The Relationship of Socioeconomic Status with Body Mass Index Depends on the Socioeconomic Measure Used
Funding agencies: TBG received support from Harvard University through the Lown Scholars Program.
Disclosure: The authors declared no conflict of interest.
Abstract
Objective
The association between socioeconomic status (SES) and body mass index (BMI) in middle-income countries is mixed. Heterogeneity in SES indicators used could explain some differences. This study aimed to identify SES indicators consistently associated with BMI in Mexican adults in 2006, 2012, and 2016.
Methods
Data were obtained from the Mexican National Health and Nutrition Surveys of 2006, 2012, and 2016, including adults 20 to 59 years old. Given expected differences by sex, sex-stratified linear regression models were fitted for each survey. Age-adjusted and multivariate models were fit by using seven noncollinear SES indicators.
Results
In age-adjusted models, most SES indicators were associated with a higher BMI in men; mixed associations were found for women. In multivariate models, living in urban areas was associated with a higher BMI for both men and women in 2006 and 2012. Across all surveys, education was associated with a lower BMI in women, while household assets were associated with a higher BMI in men.
Conclusions
The association between SES indicators and BMI is complex. Differences by sex need to be explicitly recognized when modeling this association. Approaches that rely on a single indicator could be confounded by other SES indicators. Adjusted models show the specific SES attributes that may influence BMI.
Introduction
Lower socioeconomic status (SES) is generally associated with a higher body mass index (BMI) (1). However, this association is complex and varies by sex and country-level income (2, 3). The sources of this heterogeneity and the specific mechanisms underlying the association between SES and BMI are largely unknown (4). Identifying which SES factors can be intervened upon to reduce BMI is a critical task but one that requires a different approach to SES analysis.
Most studies operationalize SES as an index (5, 6) or rely on a single indicator (7, 8), limiting our ability to explore the specific mechanisms that underlie its association with BMI. Indices obscure the direction and magnitude of the association for each individual indicator with the outcome (9), making them more suitable for covariate control than for main exposure analysis (10). Single indicators ignore alternative pathways (4, 11) and neglect other mechanisms that may confound the association (12). Consequently, calls have been made to quantify the BMI and SES association by using multiple indicators (4, 11), which, when correctly specified and theoretically ground, allows for simultaneous adjustment and quantitation of specific effects (12).
We aimed to identify the association of seven SES indicators with BMI for women and men in a representative sample of Mexican adults, analyzing the consistency of associations in the following three points in time: 2006, 2012, and 2016.
Methods
Participants
Data were obtained from the 2006, 2012, and 2016 Mexican National Health and Nutrition Surveys (ENSANUT for its Spanish acronym). ENSANUT is a cross-sectional, probabilistic survey, representative at the national, regional, state, and urban/rural levels. Each participant has a sampling weight, representing the inverse probability of being sampled, considering the nonresponse rate (13, 14). ENSANUT began in 2006 and is carried out every 6 years. However, in 2016, a midterm ENSANUT was implemented to timely inform policies related to obesity and diabetes. ENSANUT 2016 had a smaller sample size, which limits its representativeness mainly at the state level (15). ENSANUT 2006, 2012 and 2016 were approved by the Research Ethics Committee of the National Institute of Public Health of Mexico (INSP). We used the three survey waves of ENSANUT to have comparable information and to explore the consistency of associations over time. We included adults aged 20 to 59 years with complete data on BMI, sex, area of residence, household assets, beneficiary of a cash-transfer program, health insurance coverage, education level, employment status, and ethnicity. Pregnant women (417 in 2006, 109 in 2012, and 110 in 2016) and extreme values of BMI (344 in 2006, 90 in 2012, and 26 in 2016) were excluded from the analysis. The final sample included 26,735 (97%), 30,102 (99%), and 6,300 (96%) participants from the 2006, 2012, and 2016 surveys, respectively.
BMI
BMI (weight in kilograms/height in meters squared) was calculated from weight and height. Weight and height were directly measured by trained teams during home visits following standardized procedures (13, 14). Per ENSANUT guidelines, BMI was considered to be valid within the 10 and 59 kg/m2 range and height between 1.3 and 2.0 meters.
SES indicators
Area of residence
Area of residence provides an estimate of social and economic circumstances; it is less volatile than income and less static than education (9). Although area of residence has been analyzed as an individual condition, recent approaches for developing countries consider urban and rural as a reflection of SES differences (16). We categorized the area of residence as rural (less than 2,500 inhabitants), urban (2,500 to 99,999 inhabitants), or metropolitan (100,000 inhabitants or more).
Household asset index
Household assets are used in social studies as measures of an individual's wealth. Assets indicate the household's ability to face emergencies and economic shocks (9) and measure cumulative wealth over time (17). We constructed the index from the following self-reported variables: (1) ownership of durable assets (house, properties, car, motorcycle, television, radio, sound system, blender, refrigerator, stove, washing machine, water boiler, computer, microwave, phone, and air conditioner), (2) housing characteristics (wall, roof, and floor materials, kitchen as separate room, type of fuel used for cooking, and crowding), and (3) basic amenities (running water, toilet, and drain system). All these variables were introduced into a polychoric principal component analysis (PCA) with varimax rotation (18). We used PCA for item reduction and to assign weights to household variables according to their relevance for the overall household asset construct. A single component (eigenvalue > 1) explaining 73% of the variance in 2006 and 2012 and 60% in 2016 was extracted. Variables with a high percentage of unique variance (≥ 0.90) or factor loadings < 0.40 (owning a house, other properties, a motorcycle, or a radio) were removed from the analysis following standard practice (19, 20). The household asset index was then calculated for each individual by summing up the rotated PCA weights across all items. The rotated factor loadings and their uniqueness are presented in Supporting Information Table S1. The index was then categorized with respect to 2006 into low, middle, and high.
Cash-transfer program
Participants indicated whether they were part of the program “Oportunidades,” now called “Prospera.” This is a governmental conditional cash-transfer program implemented in 1997 to reduce inequities by improving well-being, health, education, and nutrition (21). The program has a national coverage and identifies households living in poverty. The beneficiaries are given cash in exchange for school attendance, periodic health visits, and compliance with nutrition guidelines. We categorized respondents into beneficiaries or not beneficiaries.
Health insurance coverage
Participants were asked if they were affiliated or covered by any health insurance. Answers were categorized into three groups: (1) uninsured, (2) covered with the “Seguro Popular” program, and (3) covered with private or social insurance (e.g., Mexican Institute of Social Security, Institute for Social Security and Services for State Workers, Mexican Petroleum, Secretariat of the Navy, and Secretariat of National Defense). Private insurance was combined with social insurance because the proportion of private insurance in Mexico is very low (0.44% in 2012) (22). In 2016, health care insurance was exclusively asked to the head of the family; thus, we assumed that other family members were covered by the same health insurance.
Education attainment
Participants were asked what was the highest degree or level of school they had completed. Responses were categorized into the following four groups: elementary school or less (≤ 6 years), middle school (7-9 years), high school (10-12 years), and college or more (> 12 years).
Employment status
Participants were asked if in the past week they had worked at least 1 hour. Those who answered “yes” or “no, but I do have a job” were considered employed and unemployed otherwise. Students, housekeepers, those who were looking for a job, retired adults, or others were classified as unemployed.
Ethnicity
Participants were classified as indigenous if they self-identified as such and as nonindigenous otherwise.
Covariates
Self-reported age in years was included as a continuous variable. Self-reported sex was collected for stratification purposes. Other potential obesity confounders, such as physical activity, diet, or smoking, were not considered in the analysis because they have been described as potential mediators through which socioeconomic factors influence obesity (23). Adjusting for them would eliminate some of the mechanisms through which social stratification influences body weight, leading to an underestimation of the net effect.
Statistical analysis
First, the mean of continuous variables and the frequency of categorical variables, as well as their respective 95% confidence intervals (CI), were computed. Given expected differences by sex on the association between SES and BMI, we fitted sex-stratified linear regression models separately for 2006, 2012, and 2016, with BMI as a dependent variable and the SES indicators as independent variables. Regression models were first adjusted for age (as a continuous variable in years) and then adjusted for all SES indicators in addition to age. There were no multicollinearity problems detected (variation inflation factor < 5) (24). A correlation matrix between SES variables for each survey year was calculated (see Supporting Information Table S2). Correlations were moderate (≤ 0.55) in all survey years. All analyses considered the sampling design and survey sampling weights by using the SVY suite in Stata (StataCorp LLC, College Station, Texas).
Results
Sample characteristics are shown in Table 1. In 2006, the mean BMI was 27.3 kg/m2 and 28.4 kg/m2 for men and women, respectively. BMI increased 0.4 kg/m2 in 2012 and 0.5 kg/m2 in 2016 for both sexes with respect to 2006. From 2006 to 2016, the percent of uninsured decreased from 51.9% to 13.1% for men and from 50.6% to 13.7% for women, while “Seguro Popular” coverage increased from 8.1% in men and 10.7% in women to 42.0% and 44.7%, respectively. In 2006, 38.8% of men and 47.9% of women had completed elementary school or less, but the gap reduced in 2016 to 25.2% and 26.8%, respectively. Only 14.5% of men were unemployed in 2016, but the percent was nearly four times greater in women (55.0%).
| Men | Women | |||||
|---|---|---|---|---|---|---|
| 2006, | 2012, | 2016, | 2006, | 2012, | 2016, | |
| n = 10,509, N = 19,295,427 | n = 12,542, N = 27,040,989 | n = 2,049, N = 27,348,027 | n = 16,226, N = 27,568,523 | n = 17,560, N = 29,523,737 | n = 4,251, N = 29,271,746 | |
| Mean or n (95% CI) | Mean or n (95% CI) | Mean or n (95% CI) | Mean or n (95% CI) | Mean or n (95% CI) | Mean or n (95% CI) | |
| BMI (kg/m2) | 27.3 (27.2-27.5) | 27.7 (27.6-27.9) | 27.8 (27.3-28.4) | 28.4 (28.3-28.6) | 28.8 (28.7-29.0) | 28.9 (28.6-29.3) |
| Area of residence (%) | ||||||
| Rural | 19.0 (17.3-20.8) | 21.1 (20.2-22.1) | 23.3 (20.2-26.9) | 20.6 (18.9-22.5) | 20.9 (20.1-21.7) | 22.2 (19.4-25.3) |
| Urban | 23.3 (21.4-25.3) | 18.5 (17.5-19.5) | 14.0 (10.2-18.8) | 23.6 (21.8-25.5) | 18.8 (17.9-19.6) | 15.1 (11.0-20.3) |
| Metropolitan | 57.7 (55.1-60.3) | 60.4 (59.0-61.7) | 62.7 (56.9-68.1) | 55.8 (53.3-58.3) | 60.3 (59.1-61.5) | 62.7 (57.2-67.9) |
| Household assets (%) | ||||||
| Low | 31.9 (29.9-34.0) | 34.8 (33.4-36.2) | 30.3 (26.0-34.9) | 34.7 (32.7-36.7) | 35.4 (33.9-36.9) | 28.9 (25.8-32.3) |
| Middle | 33.8 (31.9-35.7) | 36.5 (34.9-38.1) | 29.0 (24.6-33.9) | 33.6 (32.0-35.2) | 35.9 (34.5-37.2) | 30.3 (27.0-33.8) |
| High | 34.3 (31.9-36.8) | 28.7 (27.1-30.5) | 40.7 (35.5-46.2) | 31.7 (29.9-33.7) | 28.8 (27.2-30.4) | 40.7 (36.4-45.2) |
| Cash-transfer program (%) | ||||||
| Not beneficiary | 84.4 (82.9-85.8) | 79.6 (78.5-80.7) | 78.1 (74.5-81.3) | 81.5 (79.9-83.1) | 79.4 (78.4-80.4) | 77.2 (74.5-79.7) |
| Beneficiary | 15.6 (14.2-17.1) | 20.4 (19.3-21.5) | 21.9 (18.7-25.5) | 18.5 (16.9-20.1) | 20.6 (19.6-21.6) | 22.8 (20.3-25.5) |
| Health insurance coverage (%) | ||||||
| Uninsured | 51.9 (50.0-53.8) | 29.4 (28.1-30.7) | 13.1 (10.2-16.7) | 50.6 (48.8-52.3) | 22.2 (21.1-23.4) | 13.7 (11.7-16.0) |
| “Seguro Popular” | 8.1 (7.3-9.1) | 31.6 (30.2-33.0) | 42.0 (37.5-46.5) | 10.7 (9.6-12.0) | 39.3 (38.0-40.6) | 44.7 (40.8-48.7) |
| Private or social insurance | 40.0 (38.1-41.9) | 39.0 (37.5-40.6) | 44.9 (40.4-49.5) | 38.7 (37.0-40.4) | 38.5 (37.1-39.9) | 41.6 (38.0-45.3) |
| Education level (%) | ||||||
| Elementary school or less | 38.8 (36.8-40.8) | 30.2 (28.8-31.6) | 25.2 (21.8-28.8) | 47.9 (46.2-49.5) | 35.2 (33.9-36.6) | 26.8 (23.9-29.8) |
| Middle school | 27.7 (26.4-29.2) | 32.0 (30.6-33.5) | 32.0 (27.9-36.4) | 24.3 (23.0-25.6) | 29.2 (28.0-30.4) | 35.6 (32.0-39.4) |
| High school | 17.6 (16.1-19.1) | 19.2 (18.1-20.4) | 22.1 (18.6-26.2) | 15.9 (14.9-17.0) | 18.9 (17.9-19.9) | 18.1 (16.1-20.2) |
| College or more | 15.9 (14.3-17.7) | 18.5 (17.2-19.9) | 20.7 (16.8-25.2) | 11.9 (10.7-13.2) | 16.7 (15.6-17.8) | 19.5 (15.4-24.5) |
| Employment status (%) | ||||||
| Unemployed | 15.7 (14.3-17.2) | 15.2 (14.1-16.3) | 14.5 (10.9-19.0) | 68.0 (66.7-69.3) | 61.3 (60.0-62.5) | 55.0 (51.4-58.5) |
| Employed | 84.3 (82.8-85.7) | 84.8 (83.7-85.9) | 85.5 (81.0-89.1) | 32.0 (30.7-33.3) | 38.7 (37.5-40.0) | 45.0 (41.5-48.6) |
| Ethnicity (%) | ||||||
| Nonindigenous | 82.6 (81.1-84.0) | 78.8 (77.3-80.2) | 75.1 (71.6-78.3) | 81.0 (79.5-82.5) | 78.6 (77.2-79.9) | 71.0 (68.0-73.9) |
| Indigenous | 17.4 (16.0-18.9) | 21.2 (19.8-22.7) | 24.9 (21.7-28.4) | 19.0 (17.5-20.5) | 21.4 (20.1-22.8) | 29.0 (26.1-32.0) |
| Age (y) | 37.5 (37.1-37.9) | 37.0 (36.7-37.3) | 36.3 (35.5-37.1) | 37.2 (36.9-37.4) | 37.4 (37.2-37.7) | 37.0 (36.4-37.6) |
Table 2 presents age and multivariate models for BMI and SES for men. In 2006, most SES indicators, except employment and being covered by “Seguro Popular” insurance, were associated with BMI after age adjustment. In the multivariate model, only area of residence (urban 0.53 kg/m2 higher than rural) and household assets (1.07 and 0.95 kg/m2 higher for middle/high than low assets) remained associated with BMI. Results were similar in the multivariate model for 2012; a higher BMI was observed in people living in urban and metropolitan areas (urban 1.29 and metropolitan 0.62 kg/m2 higher BMI than rural) and among those with middle/high household assets (0.41 middle and 0.90 kg/m2 high compared with low assets). Employed people had higher BMI (0.79 kg/m2) than unemployed but only in 2012. Only high school education, and only in 2012, was associated with 0.5 kg/m2 higher BMI than primary school or less. In 2016, area of residence became unassociated with BMI, and only the high tertile of household assets was associated with higher BMI (1.25 kg/m2 higher BMI than the low-assets group).
| 2006 | 2012 | 2016 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Age-adjusted | Multivariate | Age-adjusted | Multivariate | Age-adjusted | Multivariate | |||||||
| β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | |
| Area of residence | ||||||||||||
| Rural | REF | REF | REF | REF | REF | REF | ||||||
| Urban | 0.99 (0.60 to 1.39) | < 0.001 | 0.53 (0.08 to 0.98) | 0.021 | 1.55 (1.21 to 1.90) | < 0.001 | 1.29 (0.93 to 1.64) | < 0.001 | 1.04 (0.00 to 2.08) | 0.050 | 0.48 (−0.54 to 1.51) | 0.354 |
| Metropolitan | 0.96 (0.58 to 1.35) | < 0.001 | 0.22 (−0.25 to 0.69) | 0.359 | 1.20 (0.91 to 1.50) | < 0.001 | 0.62 (0.29 to 0.95) | < 0.001 | 0.33 (−0.58 to 1.25) | 0.472 | −0.61 (−1.50 to 0.29) | 0.182 |
| Household assets | ||||||||||||
| Low | REF | REF | REF | REF | REF | REF | ||||||
| Middle | 1.24 (0.92 to 1.55) | < 0.001 | 1.07 (0.72 to 1.41) | < 0.001 | 0.74 (0.44 to 1.04) | < 0.001 | 0.41 (0.08 to 0.74) | 0.015 | 0.63 (−0.49 to 1.75) | 0.271 | 0.66 (−0.21 to 1.54) | 0.138 |
| High | 1.14 (0.77 to 1.51) | < 0.001 | 0.95 (0.54 to 1.35) | < 0.001 | 1.33 (0.95 to 1.70) | < 0.001 | 0.90 (0.41 to 1.38) | < 0.001 | 1.47 (0.44 to 2.51) | 0.005 | 1.25 (0.28 to 2.21) | 0.012 |
| Cash-transfer program | ||||||||||||
| Not beneficiary | REF | REF | REF | REF | REF | REF | ||||||
| Beneficiary | −0.94 (−1.27 to −0.62) | < 0.001 | −0.07 (−0.48 to 0.34) | 0.738 | −0.93 (−1.25 to −0.61) | < 0.001 | −0.27 (−0.66 to 0.13) | 0.185 | −0.45 (−1.38 to 0.48) | 0.346 | 0.09 (−0.82 to 1.00) | 0.845 |
| Health insurance coverage | ||||||||||||
| Uninsured | REF | REF | REF | REF | REF | REF | ||||||
| “Seguro Popular” | −0.25 (−0.63 to 0.14) | 0.207 | 0.14 (−0.24 to 0.51) | 0.481 | −0.03 (−0.36 to 0.3) | 0.860 | 0.32 (−0.03 to 0.67) | 0.069 | 0.48 (−1.60 to 2.55) | 0.650 | 0.60 (−1.28 to 2.47) | 0.530 |
| Private or social insurance | 0.55 (0.21 to 0.88) | 0.001 | 0.27 (−0.06 to 0.60) | 0.110 | 0.61 (0.27 to 0.95) | < 0.001 | 0.26 (−0.08 to 0.61) | 0.138 | 1.51 (−0.64 to 3.66) | 0.168 | 1.19 (−0.79 to 3.17) | 0.237 |
| Education level | ||||||||||||
| Elementary school or less | REF | REF | REF | REF | REF | REF | ||||||
| Middle school | 0.54 (0.19 to 0.90) | 0.003 | 0.17 (−0.20 to 0.54) | 0.373 | 0.65 (0.29 to 1.01) | < 0.001 | 0.24 (−0.13 to 0.61) | 0.205 | 0.16 (−1.03 to 1.35) | 0.794 | −0.04 (−1.16 to 1.08) | 0.942 |
| High school | 0.70 (0.20 to 1.19) | 0.006 | 0.15 (−0.40 to 0.69) | 0.599 | 1.05 (0.63 to 1.46) | < 0.001 | 0.50 (0.07 to 0.93) | 0.023 | 1.40 (0.40 to 2.40) | 0.006 | 1.02 (−0.08 to 2.13) | 0.068 |
| College or more | 0.64 (0.04 to 1.24) | 0.038 | 0.00 (−0.61 to 0.62) | 0.995 | 1.09 (0.69 to 1.50) | < 0.001 | 0.38 (−0.11 to 0.87) | 0.124 | 1.39 (−0.3 to 3.07) | 0.106 | 0.74 (−0.76 to 2.24) | 0.331 |
| Employment status | ||||||||||||
| Unemployed | REF | REF | REF | REF | REF | REF | ||||||
| Employed | 0.20 (−0.28 to 0.69) | 0.413 | 0.11 (−0.36 to 0.58) | 0.636 | 0.77 (0.39 to 1.15) | < 0.001 | 0.79 (0.41 to 1.18) | < 0.001 | 0.24 (−1.01 to 1.5) | 0.705 | 0.63 (−0.53 to 1.79) | 0.285 |
| Ethnicity | ||||||||||||
| Nonindigenous | REF | REF | REF | REF | REF | REF | ||||||
| Indigenous | −0.39 (−0.75 to −0.03) | 0.035 | −0.01 (−0.37 to 0.35) | 0.948 | 0.37 (0.03 to 0.72) | 0.033 | 0.09 (−0.29 to 0.46) | 0.646 | −0.11 (−1.04 to 0.81) | 0.809 | 0.18 (−0.7 to 1.07) | 0.684 |
- Age-adjusted column presents estimates of models for each SES indicator adjusted by age; multivariate models present estimates of models including all SES indicators and age.
Results for women are in Table 3. In 2006, most SES indicators were associated with BMI in the age and multivariate models. In the multivariate model in 2012, being a beneficiary of the cash-transfer program and being covered by “Seguro Popular” were no longer associated with BMI, while household assets were inconsistently associated with BMI. In 2016, area of residence was no longer associated with BMI. In the multivariate model, being a beneficiary of the cash-transfer program became, again, associated with lower BMI in comparison with nonbeneficiaries, and having a high school education was associated with lower BMI in comparison with elementary school. Across all survey waves, in the multivariate model, women with a higher education had a lower BMI compared with women with elementary education or less (−1.20, −1.02, and −1.18 kg/m2 for women with high school and −1.96, −1.75, and −0.76 kg/m2 for women with college in 2006, 2012, and 2016, respectively).
| 2006 | 2012 | 2016 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Age-adjusted | Multivariate | Age-adjusted | Multivariate | Age-adjusted | Multivariate | |||||||
| β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | β (95% CI) | P value | |
| Area of residence | ||||||||||||
| Rural | REF | REF | REF | REF | REF | REF | ||||||
| Urban | 0.76 (0.39 to 1.13) | < 0.001 | 0.50 (0.08 to 0.92) | 0.019 | 0.86 (0.49 to 1.23) | < 0.001 | 0.94 (0.56 to 1.33) | < 0.001 | 0.09 (−0.60 to 0.78) | 0.804 | 0.18 (−0.51 to 0.88) | 0.605 |
| Metropolitan | 0.58 (0.23 to 0.94) | 0.001 | 0.37 (−0.06 to 0.80) | 0.088 | 0.47 (0.17 to 0.77) | 0.003 | 0.66 (0.30 to 1.01) | < 0.001 | 0.33 (−0.35 to 1.00) | 0.341 | 0.36 (−0.28 to 1.01) | 0.269 |
| Household assets | ||||||||||||
| Low | REF | REF | REF | REF | REF | REF | ||||||
| Middle | 0.79 (0.50 to 1.07) | < 0.001 | 0.64 (0.34 to 0.94) | < 0.001 | 0.32 (0.00 to 0.64) | 0.051 | 0.35 (0.00 to 0.71) | 0.050 | 0.67 (−0.14 to 1.49) | 0.106 | 0.44 (−0.39 to 1.28) | 0.296 |
| High | 0.03 (−0.35 to 0.40) | 0.887 | 0.27 (−0.14 to 0.69) | 0.199 | −0.31 (−0.67 to 0.05) | 0.095 | 0.11 (−0.34 to 0.55) | 0.644 | −0.42 (−1.30 to 0.45) | 0.342 | −0.52 (−1.55 to 0.50) | 0.316 |
| Cash-transfer program | ||||||||||||
| Not beneficiary | REF | REF | REF | REF | REF | REF | ||||||
| Beneficiary | −0.79 (−1.10 to −0.47) | < 0.001 | −0.74 (−1.14 to −0.34) | < 0.001 | −0.16 (−0.44 to 0.12) | 0.273 | −0.13 (−0.46 to 0.2) | 0.440 | −0.47 (−1.09 to 0.15) | 0.135 | −0.69 (−1.33 to −0.05) | 0.035 |
| Health insurance coverage | ||||||||||||
| Uninsured | REF | REF | REF | REF | REF | REF | ||||||
| “Seguro Popular” | 0.30 (−0.08 to 0.68) | 0.127 | 0.53 (0.15 to 0.90) | 0.006 | −0.10 (−0.50 to 0.31) | 0.647 | −0.11 (−0.53 to 0.3) | 0.590 | −0.15 (−1.62 to 1.32) | 0.843 | −0.06 (−1.41 to 1.30) | 0.934 |
| Private or social insurance | 0.11 (−0.21 to 0.42) | 0.505 | 0.20 (−0.13 to 0.52) | 0.240 | −0.25 (−0.66 to 0.16) | 0.230 | −0.06 (−0.5 to 0.37) | 0.773 | −0.06 (−1.53 to 1.41) | 0.934 | 0.11 (−1.34 to 1.56) | 0.882 |
| Education level | ||||||||||||
| Elementary school or less | REF | REF | REF | REF | REF | REF | ||||||
| Middle school | 0.10 (−0.26 to 0.46) | 0.579 | −0.23 (−0.60 to 0.13) | 0.215 | 0.04 (−0.28 to 0.36) | 0.812 | −0.15 (−0.47 to 0.18) | 0.380 | 0.39 (−0.50 to 1.27) | 0.388 | 0.27 (−0.60 to 1.14) | 0.539 |
| High school | −0.71 (−1.09 to −0.32) | < 0.001 | −1.20 (−1.62 to −0.78) | < 0.001 | −0.73 (−1.12 to −0.34) | < 0.001 | −1.02 (−1.44 to −0.60) | < 0.001 | −1.14 (−2.07 to −0.21) | 0.016 | −1.18 (−2.20 to −0.15) | 0.024 |
| College or more | −1.44 (−2.01 to −0.86) | < 0.001 | −1.96 (−2.56 to −1.36) | < 0.001 | −1.41 (−1.87 to −0.94) | < 0.001 | −1.75 (−2.29 to −1.22) | < 0.001 | −0.76 (−1.80 to 0.28) | 0.151 | −0.76 (−1.84 to 0.32) | 0.166 |
| Employment status | ||||||||||||
| Unemployed | REF | REF | REF | REF | REF | REF | ||||||
| Employed | 0.03 (−0.29 to 0.35) | 0.858 | 0.13 (−0.21 to 0.46) | 0.457 | 0.00 (−0.27 to 0.27) | 0.979 | 0.11 (−0.18 to 0.4) | 0.463 | −0.22 (−0.95 to 0.51) | 0.555 | −0.13 (−0.87 to 0.62) | 0.737 |
| Ethnicity | ||||||||||||
| Nonindigenous | REF | REF | REF | REF | REF | REF | ||||||
| Indigenous | −0.36 (−0.70 to −0.02) | 0.036 | −0.25 (−0.59 to 0.08) | 0.138 | 0.17 (−0.14 to 0.48) | 0.276 | 0.20 (−0.14 to 0.54) | 0.239 | −0.24 (−0.93 to 0.45) | 0.495 | −0.17 (−0.84 to 0.50) | 0.615 |
- Age-adjusted column presents estimates of models for each SES indicator adjusted by age; multivariate models present estimates of models including all SES indicators and age.
Discussion
We found that the association between SES indicators and BMI depended on the method used to measure SES. A large number of SES indicators were associated with BMI in age-adjusted models. However, in multivariate models, only some variables remained associated with important differences by sex. For men and women in 2006 and 2012, living in urban areas was associated with higher BMI; differences between urban and rural areas disappeared in 2016. For men, higher household assets were associated with higher BMI in the three survey waves; contrastingly, for women, higher education was consistently associated with lower BMI across surveys.
SES indicators consistently associated with BMI varied by sex. For women, higher education was associated with lower BMI in all years. For men, more household assets were associated with higher BMI in all years. This is consistent with other studies (25, 26). Although explaining the underlying mechanisms of how SES modifies body weight is beyond the scope of this article, differences in ideal body weight may explain the differential sex effects. The ideal body weight is socially constructed, being time and culturally dependent. Yet, prior research has shown that women tend to consider a thinner body to be ideal, while for men, body weight becomes a physical marker of dominance, and a larger body is preferred (3, 23, 27). In a recent study, high-SES men aspired to a higher body weight compared with low-SES men; in contrast, high-SES women considered a lower body weight ideal compared with low-SES women, regardless of the SES indicator (28). In that study, the association between SES and BMI was attenuated after ideal body weight was adjusted for, suggesting a strong mediating effect of ideal body weight (28). Unfortunately, the potential mediating effect of the ideal body type has been scarcely investigated, and other mechanisms that are perhaps more important could also be at play. Further research is needed to fully understand the sex-specific mechanism linking SES and obesity.
Education is the most frequently used SES indicator in obesity studies (2). We found that education and BMI were associated in men in unadjusted models; this is in line with other studies in Mexico (5) and high-income countries (2). However, this association disappeared after adjusting for other SES indicators, showing the importance of considering other SES pathways for weight gain. This is consistent with the results reported by Quezada et al. by using ENSANUT data for 2006 and 2012 (29). For women, the association between education and BMI has been previously described in Mexico (5, 7, 8, 29), but nearly all studies failed to adjust for other SES indicators. In our study, education remained associated with BMI in 2006, 2012, and 2016 after taking into account age and other SES variables, suggesting that for women, education is a consistently associated factor. Sex was also described as an effect modifier in 232 countries for the association between education and obesity (30). Potential explanations for the role of education in BMI for women include being responsible for cooking and preparing food, while men have lunch more often away from home (31).
Our study showed a consistent association between household assets and BMI only for men (28). The association for 2006 and 2012 was significant for middle and high assets; in 2016, the association held only for the high-asset group. Studies using the National Health Survey of 2000 (32) and ENSANUT of 2006 and 2012 showed similar associations for men (29), although other SES indicators, such as insurance coverage, participating in a cash-transfer program, or ethnicity, were not considered as confounders. In women, assets were not associated with BMI in any of the survey years. Our findings differ from those of a systematic review, which found no association between household assets and obesity in men but a positive association for women in middle-income countries (3). However, most included studies fail to adjust for other confounders besides age and sex, which could explain the different results obtained in the systematic review.
Urban areas in Mexico presented higher obesity for both men and women in multivariate models in comparison with rural areas in 2006 and 2012, but this association disappears in 2016. In contrast, in developed countries, obesity tends to be higher in rural areas (33). The increased obesity prevalence in 2006 and 2012 in urban areas may be explained by the nutritional transition affecting urban and metropolitan areas in Mexico. In Latin America over the past 30 years, urbanization has boomed and over 80% of the population now live in urban and metropolitan areas (34). The rural exodus and the rapid urbanization are decreasing the quality of life in urban areas and have been previously associated with an increase in obesity (35). The lack of association in 2016 suggests that differences between urban and rural areas could be decreasing; in 2016, rural areas showed a decrease in physical activity, which could have equated the rural to the urban population (15). An in-depth analysis of this finding and its potential mediators is needed.
By incorporating multiple aspects of SES, our study had some advantages to prior studies. In our study, the moderate correlations (all ≤ 0.55) between the SES variables by survey year and sex allowed us to simultaneously adjust for all individual SES indicators without creating collinearity issues in the model. Previous studies have not been able to perform simultaneous analyses of multiple SES indicators because of higher correlations. This is likely due to most studies having been conducted in high-income countries where collinearity of SES indicators seems to be much higher than in middle-income countries such as Mexico, where there is higher SES heterogeneity (e.g., a higher education level does not necessarily imply a higher income). Moreover, by using multiple SES indicators at once, our study provides further insight into the complex relationship between SES and obesity and into the relative contribution of different aspects of SES to obesity. Our approach is key to inform policy interventions, which should be multidimensional in nature.
A limitation of our study was to rely on cross-sectional data, which precluded any causal interpretation of the SES and obesity association. We did not consider income as an additional SES indicator. Household assets and household characteristics are more consistently reported in Mexico in comparison with monthly income (36). Still, in a sensitivity analysis, we added income to final models and did not observe a change in the estimates for the association of other SES indicators, and we did not find an association between BMI and income in any of the years. To test the sensitivity of our results, we (1) excluded beneficiaries of the cash-transfer program and (2) dichotomized BMI (people with vs. without obesity) by using Poisson regression models, which, given the prevalence of obesity (30%), are superior to logistic regression models (37). The conclusions of both analyses were in line with our main results. A second limitation concerns the definition we used to measure employment. Our indicator of employment measured if the person worked at least 1 hour per week but did not identify full-time or part-time jobs. Only in 2012 were working participants asked about working hours; the mean working hours during the past week for 2012 was 44.4 hours (95% CI: 43.9-44.9), suggesting that, on average, most participants worked full-time. A third limitation is related to the sample size of ENSANUT 2016. Comparisons between survey waves should be interpreted carefully, as the smaller sample size of the 2016 survey could affect the power of some associations.
Conclusion
The association between BMI and SES has been studied by using different analytical approaches that include SES indices, a single SES indicator, or multiple SES adjustment. Understanding the relative contribution of different aspects of SES to obesity is vital to inform policy interventions. If the aim of a study is to identify which SES condition is more relevant to obesity, using a single SES indicator or an index will be a limited approach. Reliance on a single indicator will ignore or be confounded by other social pathways that may be important, while the index could conflate the overall importance of SES without giving clear clues as to which mechanism is more important. Instead, we recommend using multiple SES indicators, which allows one to consider a multiplicity of factors adjusting for each other and opening new venues of understanding on the complex relationship between SES and obesity.
