44 and 32% at ages 45–64), and significantly higher by 8–18 points for the less educated.

Although male educational differences were smaller than female ones, education and obesity tended to be negatively related among white men in both periods as well. One exception to the overall pattern was younger men who dropped out of high school. They had the highest obese percentage in 1971–1980 (15.7% vs. 6.3% for college graduates), but in 1999–2006, they were nonsignificantly less likely to be obese than the most educated (18.3% vs. 24.8%). The 16-point decrease in the difference was significant. Noteworthy is that in no other comparisons had white male educational differences declined over time, and at ages 45–64, the obese prevalence in 1999–2006 was highest among high-school dropouts (42%).

The black samples were substantially smaller, especially in the earlier period for the two college groups. Therefore, results for blacks were more uncertain. For younger black women, the less educated were more or no less likely to be obese than college graduates, and the obesity gap between the two college groups increased with moderate significance, which was similar to white women. For older black women, educational differences appeared to be declining to nonsignificance in 1999–2006. But the decline was largely caused by the unusually low obese percentage of college graduates in the small 1971–1980 sample (19 cases).

Educational differences in obesity were negligible among older black men in both periods, and among younger black men in 1999–2006. The elevation in obesity in 1971–1980 for younger less educated blacks was significant if all noncollege graduates were combined, thus suggesting a decline in the differences. But the number of college graduates was small in the 1971–1980 sample (27 cases).

Finally, obesity increased over time at all educational levels for all age-gender-race groups (last column, Table 1). Educational differences in the changes are equivalent to changes in the educational differences (last but one column, Table 1). Thus, the increase was similar across most educational groups, but significantly greater for younger women with some college and smaller for younger white men without a high-school degree. The significantly smaller increase for the least educated older black women was dubious due to a small sample, as pointed out above.

DISCUSSION

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

Since the 1970s, educational differences in obesity have remained stable overall. Among whites and younger black women aged 25–44, the less educated continue to be more likely to be obese than college graduates, but the differentials are larger for women than men, and weak or nonexistent among black men and older black women aged 45–64. The stability and persistence of educational differences contrast with previous claims that obesity differentials have declined over time, and the most rapid increase in obesity has occurred with the most educated (3,4,5,7,12). Given the heterogeneities between college graduates and nongraduates (unchanged among men but widening among younger women), lumping them together as done previously (3,4,12) is perhaps unwise, biasing downward estimates of educational differences and leading to spurious trends.

Prior research has also examined obesity differentials by income (13). Income and education may tap different features of one's socioeconomic position, and thus their respective associations with body weight require different interpretations. Education signals not only potential wages and earnings that could be used to purchase health-related goods and services, but also cognitive skills and access to information that affect health behavior. Education may be less affected by obesity and other health conditions in adulthood than income. Our additional analysis adjusting for income (measured by poverty income ratio) found no substantial change in the results.

There are exceptions to the stable trend. The increasing obese tendency of younger women with some college relative to college graduates could be due to the changed population composition. The dramatic expansion of the higher education system made college enrollment possible for people who would not have been able to enter college in the past.

That the obesity gap has declined between the least and most educated white men, and the former have become less likely to be obese (albeit nonsignificantly so as shown here) than the latter has attracted much attention (3,4). What is missing in the discussion is that this pertains to younger men only. Among older men, high-school dropouts continued to be more likely to be obese than college graduates, and actually had the highest obese prevalence of all educational groups in 1999–2006. Additional analysis found that despite a lower overall obese percentage, the least educated younger men had a slightly higher percentage of severe obesity (BMI ≥35) than the most educated. The extreme tendencies (to be both severely obese and lean) could result from a clustering of risk factors such as unbalanced diet, lack of exercise and smoking.

Findings about black Americans are less certain due to small samples. The weak or lack of association may reflect racial differences in weight ideals. Obesity has been more prevalent among blacks than whites at the population level, likely affecting how individual characteristics such as race, education, and BMI are associated. Such normative differences may not be universal, because obesity differentials among high-school graduated younger black women are as strong as among their white counterparts. Regardless of race and gender, smoking, which could lead to wasting illnesses, has become more concentrated among the less educated in more recent birth cohorts (17). Further research is needed to obtain a more holistic view of health.

While highlighting the continued educational differences in obesity, this research provides strong evidence of the obesogenic environment to which the total population is subject. Obesity has risen for all. College-graduated white women have been least likely to be obese, but with a 14–18 point increase, their obese percentage is now as high as that of women who did not finish high school 30 years ago. Public health programs should target both less educated groups and the general environment that affects the way of living for all.

ACKNOWLEDGMENTS

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

This paper received no specific funding. For comments and suggestions, I am grateful to Larry Wu and 3 anonymous reviewers.

REFERENCES

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. METHODS AND PROCEDURES
  4. RESULTS
  5. DISCUSSION
  6. SUPPLEMENTARY MATERIAL
  7. ACKNOWLEDGMENTS
  8. DISCLOSURE
  9. REFERENCES
  10. Supporting Information

supporting Information

FilenameFormatSizeDescription
oby_904_sm_x1.doc125KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.