p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Available online at www.sciencedirect.com
Public Health journal homepage: www.elsevier.com/puhe
Original Research
Is obesity associated with global warming? J. Squalli* Department of Economics, American University of Sharjah, P.O. Box 26666, Sharjah, United Arab Emirates
article info
abstract
Article history:
Objectives: Obesity is a national epidemic that imposes direct medical and indirect eco-
Received 13 November 2013
nomic costs on society. Recent scholarly inquiries contend that obesity also contributes to
Received in revised form
global warming. The paper investigates the relationship between greenhouse gas emis-
9 September 2014
sions and obesity.
Accepted 16 September 2014
Study design: Cross-sectional state-level data for the year 2010.
Available online 26 November 2014
Methods: Multiple regression analysis using least squares with bootstrapped standard errors and quantile regression.
Keywords:
Results: States with higher rates of obesity are associated with higher CO2 and CH4 emis-
Greenhouse gas emissions
sions (p < 0.05) and marginally associated with higher N2O emissions (p < 0.10), net of other
Obesity
factors. Reverting to the obesity rates of the year 2000 across the entire United States could
Global warming
decrease greenhouse gas emissions by about two percent, representing more than 136 million metric tons of CO2 equivalent. Conclusions: Future studies should establish clear causality between obesity and emissions by using longitudinal data while controlling for other relevant factors. They should also consider identifying means to net out the potential effects of carbon sinks, conversion of CH4 to energy, cross-state diversion, disposal, and transfer of municipal solid waste, and potentially lower energy consumption from increased sedentariness. © 2014 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.
Introduction According to OECD Health Data 2012, about one third of the US adult population is obese. Based on census and Centers for Disease Control and Prevention (CDC) data, the number of obese people in the United States amounts to approximately 63 million adults in 2010 vs 39 million in 2000. Such a drastic change has prompted the CDC to consider obesity a national epidemic that imposes a heavy burden on society. The direct medical costs associated with obesity have been estimated at $190 billion, representing more than 20% of US spending on healthcare.1 Indirect costs, on the other hand, are estimated at
about $12 billion for lost productive time from work absenteeism and presenteeism2 and around $60 billion for health insurance externality.3 Obesity is also expected to impose an environmental burden by potentially contributing to increased greenhouse gas emissions from higher fuel consumption required to transport heavier people, increased fertilizer usage for higher food production, and increased organic waste generated by larger livestock and heavier people.4 To date, the literature on the relationship between obesity and greenhouse gas emissions is very limited in scope. In fact, while a number of contributions estimate obesity to exert a substantial environmental effect, none substantiate their predictions with
* Tel.: þ971 65152318. E-mail address: [email protected]. http://dx.doi.org/10.1016/j.puhe.2014.09.008 0033-3506/© 2014 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.
1088
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 1 e Summary statistics. Variable CH4 CO2 N 2O Crop and animal farming (% of real GDP) Household size Income ($) Landfills Manufacturing (% of real GDP) Natural gas and oil (% of real GDP) Obesity Population Urbanization Transportation (% of real GDP) Transportation and utilities (% of real GDP)
Obs.
Mean
Std. Dev.
Min
Max
51 51 51 51 51 51 50 51 51 51 51 51 51 51
1,877,535 169,232.9 593,692.3 1.37 2.28 43,124.67 70.72 11.56 1.68 26.99 6,065,298 74.09 3.18 5.04
2,077,286 170,142.1 950,085.9 1.88 0.17 16,997.91 102.18 5.59 4.31 3.23 6,839,909 14.88 1.43 1.53
385 3869.35 895 0 1.83 28,596 3 0.23 0 21 564,554 38.66 0.28 1.55
9,759,024 1,024,036 5,377,550 8.34 2.83 148,605 678 27.48 20.92 34 37,338,198 100 9.44 10.70
Manufacturing and Utilities represent output as a percent of total state GDP in chained 2005 dollars. CO2 emissions are measured in thousand metric tons, whereas CH4 and N2O are measured in metric tons of CO2 equivalent.
econometric estimations that control for the various factors that can contribute to emissions. In other words, the evidence presented in the relevant literature is largely speculative. For example, a study focussing on the European Union and OECD countries uses crude estimates of car, plane, and rail passenger weight to estimate the impact of a 5 kg increase in weight on CO2 emissions.4 It estimates the weight increase to raise CO2 emissions in Germany by 0.7 million metric tons. By extrapolation, the weight increase is also estimated to raise emissions in the European Union by 3.4 million metric tons and by 10.2 million metric tons in OECD countries.4 Another study estimates obesity's contribution to fuel consumption in the United States to raise emissions by more than nine million metric tons of CO2 equivalent annually.5 Similarly, obesity's contribution to food production and car travel in a population of one billion people is estimated to potentially increase greenhouse gas emissions by up to one billion metric tons of CO2 equivalent annually.6 Finally, more recently, based on sample of 25 diabetic and obese individuals in Scotland, a 10 kg increase in body weight is estimated to raise world CO2 emissions by 49.56 million metric tons.7 This paper is a first attempt at assessing the statistical association between emissions and obesity using US statelevel data. The ensuing empirical findings are expected to have important policy implications to better inform policymakers and the public. Understanding the relationship between obesity and the environment ought to guide policymakers in identifying proper corrective measures, whenever applicable. To this end, the paper is organized as follows: Section 2 describes the data and methodology. Section 3 summarizes the estimation results and assesses the potential impact of a reduction in obesity on greenhouse gas emissions. Section 4 discusses the results and concludes.
Methods Data The variables used and their sources are described below. Summary statistics are shown in Table 1.
Greenhouse gas emissions data are from the Environmental Protection Agency (EPA) and include emissions of Methane (CH4), Carbon Dioxide (CO2), and Nitrous Oxide (N2O), measured in metric tons of CO2 equivalent. The data are at the US state level for the year 2010. The EPA implemented in 2010 a mandatory reporting rule for industrial facilities that emit over 25,000 metric tons of CO2 equivalent. While this reporting rule does not include all emitters, it does shed light on emissions at the state level. For CO2 emissions in particular, given the fact that, according to the EPA, emissions from fossil fuel combustion represent about 78% of total emissions, the EPA's data set are combined for emissions from large facilities with emissions from fossil fuel combustion. Data for CH4 and N2O, on the other hand, represent emissions from large facilities. EPA data are also used for measuring the number of municipal solid waste landfills to control for methane generated by waste. To the author's knowledge, the only data available date back to 1995. Population-related data include the population size, which are for the US census of 2010 from the US Census Bureau. Other variables include a measure of urbanization, which is from the US census of 2010 and represents the share of the population in urbanized areas (with more than one million people). The average household size is also used, which is computed using data from the 2010 US census and represents the weighted average household size in rented and owned units. Macroeconomic data include real GDP per capita, which are from the Bureau of Economic Analysis for 2010, in chained 2005 dollars. Real GDP per capita is used as a proxy for affluence. Also the sectoral output data is used, as a percent of real GDP, which are from the Bureau of Economic Analysis for 2010. The sectors included are Crop and Animal Farming, Manufacturing, Natural Gas and Oil, Transportation and Utilities, and Transportation and are used to control for the primary sources of CH4, CO2, and N2O emissions. This paper's the variable of interest represents the obesity rate, which is measured using the percentage of US adults with a body mass index exceeding 30. The data are from the Behavioural Risk Factor Surveillance Systems of the CDC for 2010. Fig. 1 provides preliminary insight on the relationship between greenhouse gas emissions and obesity. Overall, the
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
1089
Fig. 1 e Obesity rate and greenhouse gas emissions.
fitted values seem to capture a positive relationship between the two variables (despite DC being a potential outlier). The next section investigates this relationship further.
ln N2 Oi ¼ b0 þ b1 ln POPi þ b2 ln INCOMEi þ b3 lnðINCOMEi Þ2 þ b4 FARMINGi þ b5 MANUFi þ b6 TRANSi þ b7 OBESITYi þ ti (3)
Methodology The specifications for each of the three primary greenhouse gas emissions are estimated as follows: ln CO2i ¼ a0 þ a1 ln POPi þ a2 ln INCOMEi þ a3 lnðINCOMEi Þ2 þ a4 MANUFi þ a5 TRANSUTILi þ a6 ln HHOLDi þ a7 URBANi þ a8 OBESITYi þ 3 i (1) where CO2 emissions for state i are estimated with respect to population, income, real manufacturing output as a percent of state GDP (MANUF), real transportation and utilities output as a percent of state GDP (TRANSUTIL), average state household size (HHOLD), the urbanization rate (URBAN), and the obesity rate. The variables MANUF and TRANSUTIL are used to control for the primary sources of CO2 emissions, namely industrial processes, electricity production, and fossil fuel combustion. Furthermore, to control for the potential impact of populationrelated factors on emissions, the variables HHOLD and URBAN are used to account for scale economies in energy consumption and for relatively higher energy consumption in urban areas, respectively.9 As for CH4 and N2O emissions, the following specifications are estimated: ln CH4i ¼ g0 þ g1 ln POPi þ g2 ln INCOMEi þ g3 lnðINCOMEi Þ2 þ g4 FARMINGi þ g5 ln LANDFILLi þ g6 OILi þ g7 OBESITYi þ mi (2)
where FARMING represents the real output of crop and animal farming as a percent of real state GDP, LANDFILL represents the number of landfills, and OIL represents the real output of natural gas and oil as a percent of real state GDP. The variables FARMING, LANDFILL, and OIL are used in Equation (2) to control for the primary sources of CH4 emissions, namely livestock farming, municipal solid waste from homes and businesses, and the production, processing, transportation, storage, and distribution of natural gas and petroleum, respectively. It is important to note that it is not necessary to scale the LANDFILL variable since population, a typical scale variable, is included as an explanatory variable. As for Equation (3), FARMING, MANUF, and TRANS are used to control for the primary sources of N2O emissions, namely agricultural soil management (i.e. the use of nitrogen-based fertilizers), industrial or chemical production, and transportation, amongst others. In order to determine the proper estimation procedure, first, a Least Squares estimations of all three specifications are completed. Second, leverage values for each observation are derived and plotted against their corresponding normalized squared residuals. A leverage represents the diagonal component of the hat matrix, which is bounded by the limits 1/n and 1. Observations with high leverage values are likely to be outliers. As Fig. 2 shows, most data points have leverage values that do not exceed 0.6. However, for CO2 and N2O estimations, DC has a leverage value close to unity, suggesting that estimation results may be influenced by DC data. To
1090
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
assess how DC data points affect the estimation results, there is need to exclude DC from the data set and complete Least Squares estimations for CO2 and N2O emissions. The estimation results show no discernible difference warranting the exclusion of DC. After completing Least Squares estimations for the three greenhouse gases, there is no evidence of heteroskedasticity from a visual inspection of plots of the residuals against the fitted values. Applying the CookeWeisberg test for heteroskedasticity, the author fails to reject the null hypothesis of constant variance for CH4 and N2O estimations but not for CO2 estimations. Finally, based on the ShapiroeWilk test, he fails to reject the null hypothesis that residuals are normally distributed. Although these tests may suggest limiting estimations to Ordinary Least Squares the econometric specifications are estimated further using: (a) a Least Squares procedure with bootstrapped standard errors and (b) quantile regression. The benefit of the first procedure lies in its ability to derive estimates of standard errors and confidence intervals based on the underlying distribution of the sample. Given the complex nature of the relationship between greenhouse gas emissions and its determinants and the small sample size, bootstrapping is distribution-independent and can allay concerns about within-sample distortions. Bootstrapping estimations are completed using 500 replications as a means to minimize potential random sampling errors. Furthermore, this procedure is less sensitive to outliers and heteroskedasticity and involves estimating the median (rather than the mean) of the dependent variable by minimizing the sum of absolute residuals. For robustness, both procedures are used, but interpretations focus on quantile regression due to the latter's ability to control for outliers and hetereoskedasticity (especially for CO2 emissions).
Results Columns (1) and (2) in Table 2 report Least Squares estimations with bootstrapped standard errors and quantile regression, respectively. Estimations are also completed using quantile regression with bootstrapped standard errors, which yield very similar results to those reported by the standard quantile regression. Overall, the coefficient estimates for population are close to unity and statistically significant at the 0.01 level across both estimation procedures and for all three greenhouse gases. For CH4 estimations, states with larger farming and natural gas and oil sectors are associated with higher emissions, statistically significant at the 0.10 and 0.05 levels, respectively. As for the other two greenhouse gases, states with larger transportation and utilities sector are associated with higher CO2 emissions, statistically significant at the 0.01 level, whereas states with a larger farming sector are associated with higher N2O emissions, statistically significant at the 0.05 level. On the other hand, the coefficient estimate for obesity is positive and statistically significant across both estimation procedures and for CO2 and CH4 at the 0.05 level and for N2O at the 0.10 level. While the fact that the two estimation procedures yield similar results is undoubtedly a good sign, the absence of statistical significance for a number of included explanatory variables could be due to potential collinearity. The Variance Inflation Factors (VIFs) test can be used to assess the severity of collinearity among variables. VIFs are computed by estimating each variable with respect to other explanatory variables. The corresponding coefficient of determination (R2) values from each estimated equation are then used to compute VIFs as VIF ¼ 1/(1 R2). This test finds that the VIFs for income and income squared are the only ones that are consistently very high. These variables are evidently collinear
Fig. 2 e Potential influence of individual states on least squares estimates.
1091
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 2 e Estimation results. CO2
CH4 (1) Population Income Income squared Farming Landfills
0.99*** (0.07) 25.93 (37.92) 1.26 (1.79) 0.04 (0.04) 0.06 (0.08)
(2) 0.99*** (0.12) 32.76 (46.44) 1.59 (2.18) 0.08* (0.04) 0.03 (0.11)
Manufacturing Natural gas and Oil
0.06** (0.02)
N2O
(1)
(2)
(1)
(2)
0.87*** (0.09) 15.41 (27.52) 0.70 (1.29)
0.90*** (0.10) 10.66 (11.16) 0.47 (0.51)
0.98*** (0.25) 3.07 (67.19) 0.07 (3.14) 0.07 (0.11)
1.08*** (0.20) 11.32 (27.66) 0.42 (1.26) 0.24** (0.10)
0.006 (0.01)
0.02 (0.01)
0.01 (0.03)
0.01 (0.04)
0.30 (0.21)
0.12 (0.13)
0.15** (0.07) 15.89 (357.40) 51 0.68 0.14 0.70 0.97 0.35
0.13* (0.07) 64.16 (151.14) 51 0.52
0.05** (0.02)
Transportation Transportation and utilities Household Size Urbanization Obesity Intercept Observations Adjusted R2/Pseudo-R2 CookeWeisberg c2 P-value ShapiroeWilk W P-value
0.08*** (0.02) 136.47 (200.97) 50 0.90 1.61 0.20 0.97 0.39
0.07** (0.03) 172.15 (246.68) 50 0.74
0.27*** (0.07) 0.57 (1.55) 0.002 (0.008) 0.07* (0.03) 82.13 (146.77) 51 0.84 7.61 0.00 0.96 0.19
0.29*** (0.05) 0.54 (1.44) 0.001 (0.009) 0.07** (0.03) 58.13 (60.84) 51 0.65
Notes: Coefficient estimates are reported with their corresponding standard errors between parentheses. Asterisks, *, **, and *** denote statistical significance respectively at the 0.10, 0.05, and 0.01 levels. Columns (1) and (2) show the results for least squares estimations with bootstrapped standard errors and quantile regression, respectively. Bootstrapped standard errors are computed using 500 replications. Pseudo R2 values are reported for quantile regression.
and do not require any attention. As for the remaining variables, their VIF values (results not included for brevity) range between 1.5 and 2.8 for CH4 estimations, 1.5 and 4.2 for CO2 estimations, and 1.2 and 2.07 for N2O estimations. This suggests that the lack of statistical significance for many of the explanatory variables is not caused by collinearity.
The expected effect of obesity reduction on greenhouse gas emissions The coefficient estimates for obesity, as reported in Table 2, can be interpreted as elasticities measuring the responsiveness of emissions to changes in the obesity rate. Assuming such an interpretation, this would imply that a 10% decrease in the obesity rate can result in a 0.7% decrease in CH4 and CO2, and a 1.3% decrease in N2O. This section assesses the potential impact of reverting to the obesity rates of the year 2000 on greenhouse gas emissions. Table 3 shows the state-level obesity rates for the years 200010 and 2010. For instance, consider Delaware and Alaska, the two states that require the largest and smallest decrease in the obesity rate to revert to the 2000 levels, respectively. In
order for Delaware to revert to its 2000 obesity rate, its 2010 rate must decrease by 42.14%. As a result, based on the reported elasticity coefficients, CH4 emissions would decline by 3.28% (42.14 0.078) or by 3810 metric tons of CO2 equivalent. As for Alaska, reverting to its 2000 obesity rate of 20.5 requires that this state reduces its 2010 obesity rate of 24.3 by 16.33%. Based on the coefficient estimates for CH4, CO2, and N2O emissions, this would result in a reduction of CH4 emissions by about 6910 metric tons (16.33 542,358 0.078/100), a reduction of CO2 by about 0.68 million metric tons (16.33 55 million metric tons 0.076/100), and a reduction of N2O by about 420 metric tons (16.33 19,456 0.133/100). Reverting to 2000 obesity rates in Alaska can result in a total reduction of 689,778 metric tons of CO2 equivalent. Applying the same calculations to the remaining states shows that reverting to 2000 obesity rates across the entire United States can decrease greenhouse gas emissions by more than 180 million metric tons of CO2 equivalent. It is important to note that since the disaggregated data combine the EPA's data for emissions from large facilities with those for emissions from fossil fuel combustion, the aggregated figures may not necessarily match those reported by the
1092
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 3 e Impact of reverting to 2000 obesity rates. State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total
Obesity rate (2000)
Obesity rate (2010)
% change
CH4 decrease (thousand MT)
CO2 decrease (million MT)
N2O decrease (thousand MT)
Total decrease (MT)
23.5 20.5 18.8 22.6 19.2 13.8 16.9 16.2 21.2 18.1 20.9 15.1 18.4 20.9 21.3 20.8 20.1 22.3 22.8 19.7 19.5 16.4 21.8 16.8 24.3 21.6 15.2 20.6 17.2 17.1 17.6 18.8 17.2 21.3 19.8 21 19 21 20.7 16.8 21.5 19.2 22.7 22.7 18.5 17.7 17.5 18.5 22.8 19.4 17.6
32.2 24.5 24.3 30.1 24 21 22.5 28 22.2 26.6 29.6 22.7 26.5 28.2 29.6 28.4 29.4 31.3 31 26.8 27.1 23 30.9 24.8 34 30.5 23 26.9 22.4 25 23.8 25.1 23.9 27.8 27.2 29.2 30.4 26.8 28.6 25.5 31.5 27.3 30.8 31 22.5 23.2 26 25.5 32.5 26.3 25.1
27.02 16.33 22.63 24.92 20.00 34.29 24.89 42.14 4.50 31.95 29.39 33.48 30.57 25.89 28.04 26.76 31.63 28.75 26.45 26.49 28.04 28.70 29.45 32.26 28.53 29.18 33.91 23.42 23.21 31.60 26.05 25.10 28.03 23.38 27.21 28.08 37.50 21.64 27.62 34.12 31.75 29.67 26.30 26.77 17.78 23.71 32.69 27.45 29.85 26.24 29.88
50.58 6.91 26.48 33.04 126.36 29.42 4.27 3.81 0.00 171.98 96.29 6.86 7.52 74.71 83.11 20.87 37.01 42.61 47.45 6.76 28.44 10.92 103.37 17.54 31.15 47.80 12.36 15.58 4.02 6.89 46.73 11.34 52.48 60.18 6.77 118.28 49.73 15.95 71.10 5.93 39.62 3.10 49.23 204.66 8.87 2.03 73.95 20.82 20.74 34.79 5.45 2056
4.72 0.68 2.58 2.08 7.14 3.65 0.89 0.54 0.01 9.23 5.81 0.71 0.44 7.08 7.89 3.05 2.88 5.68 6.63 0.48 2.13 2.05 5.82 3.43 2.17 4.88 1.52 1.39 1.02 0.55 2.77 1.72 4.69 3.91 1.78 8.54 4.85 0.86 8.40 0.39 3.14 0.47 3.20 20.35 1.35 0.11 3.73 2.09 4.08 2.99 2.73 179
25.78 0.42 27.32 23.96 4.60 9.42 1.67 0.78 0.01 229.50 50.82 1.64 1.62 30.40 23.11 42.71 15.26 22.59 105.54 4.05 7.21 3.34 16.99 10.06 72.09 33.02 5.07 4.23 1.67 1.14 2.87 4.71 5.56 14.96 5.91 35.14 74.81 3.30 30.45 0.09 14.19 0.89 10.62 103.27 6.25 0.18 13.06 9.17 15.60 10.98 14.01 1152
4,799,212 689,778 2,634,047 2,140,059 7,274,963 3,686,839 894,980 549,077 13,358 9,630,053 5,954,944 720,957 450,515 7,187,665 7,991,222 3,114,287 2,937,164 5,746,988 6,787,061 494,034 2,167,011 2,064,279 5,939,643 3,459,852 2,271,483 4,959,769 1,538,092 1,408,157 1,028,982 560,397 2,821,372 1,732,859 4,745,236 3,984,497 1,790,861 8,690,409 4,969,538 874,536 8,498,750 394,967 3,190,320 477,529 3,257,774 20,656,317 1,366,237 112,110 3,813,929 2,116,264 4,119,294 3,036,644 2,744,561 182,488,870
Source: The author's own calculations based on coefficient estimates reported in Table 2, census data, the CDC's obesity data, and EPA greenhouse gas emissions data (in metric tons). Note: The % change column represents the required decrease in the obesity rate to revert to the 2000 levels. MT denotes metric tons.
EPA. This is likely due to unidentifiable duplicated entries. In fact, the EPA reports total emissions of 6810 million metric tons in 2010, whereas this data report around 8692 million metric tons. This problem is unavoidable due to the absence of
suitable and accurate disaggregated data. Thus, based on this data, reverting to the obesity rate of 2000 is estimated to result in a two percent (180/8700) decrease in greenhouse gas emissions. Applying this rate of change to EPA's data implies
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
that reverting to the 2000 obesity rates can therefore reduce greenhouse gas emissions by more than 136 million metric tons.
1093
Funding None declared.
Competing interests
Discussion None declared. This paper provides cross-sectional empirical evidence supporting the contention that obesity is statistically associated with higher greenhouse gas emissions, namely CH4, CO2, and N2O, the primary sources of global warming. If appropriate steps are taken to revert the current epidemic to the obesity rates of the year 2000, the country could benefit from a two percent decrease in greenhouse gas emissions estimated at more than 136 million metric tons of CO2 equivalent. While this figure may seem small, it exceeds the combined emissions generated by Sweden and Denmark in 2005. While this paper provides initial empirical insight on the environment-obesity relationship, the author hopes that it represents a foundation for future research in this area. Future studies should establish clear causality between obesity and emissions by using longitudinal data while controlling for other relevant factors. They should also consider identifying means to net out the potential effects of carbon sinks, conversion of CH4 to energy, cross-state diversion, disposal, and transfer of municipal solid waste, and potentially lower energy consumption from increased sedentariness.
Author statements Ethical approval None sought.
references
1. Cawley J, Meyerhoefer C. The medical care costs of obesity: an instrumental variables approach. J Health Econ 2012;3: 219e30. 2. Frone MR. Obesity and absenteeism among US workers: do physical health and mental health explain the relation? J Workplace Behav Health 2007;22:65e79. 3. Bhattacharya J, Sood N. Health insurance and the obesity externality. Adv Health Econ Health Serv Res 2007;17:279e318. 4. Michaelowa A, Dransfeld B. Greenhouse gas benefits of fighting obesity. Ecol Econ 2008;66:298e308. 5. Jacobson SH, King DM. Measuring the potential for automobile fuel savings in the US: the impact of obesity. Transp Res Part D: Transp Environ 2009;14:6e13. 6. Edwards P, Roberts I. Population adiposity and climate change. Int J Epidemiol 2009;2009:1e4. 7. Gryka A, Broom J, Rolland C. Global warming: is weight loss a solution? Int J Obes 2012;36:474e6. 9. Martinez-Zarzosoa I, Maruotti A. The impact of urbanization on CO2 emissions: evidence from developing countries. Ecol Econ 2011;70:1344e53. 10. Mokdad AH, Bowman BA, Ford ES, Vinicor F, Marks JS, Koplan JP. The continuing epidemics of obesity and diabetes in the United States. J Am Med Assoc 2001;286: 1195e200.
Available online at www.sciencedirect.com
Public Health journal homepage: www.elsevier.com/puhe
Original Research
Is obesity associated with global warming? J. Squalli* Department of Economics, American University of Sharjah, P.O. Box 26666, Sharjah, United Arab Emirates
article info
abstract
Article history:
Objectives: Obesity is a national epidemic that imposes direct medical and indirect eco-
Received 13 November 2013
nomic costs on society. Recent scholarly inquiries contend that obesity also contributes to
Received in revised form
global warming. The paper investigates the relationship between greenhouse gas emis-
9 September 2014
sions and obesity.
Accepted 16 September 2014
Study design: Cross-sectional state-level data for the year 2010.
Available online 26 November 2014
Methods: Multiple regression analysis using least squares with bootstrapped standard errors and quantile regression.
Keywords:
Results: States with higher rates of obesity are associated with higher CO2 and CH4 emis-
Greenhouse gas emissions
sions (p < 0.05) and marginally associated with higher N2O emissions (p < 0.10), net of other
Obesity
factors. Reverting to the obesity rates of the year 2000 across the entire United States could
Global warming
decrease greenhouse gas emissions by about two percent, representing more than 136 million metric tons of CO2 equivalent. Conclusions: Future studies should establish clear causality between obesity and emissions by using longitudinal data while controlling for other relevant factors. They should also consider identifying means to net out the potential effects of carbon sinks, conversion of CH4 to energy, cross-state diversion, disposal, and transfer of municipal solid waste, and potentially lower energy consumption from increased sedentariness. © 2014 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.
Introduction According to OECD Health Data 2012, about one third of the US adult population is obese. Based on census and Centers for Disease Control and Prevention (CDC) data, the number of obese people in the United States amounts to approximately 63 million adults in 2010 vs 39 million in 2000. Such a drastic change has prompted the CDC to consider obesity a national epidemic that imposes a heavy burden on society. The direct medical costs associated with obesity have been estimated at $190 billion, representing more than 20% of US spending on healthcare.1 Indirect costs, on the other hand, are estimated at
about $12 billion for lost productive time from work absenteeism and presenteeism2 and around $60 billion for health insurance externality.3 Obesity is also expected to impose an environmental burden by potentially contributing to increased greenhouse gas emissions from higher fuel consumption required to transport heavier people, increased fertilizer usage for higher food production, and increased organic waste generated by larger livestock and heavier people.4 To date, the literature on the relationship between obesity and greenhouse gas emissions is very limited in scope. In fact, while a number of contributions estimate obesity to exert a substantial environmental effect, none substantiate their predictions with
* Tel.: þ971 65152318. E-mail address: [email protected]. http://dx.doi.org/10.1016/j.puhe.2014.09.008 0033-3506/© 2014 The Royal Society for Public Health. Published by Elsevier Ltd. All rights reserved.
1088
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 1 e Summary statistics. Variable CH4 CO2 N 2O Crop and animal farming (% of real GDP) Household size Income ($) Landfills Manufacturing (% of real GDP) Natural gas and oil (% of real GDP) Obesity Population Urbanization Transportation (% of real GDP) Transportation and utilities (% of real GDP)
Obs.
Mean
Std. Dev.
Min
Max
51 51 51 51 51 51 50 51 51 51 51 51 51 51
1,877,535 169,232.9 593,692.3 1.37 2.28 43,124.67 70.72 11.56 1.68 26.99 6,065,298 74.09 3.18 5.04
2,077,286 170,142.1 950,085.9 1.88 0.17 16,997.91 102.18 5.59 4.31 3.23 6,839,909 14.88 1.43 1.53
385 3869.35 895 0 1.83 28,596 3 0.23 0 21 564,554 38.66 0.28 1.55
9,759,024 1,024,036 5,377,550 8.34 2.83 148,605 678 27.48 20.92 34 37,338,198 100 9.44 10.70
Manufacturing and Utilities represent output as a percent of total state GDP in chained 2005 dollars. CO2 emissions are measured in thousand metric tons, whereas CH4 and N2O are measured in metric tons of CO2 equivalent.
econometric estimations that control for the various factors that can contribute to emissions. In other words, the evidence presented in the relevant literature is largely speculative. For example, a study focussing on the European Union and OECD countries uses crude estimates of car, plane, and rail passenger weight to estimate the impact of a 5 kg increase in weight on CO2 emissions.4 It estimates the weight increase to raise CO2 emissions in Germany by 0.7 million metric tons. By extrapolation, the weight increase is also estimated to raise emissions in the European Union by 3.4 million metric tons and by 10.2 million metric tons in OECD countries.4 Another study estimates obesity's contribution to fuel consumption in the United States to raise emissions by more than nine million metric tons of CO2 equivalent annually.5 Similarly, obesity's contribution to food production and car travel in a population of one billion people is estimated to potentially increase greenhouse gas emissions by up to one billion metric tons of CO2 equivalent annually.6 Finally, more recently, based on sample of 25 diabetic and obese individuals in Scotland, a 10 kg increase in body weight is estimated to raise world CO2 emissions by 49.56 million metric tons.7 This paper is a first attempt at assessing the statistical association between emissions and obesity using US statelevel data. The ensuing empirical findings are expected to have important policy implications to better inform policymakers and the public. Understanding the relationship between obesity and the environment ought to guide policymakers in identifying proper corrective measures, whenever applicable. To this end, the paper is organized as follows: Section 2 describes the data and methodology. Section 3 summarizes the estimation results and assesses the potential impact of a reduction in obesity on greenhouse gas emissions. Section 4 discusses the results and concludes.
Methods Data The variables used and their sources are described below. Summary statistics are shown in Table 1.
Greenhouse gas emissions data are from the Environmental Protection Agency (EPA) and include emissions of Methane (CH4), Carbon Dioxide (CO2), and Nitrous Oxide (N2O), measured in metric tons of CO2 equivalent. The data are at the US state level for the year 2010. The EPA implemented in 2010 a mandatory reporting rule for industrial facilities that emit over 25,000 metric tons of CO2 equivalent. While this reporting rule does not include all emitters, it does shed light on emissions at the state level. For CO2 emissions in particular, given the fact that, according to the EPA, emissions from fossil fuel combustion represent about 78% of total emissions, the EPA's data set are combined for emissions from large facilities with emissions from fossil fuel combustion. Data for CH4 and N2O, on the other hand, represent emissions from large facilities. EPA data are also used for measuring the number of municipal solid waste landfills to control for methane generated by waste. To the author's knowledge, the only data available date back to 1995. Population-related data include the population size, which are for the US census of 2010 from the US Census Bureau. Other variables include a measure of urbanization, which is from the US census of 2010 and represents the share of the population in urbanized areas (with more than one million people). The average household size is also used, which is computed using data from the 2010 US census and represents the weighted average household size in rented and owned units. Macroeconomic data include real GDP per capita, which are from the Bureau of Economic Analysis for 2010, in chained 2005 dollars. Real GDP per capita is used as a proxy for affluence. Also the sectoral output data is used, as a percent of real GDP, which are from the Bureau of Economic Analysis for 2010. The sectors included are Crop and Animal Farming, Manufacturing, Natural Gas and Oil, Transportation and Utilities, and Transportation and are used to control for the primary sources of CH4, CO2, and N2O emissions. This paper's the variable of interest represents the obesity rate, which is measured using the percentage of US adults with a body mass index exceeding 30. The data are from the Behavioural Risk Factor Surveillance Systems of the CDC for 2010. Fig. 1 provides preliminary insight on the relationship between greenhouse gas emissions and obesity. Overall, the
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
1089
Fig. 1 e Obesity rate and greenhouse gas emissions.
fitted values seem to capture a positive relationship between the two variables (despite DC being a potential outlier). The next section investigates this relationship further.
ln N2 Oi ¼ b0 þ b1 ln POPi þ b2 ln INCOMEi þ b3 lnðINCOMEi Þ2 þ b4 FARMINGi þ b5 MANUFi þ b6 TRANSi þ b7 OBESITYi þ ti (3)
Methodology The specifications for each of the three primary greenhouse gas emissions are estimated as follows: ln CO2i ¼ a0 þ a1 ln POPi þ a2 ln INCOMEi þ a3 lnðINCOMEi Þ2 þ a4 MANUFi þ a5 TRANSUTILi þ a6 ln HHOLDi þ a7 URBANi þ a8 OBESITYi þ 3 i (1) where CO2 emissions for state i are estimated with respect to population, income, real manufacturing output as a percent of state GDP (MANUF), real transportation and utilities output as a percent of state GDP (TRANSUTIL), average state household size (HHOLD), the urbanization rate (URBAN), and the obesity rate. The variables MANUF and TRANSUTIL are used to control for the primary sources of CO2 emissions, namely industrial processes, electricity production, and fossil fuel combustion. Furthermore, to control for the potential impact of populationrelated factors on emissions, the variables HHOLD and URBAN are used to account for scale economies in energy consumption and for relatively higher energy consumption in urban areas, respectively.9 As for CH4 and N2O emissions, the following specifications are estimated: ln CH4i ¼ g0 þ g1 ln POPi þ g2 ln INCOMEi þ g3 lnðINCOMEi Þ2 þ g4 FARMINGi þ g5 ln LANDFILLi þ g6 OILi þ g7 OBESITYi þ mi (2)
where FARMING represents the real output of crop and animal farming as a percent of real state GDP, LANDFILL represents the number of landfills, and OIL represents the real output of natural gas and oil as a percent of real state GDP. The variables FARMING, LANDFILL, and OIL are used in Equation (2) to control for the primary sources of CH4 emissions, namely livestock farming, municipal solid waste from homes and businesses, and the production, processing, transportation, storage, and distribution of natural gas and petroleum, respectively. It is important to note that it is not necessary to scale the LANDFILL variable since population, a typical scale variable, is included as an explanatory variable. As for Equation (3), FARMING, MANUF, and TRANS are used to control for the primary sources of N2O emissions, namely agricultural soil management (i.e. the use of nitrogen-based fertilizers), industrial or chemical production, and transportation, amongst others. In order to determine the proper estimation procedure, first, a Least Squares estimations of all three specifications are completed. Second, leverage values for each observation are derived and plotted against their corresponding normalized squared residuals. A leverage represents the diagonal component of the hat matrix, which is bounded by the limits 1/n and 1. Observations with high leverage values are likely to be outliers. As Fig. 2 shows, most data points have leverage values that do not exceed 0.6. However, for CO2 and N2O estimations, DC has a leverage value close to unity, suggesting that estimation results may be influenced by DC data. To
1090
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
assess how DC data points affect the estimation results, there is need to exclude DC from the data set and complete Least Squares estimations for CO2 and N2O emissions. The estimation results show no discernible difference warranting the exclusion of DC. After completing Least Squares estimations for the three greenhouse gases, there is no evidence of heteroskedasticity from a visual inspection of plots of the residuals against the fitted values. Applying the CookeWeisberg test for heteroskedasticity, the author fails to reject the null hypothesis of constant variance for CH4 and N2O estimations but not for CO2 estimations. Finally, based on the ShapiroeWilk test, he fails to reject the null hypothesis that residuals are normally distributed. Although these tests may suggest limiting estimations to Ordinary Least Squares the econometric specifications are estimated further using: (a) a Least Squares procedure with bootstrapped standard errors and (b) quantile regression. The benefit of the first procedure lies in its ability to derive estimates of standard errors and confidence intervals based on the underlying distribution of the sample. Given the complex nature of the relationship between greenhouse gas emissions and its determinants and the small sample size, bootstrapping is distribution-independent and can allay concerns about within-sample distortions. Bootstrapping estimations are completed using 500 replications as a means to minimize potential random sampling errors. Furthermore, this procedure is less sensitive to outliers and heteroskedasticity and involves estimating the median (rather than the mean) of the dependent variable by minimizing the sum of absolute residuals. For robustness, both procedures are used, but interpretations focus on quantile regression due to the latter's ability to control for outliers and hetereoskedasticity (especially for CO2 emissions).
Results Columns (1) and (2) in Table 2 report Least Squares estimations with bootstrapped standard errors and quantile regression, respectively. Estimations are also completed using quantile regression with bootstrapped standard errors, which yield very similar results to those reported by the standard quantile regression. Overall, the coefficient estimates for population are close to unity and statistically significant at the 0.01 level across both estimation procedures and for all three greenhouse gases. For CH4 estimations, states with larger farming and natural gas and oil sectors are associated with higher emissions, statistically significant at the 0.10 and 0.05 levels, respectively. As for the other two greenhouse gases, states with larger transportation and utilities sector are associated with higher CO2 emissions, statistically significant at the 0.01 level, whereas states with a larger farming sector are associated with higher N2O emissions, statistically significant at the 0.05 level. On the other hand, the coefficient estimate for obesity is positive and statistically significant across both estimation procedures and for CO2 and CH4 at the 0.05 level and for N2O at the 0.10 level. While the fact that the two estimation procedures yield similar results is undoubtedly a good sign, the absence of statistical significance for a number of included explanatory variables could be due to potential collinearity. The Variance Inflation Factors (VIFs) test can be used to assess the severity of collinearity among variables. VIFs are computed by estimating each variable with respect to other explanatory variables. The corresponding coefficient of determination (R2) values from each estimated equation are then used to compute VIFs as VIF ¼ 1/(1 R2). This test finds that the VIFs for income and income squared are the only ones that are consistently very high. These variables are evidently collinear
Fig. 2 e Potential influence of individual states on least squares estimates.
1091
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 2 e Estimation results. CO2
CH4 (1) Population Income Income squared Farming Landfills
0.99*** (0.07) 25.93 (37.92) 1.26 (1.79) 0.04 (0.04) 0.06 (0.08)
(2) 0.99*** (0.12) 32.76 (46.44) 1.59 (2.18) 0.08* (0.04) 0.03 (0.11)
Manufacturing Natural gas and Oil
0.06** (0.02)
N2O
(1)
(2)
(1)
(2)
0.87*** (0.09) 15.41 (27.52) 0.70 (1.29)
0.90*** (0.10) 10.66 (11.16) 0.47 (0.51)
0.98*** (0.25) 3.07 (67.19) 0.07 (3.14) 0.07 (0.11)
1.08*** (0.20) 11.32 (27.66) 0.42 (1.26) 0.24** (0.10)
0.006 (0.01)
0.02 (0.01)
0.01 (0.03)
0.01 (0.04)
0.30 (0.21)
0.12 (0.13)
0.15** (0.07) 15.89 (357.40) 51 0.68 0.14 0.70 0.97 0.35
0.13* (0.07) 64.16 (151.14) 51 0.52
0.05** (0.02)
Transportation Transportation and utilities Household Size Urbanization Obesity Intercept Observations Adjusted R2/Pseudo-R2 CookeWeisberg c2 P-value ShapiroeWilk W P-value
0.08*** (0.02) 136.47 (200.97) 50 0.90 1.61 0.20 0.97 0.39
0.07** (0.03) 172.15 (246.68) 50 0.74
0.27*** (0.07) 0.57 (1.55) 0.002 (0.008) 0.07* (0.03) 82.13 (146.77) 51 0.84 7.61 0.00 0.96 0.19
0.29*** (0.05) 0.54 (1.44) 0.001 (0.009) 0.07** (0.03) 58.13 (60.84) 51 0.65
Notes: Coefficient estimates are reported with their corresponding standard errors between parentheses. Asterisks, *, **, and *** denote statistical significance respectively at the 0.10, 0.05, and 0.01 levels. Columns (1) and (2) show the results for least squares estimations with bootstrapped standard errors and quantile regression, respectively. Bootstrapped standard errors are computed using 500 replications. Pseudo R2 values are reported for quantile regression.
and do not require any attention. As for the remaining variables, their VIF values (results not included for brevity) range between 1.5 and 2.8 for CH4 estimations, 1.5 and 4.2 for CO2 estimations, and 1.2 and 2.07 for N2O estimations. This suggests that the lack of statistical significance for many of the explanatory variables is not caused by collinearity.
The expected effect of obesity reduction on greenhouse gas emissions The coefficient estimates for obesity, as reported in Table 2, can be interpreted as elasticities measuring the responsiveness of emissions to changes in the obesity rate. Assuming such an interpretation, this would imply that a 10% decrease in the obesity rate can result in a 0.7% decrease in CH4 and CO2, and a 1.3% decrease in N2O. This section assesses the potential impact of reverting to the obesity rates of the year 2000 on greenhouse gas emissions. Table 3 shows the state-level obesity rates for the years 200010 and 2010. For instance, consider Delaware and Alaska, the two states that require the largest and smallest decrease in the obesity rate to revert to the 2000 levels, respectively. In
order for Delaware to revert to its 2000 obesity rate, its 2010 rate must decrease by 42.14%. As a result, based on the reported elasticity coefficients, CH4 emissions would decline by 3.28% (42.14 0.078) or by 3810 metric tons of CO2 equivalent. As for Alaska, reverting to its 2000 obesity rate of 20.5 requires that this state reduces its 2010 obesity rate of 24.3 by 16.33%. Based on the coefficient estimates for CH4, CO2, and N2O emissions, this would result in a reduction of CH4 emissions by about 6910 metric tons (16.33 542,358 0.078/100), a reduction of CO2 by about 0.68 million metric tons (16.33 55 million metric tons 0.076/100), and a reduction of N2O by about 420 metric tons (16.33 19,456 0.133/100). Reverting to 2000 obesity rates in Alaska can result in a total reduction of 689,778 metric tons of CO2 equivalent. Applying the same calculations to the remaining states shows that reverting to 2000 obesity rates across the entire United States can decrease greenhouse gas emissions by more than 180 million metric tons of CO2 equivalent. It is important to note that since the disaggregated data combine the EPA's data for emissions from large facilities with those for emissions from fossil fuel combustion, the aggregated figures may not necessarily match those reported by the
1092
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
Table 3 e Impact of reverting to 2000 obesity rates. State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total
Obesity rate (2000)
Obesity rate (2010)
% change
CH4 decrease (thousand MT)
CO2 decrease (million MT)
N2O decrease (thousand MT)
Total decrease (MT)
23.5 20.5 18.8 22.6 19.2 13.8 16.9 16.2 21.2 18.1 20.9 15.1 18.4 20.9 21.3 20.8 20.1 22.3 22.8 19.7 19.5 16.4 21.8 16.8 24.3 21.6 15.2 20.6 17.2 17.1 17.6 18.8 17.2 21.3 19.8 21 19 21 20.7 16.8 21.5 19.2 22.7 22.7 18.5 17.7 17.5 18.5 22.8 19.4 17.6
32.2 24.5 24.3 30.1 24 21 22.5 28 22.2 26.6 29.6 22.7 26.5 28.2 29.6 28.4 29.4 31.3 31 26.8 27.1 23 30.9 24.8 34 30.5 23 26.9 22.4 25 23.8 25.1 23.9 27.8 27.2 29.2 30.4 26.8 28.6 25.5 31.5 27.3 30.8 31 22.5 23.2 26 25.5 32.5 26.3 25.1
27.02 16.33 22.63 24.92 20.00 34.29 24.89 42.14 4.50 31.95 29.39 33.48 30.57 25.89 28.04 26.76 31.63 28.75 26.45 26.49 28.04 28.70 29.45 32.26 28.53 29.18 33.91 23.42 23.21 31.60 26.05 25.10 28.03 23.38 27.21 28.08 37.50 21.64 27.62 34.12 31.75 29.67 26.30 26.77 17.78 23.71 32.69 27.45 29.85 26.24 29.88
50.58 6.91 26.48 33.04 126.36 29.42 4.27 3.81 0.00 171.98 96.29 6.86 7.52 74.71 83.11 20.87 37.01 42.61 47.45 6.76 28.44 10.92 103.37 17.54 31.15 47.80 12.36 15.58 4.02 6.89 46.73 11.34 52.48 60.18 6.77 118.28 49.73 15.95 71.10 5.93 39.62 3.10 49.23 204.66 8.87 2.03 73.95 20.82 20.74 34.79 5.45 2056
4.72 0.68 2.58 2.08 7.14 3.65 0.89 0.54 0.01 9.23 5.81 0.71 0.44 7.08 7.89 3.05 2.88 5.68 6.63 0.48 2.13 2.05 5.82 3.43 2.17 4.88 1.52 1.39 1.02 0.55 2.77 1.72 4.69 3.91 1.78 8.54 4.85 0.86 8.40 0.39 3.14 0.47 3.20 20.35 1.35 0.11 3.73 2.09 4.08 2.99 2.73 179
25.78 0.42 27.32 23.96 4.60 9.42 1.67 0.78 0.01 229.50 50.82 1.64 1.62 30.40 23.11 42.71 15.26 22.59 105.54 4.05 7.21 3.34 16.99 10.06 72.09 33.02 5.07 4.23 1.67 1.14 2.87 4.71 5.56 14.96 5.91 35.14 74.81 3.30 30.45 0.09 14.19 0.89 10.62 103.27 6.25 0.18 13.06 9.17 15.60 10.98 14.01 1152
4,799,212 689,778 2,634,047 2,140,059 7,274,963 3,686,839 894,980 549,077 13,358 9,630,053 5,954,944 720,957 450,515 7,187,665 7,991,222 3,114,287 2,937,164 5,746,988 6,787,061 494,034 2,167,011 2,064,279 5,939,643 3,459,852 2,271,483 4,959,769 1,538,092 1,408,157 1,028,982 560,397 2,821,372 1,732,859 4,745,236 3,984,497 1,790,861 8,690,409 4,969,538 874,536 8,498,750 394,967 3,190,320 477,529 3,257,774 20,656,317 1,366,237 112,110 3,813,929 2,116,264 4,119,294 3,036,644 2,744,561 182,488,870
Source: The author's own calculations based on coefficient estimates reported in Table 2, census data, the CDC's obesity data, and EPA greenhouse gas emissions data (in metric tons). Note: The % change column represents the required decrease in the obesity rate to revert to the 2000 levels. MT denotes metric tons.
EPA. This is likely due to unidentifiable duplicated entries. In fact, the EPA reports total emissions of 6810 million metric tons in 2010, whereas this data report around 8692 million metric tons. This problem is unavoidable due to the absence of
suitable and accurate disaggregated data. Thus, based on this data, reverting to the obesity rate of 2000 is estimated to result in a two percent (180/8700) decrease in greenhouse gas emissions. Applying this rate of change to EPA's data implies
p u b l i c h e a l t h 1 2 8 ( 2 0 1 4 ) 1 0 8 7 e1 0 9 3
that reverting to the 2000 obesity rates can therefore reduce greenhouse gas emissions by more than 136 million metric tons.
1093
Funding None declared.
Competing interests
Discussion None declared. This paper provides cross-sectional empirical evidence supporting the contention that obesity is statistically associated with higher greenhouse gas emissions, namely CH4, CO2, and N2O, the primary sources of global warming. If appropriate steps are taken to revert the current epidemic to the obesity rates of the year 2000, the country could benefit from a two percent decrease in greenhouse gas emissions estimated at more than 136 million metric tons of CO2 equivalent. While this figure may seem small, it exceeds the combined emissions generated by Sweden and Denmark in 2005. While this paper provides initial empirical insight on the environment-obesity relationship, the author hopes that it represents a foundation for future research in this area. Future studies should establish clear causality between obesity and emissions by using longitudinal data while controlling for other relevant factors. They should also consider identifying means to net out the potential effects of carbon sinks, conversion of CH4 to energy, cross-state diversion, disposal, and transfer of municipal solid waste, and potentially lower energy consumption from increased sedentariness.
Author statements Ethical approval None sought.
references
1. Cawley J, Meyerhoefer C. The medical care costs of obesity: an instrumental variables approach. J Health Econ 2012;3: 219e30. 2. Frone MR. Obesity and absenteeism among US workers: do physical health and mental health explain the relation? J Workplace Behav Health 2007;22:65e79. 3. Bhattacharya J, Sood N. Health insurance and the obesity externality. Adv Health Econ Health Serv Res 2007;17:279e318. 4. Michaelowa A, Dransfeld B. Greenhouse gas benefits of fighting obesity. Ecol Econ 2008;66:298e308. 5. Jacobson SH, King DM. Measuring the potential for automobile fuel savings in the US: the impact of obesity. Transp Res Part D: Transp Environ 2009;14:6e13. 6. Edwards P, Roberts I. Population adiposity and climate change. Int J Epidemiol 2009;2009:1e4. 7. Gryka A, Broom J, Rolland C. Global warming: is weight loss a solution? Int J Obes 2012;36:474e6. 9. Martinez-Zarzosoa I, Maruotti A. The impact of urbanization on CO2 emissions: evidence from developing countries. Ecol Econ 2011;70:1344e53. 10. Mokdad AH, Bowman BA, Ford ES, Vinicor F, Marks JS, Koplan JP. The continuing epidemics of obesity and diabetes in the United States. J Am Med Assoc 2001;286: 1195e200.
