Background We test the reversal hypothesis, which suggests that the relationship between obesity and education depends on the economic development in the country; in poor countries, obesity is more prevalent in the higher educated groups, while in rich countries the association is reversed—higher prevalence in the lower educated.
Methods We assembled a data set on obesity and education including 412 921 individuals from 70 countries in the period 2002–2013. Gross domestic product (GDP) per capita was used as a measure of economic development. We assessed the association between obesity and GDP by education using a two-stage mixed effects model. Country-specific educational inequalities in obesity were investigated using regression-based inequality indices.
Results The reversal hypothesis was supported by our results in men and women. Obesity was positively associated with country GDP only among individuals with lower levels of education, while this association was absent or reduced in those with higher levels of education. This pattern was more pronounced in women than in men. Furthermore, educational inequalities in obesity were reversed with GDP; in low-income countries, obesity was more prevalent in individuals with higher education, in medium-income and high-income countries, obesity shifts to be more prevalent among those with lower levels of education.
Conclusions Obesity and economic development were positively associated. Our findings suggest that education might mitigate this effect. Global and national action aimed at the obesity epidemic should take this into account.
- Health inequalities
- MULTILEVEL MODELLING
- INTERNATIONAL HLTH
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Worldwide, the proportion of overweight adults has increased between 1980 and 2013, from 29% to 37% in men and from 30% to 38% in women.1 There are marked variations between countries in the obesity levels and trends; hence, cross-country comparisons might identify factors that are driving these differences.1 One such factor frequently posed is economic development.2 ,3 The reversal hypothesis, which suggests that socioeconomic status (SES) and obesity are positively associated in low-development countries but negatively associated in high-development countries may be important to take into account in the development of policy action.2 ,4
The reversal hypothesis was first explored in a series of literature reviews on the relationship between SES (measured by education, income or occupation) and obesity.5–8 These reviews, which included studies from different countries and time periods, persistently found patterns in accordance with the hypothesis. This has later been tested empirically in cross-country studies using cross-sectional data, where most studies find support for the hypothesis; however, some dispute the trend.3 ,4 ,9–14
One possible explanation for why rising gross domestic product (GDP) increases obesity prevalence among those with lower education is that physical demands decrease with rising GDP.3 ,4 For example, reductions in manual labour and new transportation means can lower energy consumption. Second, rising income may be used to purchase greater quantities of food.3 ,4 ,6 ,7 ,9 ,14 ,15 These effects may be absent in those with higher education. Increased GDP may, to a lower extent, affect the degree of physical strain in the workplace of those with higher education. Furthermore, those with higher education are less affected by food scarcity and may already be exposed to environments that are conducive to obesity. Education may also affect health-related decision-making, increasing the ability to better control one’s weight level.9 ,16 Thereby, education may represent one channel which mitigates the increased risk of obesity in individuals in richer countries.
The potential reversal in the SES associations might vary by gender.4 ,5 Health-related decision-making might be influenced by socially constructed body weight norms and ideas, which often differ for men and women within the same society. For example, obesity is severely stigmatised among women, especially in developed societies.7 ,5 In addition, the differences in manual versus professional occupations in terms of physical demands are often more pronounced in men, which have resulted in differences in the SES–obesity association by gender in England.17 If the overall employment-related physical demands decrease with GDP, these effects might be different for men and women. There is evidence to suggest that the impact of obesity on health, health service use and employment is more pronounced in women;18–20 hence, the societal consequences of the obesity epidemic depend on the gender pattern.
The objective of the following study was to test the reversal hypothesis in men and women, by investigating the association between economic development and obesity across educational groups. In addition, we set out to analyse educational inequalities in obesity and determine whether or not they vary by economic development. We used individual-level data from a broad range of countries with high, medium and low GDP. The compilation of data from 70 countries, covering 69% of the world's population in 2010, allowed us to conduct an extensive analysis of how the interface between individual-level factors and societal factors shapes the obesity gradient.
We have assembled a data set consisting of nationally representative health surveys from 70 countries across the world between 2002 and 2013. The data set is a merger of the World Health Survey (WHS), the Study on Global Aging and Adult Health (SAGE), the Health and Retirement Study (HRS), the Survey of Health Ageing and Retirement in Europe (SHARE), the International Social Survey Programme (ISSP) and the Norwegian Living Conditions Survey (health section).21–26 We include these data sets as they have measures of self-reported height and weight, education and demographic characteristics. We focus on self-reported, rather than measured, height and weight. Our final data set, excluding nations with incomplete information, consists of 412 921 individuals aged 30 and above from 70 countries who were surveyed between 2002 and 2013. See the online supplementary file for more information.
Individuals were classiﬁed into weight categories based on their BMI, which is their weight in kilograms divided by the square of height in metres. Heights reported in feet and inches were converted to centimetres and weights reported in pounds were converted to kilos. One is considered obese if BMI ≥30 kg/m2 according to the WHO guidelines.27
Education was selected as our indicator of socioeconomic position as it was available for most countries in our data set. Using the International Standard Classification of Education (ISECD-97), we collapsed the levels of education in each data set to four main levels (no education: ISCED 0; primary education: ISCED 1; secondary education: ISCED 2–4; tertiary education: ISCED 5–8).
We excluded individuals with missing information on height, weight, age, gender and education.
To explore the two aims of our study, we fitted two separate models.
Multilevel mixed effects model
First, we investigated the interface between macrolevel context represented by GDP and individual education. Hence, we needed to incorporate two levels of influence in our models (individual and country levels). Multilevel mixed effects models are well suited for this purpose; by treating individuals as nested within countries, they allow for estimation of both individual and country characteristics.4 ,28 ,29
We used the following random-intercept logistic regression model. 1where i indexes the individuals, j indexes the 70 countries, y is the obesity status for individual i in country j, A is the age group, and T is a linear year trend denoting the time of the interview for the individual. E is education, X is GDP in country j and we also include an interaction between the individual education and GDP. Ϛj is a random intercept which is assumed to be independent across countries j and independent of the other covariates in the models. We ran regressions separately in men and women.
The use of survey weights in multilevel logit models is an unresolved issue.29 ,30 We thus followed the recommended practice and fitted models using both random and fixed country intercepts with and without first stage cross-sectional individual sample weights and compared the results.28 ,29 As expected, there were only minor variations in the coefficient values which did not affect the conclusions. We present results from the random intercept models without weights only.
The second aim was to investigate educational inequalities in obesity and whether or not this varied by economic development. To measure the magnitude of relative and absolute educational inequalities in obesity, we estimated the Relative Index of Inequality (RII) and Slope Index of Inequality (SII), which are recommended when making comparisons over time or across populations.31 These indices are regression-based and take the whole socioeconomic distribution into account, rather than only comparing the two most extreme groups. Educational level at each survey was transformed into a summary measure that was scaled from zero (highest level of education) to one (lowest level of education) and was weighted to reflect the share of the sample at each educational level. The population in each education category was assigned a modified ridit-score based on the midpoint of the range in the cumulative distribution of the population of participants in the given category. For example, if the most educated individuals comprise 18% of the population, the range of individuals in this category was assigned a value of 0.09 (0.18/2), and if the second category comprises 50% of the population, the corresponding value is 0.43 (0.18+(0.5/2)) and so forth.
We used generalised linear models (log-binomial regression), with a logarithmic link function to calculate RIIs (prevalence ratios), and with an identity link function to calculate SIIs (prevalence differences).32–34 Both indices were estimated using the following generalised linear model: 2where i indexes the individuals, R is the ridit-score, C is a categorical variable representing 70 categories for each country. We control for age (A) and survey year (T), and run regressions separately in men and women. This equation allows the association between obesity and the ridit-score to vary between countries, which is captured by the inclusion of the two-way interaction (R×Ci). Equation (2) is used to estimate RII, when the link function g(yi)=log(yi) and SII when the link function g(yi)=yi. Results are provided using cross-sectional individual sample weights or individual sample weights for longitudinal data.
The highest prevalence of obesity was found in Swaziland for men and in South Africa for women. The lowest obesity prevalences were found in India, South Korea, Laos and Myanmar (see online supplementary table S1)⇓.
The obesity-GDP association by education
Table 2 shows the results of the mixed effects multilevel logit models. We find that GDP was positively associated with obesity; however, the interactions between education and GDP were significant for each level of education in men and women. Hence, the association between GDP and obesity status is modified by education.
Figure 1, which is based on models with interactions in table 2, shows a sharp increase in obesity prevalence with higher GDP in those with no education, especially in women. In those with primary education, the trend was similar but weaker. In those with secondary education, we found a positive association between obesity and GDP in men but not women. In those with tertiary education, there was little or no increase in obesity with higher GDP in men and women. The relationships between obesity prevalence and GDP were significant in those with no education in men and women, significant in men and weakly significant in women with primary education. The association was significant in men with secondary education; however, it was not significant in women with secondary education. The associations were insignificant in both genders with tertiary education.
The association between educational inequalities in obesity and GDP
Both the relative and absolute educational inequalities in obesity were positively associated with the GDP of the country (figures 2 and 3). At lower levels of GDP, the relative inequality was below 1 and the absolute inequality was negative, meaning that the prevalence of obesity was higher in those with higher education compared with those with lower education.
We also observed differences by region; in Sub-Saharan Africa, the prevalence of obesity was more common in those with high education compared with those with low education. While educational inequalities were absent in Latin America and the Caribbean, in the high-income countries and in the rest of Europe obesity was more prevalent in those with lower education. These patterns were similar in men and women.
Both the USA and the UK had significantly positive absolute inequalities in men and women (see online supplementary table S2). Similar trends were observed in the Nordic countries (Denmark, Norway, Sweden and Finland) and the rest of the northern European countries. There were significant negative absolute inequalities in many African countries (Ghana in women, Kenya, Malawi in women, Mauritania in women, Namibia, Swaziland in women and in Zambia). In China, the absolute inequality was not significant, while in India highly educated men and women were significantly more likely to be obese than men and women with lower education. In Russia, the absolute educational inequalities in obesity were positive and significant in women.
Owing to concerns regarding time trends and variations by age across the data, we reran our analysis using a subsample including data for 2006/2007 over the age of 50. Although this analysis includes only 20 countries, similar trends were observed.
This study shows a stronger obesity-GDP association in those with lower levels of education than among those with higher education. The study also illustrated that at lower levels of GDP there was a positive educational gradient in obesity, while at higher levels of GDP there was a negative educational gradient. This supports the reversal hypothesis and the results were consistent in men and women.
Earlier literature reviews on the SES obesity association support the reversal hypothesis.5–7 So does a study conducted with a data set consisting of European countries.14 Of eight identified studies using individual-level data, six use the Demographic and Health Survey (DHS) data which comprised only women between the ages of 15–49 years in low-development and middle-development countries.3 ,4 ,9–13 Of these, four studies found support for the reversal hypothesis, while two dispute whether a reversal is actually taking place.3 ,9–13 Different conclusions were reached possibly because they include different selections of countries.
Only one study does not use the DHS and includes both genders and countries from Western Europe, North America and Australia. This was conducted by Pampel et al,4 who used the WHS from 2003 for 67 countries. They used imputed values due to a number of concerns regarding the accuracy of the self-reported data in the WHS as well as a high share of individuals with missing BMI.3 Their findings support the reversal hypothesis in men and women. Our study, which is conducted in a larger data set, supports their findings. In addition, we illustrate that the association between obesity and GDP in those with no education is more pronounced in women.
What might account for the reversal of the association between obesity and education by economic development? As discussed in the introduction, this might be related to food consumption and changes in physical demands.3 ,4 ,6 ,7 Economic development tends to be accompanied by more efficient food production that can decrease costs of a particularly high-calorie diet and thereby increase consumption thereof.3 Aggressive market strategies for energy-dense food can increase with economic development and we may expect a general rise in obesity.9 As populations are increasingly exposed to high-calorie products and a more sedentary lifestyle, greater knowledge and self-control are necessary to adjust everyday behaviour in order to avoid weight gain.4 Education is conducive to a better understanding of the health and social costs of excess weight as well as the ability to increase exercise levels and adjust one's diet.4 ,9 Furthermore, it has been argued that societal and cultural standards of attractiveness and health ideals differ by development status.5 Attributes such as good health and a thin body tend to be more valued in developed countries and are more easily achieved by those with higher education due to better access to nutritional information and greater engagement in non-work-related physical activity.16 In addition, the physical exertion associated with labour reduces body weight in those with lower education in less developed countries. When GDP increases jobs become more sedentary.3 Consequently, in more developed countries, physical exertion associated with labour contributes less to the educational gradient in obesity.
Our findings support the view of obesity as a social phenomenon. In order to reach the WHO's obesity target by 2025, fundamental social and political changes in the developing and developed countries are needed. This study emphasises education as a crucial variable in achieving lifestyle change at the population level, and targeted interventions by educational status could be warranted. However, more research on the causal structure between obesity and socioeconomic factors at the individual and country levels is needed. Future research must examine the credibility of these findings and their implications for policymaking.
Our study has potential limitations. Although the data sets used are gathered using standardised survey methodology, there are variations across the data sets in the methods used. The cross-sectional nature of the data limits inferences about causality and we only assess associations, which may be biased. There may be simultaneity bias: while education affects BMI, it is also possible that the reverse is true and BMI affects education. There may be omitted variable bias: it is likely that there are unobserved factors that affect both obesity and education. There may be measurement error: if BMI is mismeasured (eg, if it is based on self-reported height and weight and if the level of mismeasurement is associated with education), the relationship will be biased.35 BMI has also been criticised, for example, because it does not incorporate body fat, which is an independent predictor of ill health.36 In a similar regard, BMI is a limited measure of obesity when comparing people of different races. Compared to white people, black/African people tend to have a higher BMI and Asian people tend to have a lower BMI at similar levels of body fat.37 ,38 We note that we have also experimented with using a BMI cut-off of 27 in the Asian population, but this did not change our findings. In addition, education may measure different things in different countries and may have greater or lesser importance depending on the country in context. Note that we use a ridit-score for education, which is relative to each country; hence, the likelihood of this problem is reduced in our study. However, the pathways in which these mechanisms work are likely to be complex and vary by factors specific to each country, such as culture, which may interfere with the associations described. Nevertheless, this study contributes to the literature by including both men and women across a wide range of countries using individual-level data. This allows us to provide a comprehensive description of the complex interactions between individual-level variables and macroeconomic factors associated with obesity.
To conclude, this study provides evidence of an association between educational inequalities in obesity and economic development. Obesity increases significantly with GDP among those with lower levels of education, but this association was absent among those with higher levels of education. Hence, public health systems in emerging economies should take this into account when developing strategies for reducing obesity.
What is already known on this subject
Findings from previous studies suggest that socioeconomic status and obesity are positively associated in low-development countries, and negatively associated in high-development countries. However, some studies dispute this trend.
Most previous studies were conducted with data consisting of women from low and middle developing countries. Only one study used a data set with low-development, middle-development and high-development countries with adult men and women. Hence, little attention has been devoted to any potential difference by gender in this pattern.
What this study adds
Our study is the largest of its kind using recent data sources. Our findings support the reversal hypothesis and identify the spread of inequality in obesity across world regions. The inclusion of the large number of low-development, middle-development and high-development countries allows us to illustrate the trend and identify shifts.
Our illustrations display different patterns in men and women. In men, there is a positive association between obesity and gross domestic product (GDP) in all educational groups, though it is not significant in those with tertiary education. In women, we found a significant association between obesity and education in those with no education only. However, the positive association between obesity and GDP in those with no education was more pronounced in women than in men. Our use of indices of inequality demonstrates a reversal in absolute and relative inequality in obesity by GDP.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Contributors JMK, BHS, SEV and VS were each involved in the literature search, study design, data interpretation and writing. JMK conducted the statistical analysis and prepared the first draft. All authors read and approved the final draft.
Competing interests None declared.
Ethics approval We used cross-sectional survey data sets available if permission was given by the owners of the data sets.
Provenance and peer review Not commissioned; externally peer reviewed.