i need help

Economics and Human Biology 34 (2019) 16–25

Understanding recent trends in childhood obesity in the United States

Patricia M. Andersona, Kristin F. Butcherb, Diane Whitmore Schanzenbachc,* aDartmouth College and NBER, United States1 bWellesley College and NBER, United States cNorthwestern University and NBER, United States

A R T I C L E I N F O

Article history: Received 16 August 2018 Received in revised form 14 February 2019 Accepted 18 February 2019 Available online 28 February 2019

Keywords: Childhood obesity Trends in obesity

A B S T R A C T

The prevalence of childhood obesity in the United States has more than tripled over the last four decades from 5 percent in 1978 to 18.5 percent in 2016. There is evidence for a break in trend in recent years: after growing from 0.4 to 0.7 percentage point per year between 1978 and 2004, the rate of increase has slowed to 0.1 percentage point per year from 2004 to 2016. To better understand these trends, in this paper we analyze a range of datasets that collect information on childhood obesity. We analyze the data overall, across the age distribution, across birth cohorts, and for subgroups of interest. We find steady increases in cohort-level obesity prevalence through approximately age 10, with levels unchanged thereafter, suggesting a need for additional interventions at early ages. We find that the prevalence of obesity has diverged by race and gender in recent years, especially among children entering kindergarten. Compared with 5-year-olds in 1997, 5-year-olds in 2010 were 2 percentage points more likely to be obese overall. Black and Hispanic 5-year-olds were 5 and 3 percentage points more likely to be obese, respectively, while whites had a 1 percentage point increase in obesity. However, overall and among all subgroups the rate of growth in obesity from kindergarten through 3rd grade has declined in recent years. Together, these findings can inform a future research literature that aims to target obesity interventions where they will be most impactful.

© 2019 Elsevier B.V. All rights reserved.

Contents lists available at ScienceDirect

Economics and Human Biology

journal homepage: www.elsev ier .com/locate /ehb

1. Introduction

Over the last four decades although many correlates of life expectancy have shown marked improvement, the prevalence of obesity has risen worldwide (NCD Risk Factor Collaboration, 2017). Dubbed a public health crisis by the Surgeon General (U.S. Department of Health and Human Services, 2001), obesity is extremely prevalent in the United States: 18.5 percent of children and 39.8 percent of adults are obese (Hales et al., 2017), and the prevalence of childhood obesity in the United States has more than tripled over the last four decades. Resources have poured into addressing the issue through public health efforts, reforms to school policies and nutrition assistance programs, and more. There is evidence for a break in trend in recent years, suggesting that efforts to address childhood obesity may be having success. In particular, data from the National Health and Nutrition Examina- tion Survey (NHANES), the official source for obesity statistics in the United States, shows that the rapid growth in the prevalence of

* Corresponding author. E-mail address: [email protected] (D.W. Schanzenbach).

1 The authors are grateful to Abigail Pitts for excellent research assistance, and to Tina Kauh and Betsy Thorn for helpful comments.

https://doi.org/10.1016/j.ehb.2019.02.002 1570-677X/© 2019 Elsevier B.V. All rights reserved.

childhood obesity seen in the 1980s and 1990s has been slowed in recent years. In addition, some recent work on very young children who participate in the Supplemental Nutrition Program for Women, Infants, and Children (WIC) has found recent declines in obesity among that population.

To better understand trends in childhood obesity in the United States, in this paper we analyze a range of datasets. We examine childhood obesity overall and for subgroups of interest. We find differences in the prevalence of obesity by race that are large and persistent. We find particularly troubling evidence that the prevalence of obesity is diverging by race and gender. In addition, we find that cohort-level obesity prevalence increases through approximately age 10 and flattens out thereafter. Together, these findings can inform a future research literature that aims to target obesity interventions where they will be most impactful.

2. Literature review

Economists have interest in obesity for many reasons. First, obesity is correlated with morbidity, increased medical costs, and mortality. Second, obesity is an excellent example of time inconsistency and hyperbolic discounting, where although indi- viduals have strong preferences for not being obese, it is very challenging to make short-term decisions that are consistent with

P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25 17

that long-term outcome. Third, the sharp time trends in the prevalence of obesity suggest that something in the environment may have changed to make that short-term versus long-term decision-making more challenging: food prices or access, technol- ogy, family structure, the built environment, to name a few. Since its founding in 2003, Economics and Human Biology has published many papers that have increased our understanding of the economics of obesity. The prior economics literature can be categorized into two broad types of studies: those that examine trends in obesity and determine how those trends line up with changes in behavior and/or the environment, and those that attempt to estimate the causal impacts of particular behaviors, policies or environmental factors.

Investigating trends among adults in the United States, Lakdawalla and Philipson (2009) decompose the increase in obesity among adults from 1976 to 1994 into two factors: decreasing food prices vs. demand-side factors such as physical activity levels. They conclude that 40 percent of the growth in obesity can be attributed to decreasing food prices. Over essentially the same time frame, Cutler et al. (2003) find that much of the obesity increase is due to increased caloric intake (instead of decreased calorie burn). They attribute the increase to declines in the costs of food preparation and purchase driven by technological innovations.

Childhood obesity is a particularly compelling area of research because, while a case can be made that if adults have full information about the impact of their behaviors on their outcomes, they should be free to make those decisions and live with their consequences, no such case can be made for children. Further, childhood obesity is a strong predictor of adult obesity, and as we show below, obesity patterns are set by about age 11. Thus, if one wants to address adult obesity, understanding the drivers of childhood obesity, and understanding interventions that may address childhood obesity, are of paramount importance.

There is some uncertainty to what extent the same factors that have been found to predict increases in adults’ obesity also predict the increase in children’s obesity. Adults’ (parents’) and children’s obesity prevalence have tended to move in tandem over time. A mother’s obesity status is positively associated with her children’s obesity status (Classen and Hokayem, 2005), and the correlation between parent’s and child’s BMI has increased somewhat over time (Anderson et al., 2009). The correlation between parents’ and children’s caloric intake is also substantial, especially regarding calories eaten away from home or calories from fast food. Cawley (2010) reviews the literature and finds that many of the factors influencing adult obesity also affect children, noting important roles for food prices and technology, as well as increased impatience. Anderson and Butcher (2006a,b) find that many factors have influenced the increase in childhood obesity, including children’s access to calorie-dense foods, increases in maternal employment, and reductions in children’s energy expenditure. Komlos et al. (2009) find that increased sedentary lifestyles among children play a role. They trace the long-term increases in BMI among black and white U.S. children from 1941 to 2004, finding that increased television viewing over this time is an important predictor of increases in BMI. As children age through elementary school, socio-economic status (SES) gradients in obesity increases, as documented by Jo (2014). Interestingly, Jo finds that the SES gradient is attenuated when school-level fixed effects are included in the prediction model, suggesting that different schools attended (and neighborhoods lived in) can explain much of the divergence as children age. On the other hand, controlling for parent employment and measured family behaviors does little to explain the growth of the SES gradient.

The second strain of literature investigates the impact of various characteristics on obesity, typically using quasi-experimental

research designs. These studies span a wide range of contributing factors, including family characteristics, school policies, preschool attendance, child health influences, and the built environment. Among the studies on children, a substantial literature investigates the impact of maternal employment and concludes that mothers’ employment plays a causal role in increasing children’s obesity (Anderson et al., 2003). There has been less consensus on the mechanisms driving the relationship, though. For example, Anderson (2012) finds that differences in reported family routines (e.g. eating meals together) do not explain the relationship between maternal employment and obesity.

Another literature investigates the role of school environments and policies. Schanzenbach (2009) finds that eating school lunches increases childhood obesity, while Anderson and Butcher (2006) find that access to “junk food” in schools increases obesity—though the impact is concentrated among students with obese parents. Using variation in state physical education (PE) requirements, Cawley et al. (2013) find that PE classes reduce obesity among 5th

grade students, with the reduction concentrated among boys. Over recent decades, more children have been attending preschools (Cascio and Schanzenbach, 2013), but the literature has mixed results on the impacts this would be expected to have on obesity. Using variation in the receipt of a childcare subsidy to identify the impact of attending childcare, Herbst and Tekin (2011) find that attending center-based care increases obesity among kindergarten entrants, with the largest impacts at the upper end of the BMI distribution. On the other hand, Belfield and Kelly (2013) find that preschool or Head Start attendance improves obesity outcomes, potentially through the mechanisms of improved nutrition intake and health screening. Frisvold and Lumeng (2011) find that attending a full-day Head Start program instead of a partial-day program reduces obesity.

The onset of child health conditions has also been shown to increase obesity, and to the extent that these conditions are disproportionately found among more disadvantaged populations this may contribute to an increasing disparity in obesity across groups. Green (2014) finds that asthma onset is related to subsequent weight gain. Bowling et al. (2017) find that taking medication for ADHD increases a child’s BMI growth trajectory between fifth and eighth grade.

There are many correlational links between the built environ- ment and childhood obesity (Sallis and Glanz, 2006, for a review), but causal linkages have been harder to establish. Both the food environment and access to exercise opportunities appear to influence childhood obesity. For example, Alviola et al. (2014) find that closer proximity to fast food restaurants increases obesity, using distance to the nearest highway as an instrument for fast food access. Sandy et al. (2013) combine information on recreational walking trails, violent crime, and children’s health records in Indiana and find that having a walking trail near a child’s home reduces obesity, with more pronounced effects for boys and older children. The effects are only significant in areas with low crime, where presumably families are more able to utilize the trails safely.

3. Data

To study trends in childhood obesity in the United States, we analyze data from six different nationally representative datasets, each with its own strengths and weaknesses. The datasets are summarized in Appendix Table 1.

The National Health and Nutrition Examination Survey (NHANES) is the official source for prevalence of obesity in the United States. Participants are directly weighed and measured by the survey collectors. The data are collected in an ongoing manner—spanning from 1976 to 2016—and are comprehensive,

Fig. 1. Overall Prevalence of Childhood Obesity in the United States.

3 See https://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf. Results are qualitatively similar, except where noted, when obesity is defined using international cutoff points recommended by the Childhood Obesity Working Group of the International Obesity Taskforce.

4 We also conduct the analysis for adolescents in the YRBSS data. 5

18 P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25

allowing for analysis by age, gender, racial/ethnic and income groups. On the other hand, it is a fairly small dataset with fewer than 4000 children measured in each wave. As a result, the confidence intervals on statistics calculated from NHANES are relatively wide, especially among subgroups.

We also analyze measured height and weight in the Early Childhood Longitudinal Study (ECLS-K) data across two cohorts of students: those who entered kindergarten in the fall of 1998 and 2010. These are large datasets with approximately 14,000 children per cohort, allowing us to follow individual children as they age from the start of kindergarten through selected subsequent grades.2 Conveniently, the first cohort is observed during the sharp increase in national childhood obesity prevalence, while the second is observed during the period when the national prevalence leveled off (see Fig. 1, below).

We supplement the work using the NHANES and ECLS-K using a range of additional datasets. Two additional data sources cover adolescents, but they are limited because they rely on self- reported height and weight data; as a result, levels of obesity are understated. Nonetheless, the trends in obesity measured in these datasets are likely to be meaningful. The Youth Risk Behavior Surveillance System (YRBSS) collects cross-sectional waves on approximately 13,500 adolescents between ages 12 and 18, every other year from 1991 to 2015. The National Health Interview Survey (NHIS) complements the YRBSS, surveying approximately 4000 adolescents per wave annually between 2008 and 2015. We use both of these datasets to analyze trends, and also to create synthetic cohort groups to investigate growth in obesity as a cohort ages. To gain insights into the youngest children, we also analyze measured BMI on participants in the Supplemental Nutrition Program for Women, Infants and Children (WIC) who range between ages 2 and 4, using the WIC Participant and Program Characteristics (WIC-PC) data. This is a large sample of young children, with approximately 20,000 observations in each biennial wave, which allows us to examine racial and ethnic groups, including Native Americans, that cannot be reliably measured in smaller nationally representative datasets. A limita- tion of these data, however, is they only include WIC participants, and participation rates vary by child age and race/ethnicity, and over time.

2 The ECLS-K 1998 surveys children at the beginning and end of kindergarten and first grade, and the end of grades 3, 5 and 8. The ECLS-K 2010 surveys children at the beginning and end of kindergarten and first grade, and at the end of grades 2, 3, and 4.

4. Methodology

We start by calculating the average prevalence of obesity over time across a wide range of characteristics, including gender, age, race/ethnicity, socio-economic status, geography, and interactions of these measures. In addition, we fit linear trend models to the data and allow for break points in the growth rate of obesity. Using microdata in the NHANES, YRBSS, and WIC-PC, we fit the best linear prediction model across years using ordinary least squares. In particular, we run models of the following form:

Obeseit = β0 + β1trendt + β2Post1999t + β3Post2002t + β4trendt*- Post1999t + β5trendt*Post2002t + eit (1)

Where trendt represents a linear time trend, and Post1999t (Post2002t) is an indicator variable for whether the observation occurred after 1999 (2002). Coefficients on the interaction terms, β4 and β5, show to what extent growth has changed over time. We present results separately by age, race, and sex. Obesity is calculated as an indicator variable for whether the child’s body- mass-index (BMI) is at or above the 95th percentile of a fixed gender and age specific BMI calculated using the 2000 CDC growth charts, drawn from national data from 1963–65 to 1988–94.3

Next, we examine the prevalence of obesity across birth-year cohorts as the cohort ages, estimating regressions of the form:

Obeseit = β0 + β1aget + β2AgeBreak1t + β3 aget*AgeBreak1t + eit, (2)

Where AgeBreak1t is an indicator variable for the observation being from an age above a breakpoint age determined by the data. Due to differences in data availability, we conduct this analysis for children age 2 to 12 and 6 to 18 in the NHANES, and for children from 5 to 10 or 5 to 14, in the ECLS-K, depending on the ECLS-K sample.4 While the ECLS-K is a true cohort study, following children from a given cohort over time as they age, the NHANES is a series of repeated cross-sections. Thus, for the NHANES we create synthetic cohorts by calculating birth year in each cross section. Then, by pooling cross sections over time we can observe a birth cohort “age.”

5. Results: national trends in childhood obesity

Table 1, calculated from NHANES data, shows the prevalence of obesity among children aged 2–19 in the U.S. from 1978 through 2016, along with standard errors. The results are also shown graphically in Fig. 1, with each dot representing the average prevalence of obesity in a particular survey wave, the solid line representing time trends, and the dotted lines surrounding the data representing standard errors. Note that the frequency of data collection changed over time: data were collected about every decade in the 1970s through 1990s, then starting in 2002 the survey was fielded continuously, allowing the measurement of obesity every two years.5 The prevalence of obesity more than tripled over this period, growing from 5 percent in 1978 to 18.5 percent in 2016.

We use NHANES II (1976–1980, coded as 1978), NHANES III (1988–1994, coded as 1991), and the continuously collected data released every two years covering the following years: 1999–2000, 2001–2002, 2003–2004, 2005–2006, 2007–2008, 2009-2010, 2011–2012, 2013–2014, 2015–2016, 2017–2018, coded as the latter of each pair of years. The “dots” on the graphs are placed at the midpoint of the years for NHANES II and III, and the final year of the continuous NHANES surveys (2000, 2002, etc.).

Table 1 Prevalence of Childhood Obesity Over Time: NHANES.

Year All White Black Hispanic Age 2–5 Age 6– 11 Age 12– 19 Male Female (1) (2) (3) (4) (5) (6) (7) (8) (9)

1978 0.056 0.052 0.080 0.111 0.048 0.069 0.049 0.055 0.056 (0.003) (0.003) (0.010) (0.017) (0.005) (0.007) (0.005) (0.005) (0.005)

1991 0.106 0.096 0.127 0.148 0.074 0.122 0.115 0.107 0.106 (0.005) (0.007) (0.006) (0.007) (0.005) (0.010) (0.009) (0.007) (0.007)

2000 0.141 0.111 0.187 0.187 0.106 0.148 0.152 0.140 0.141 (0.008) (0.011) (0.012) (0.015) (0.016) (0.016) (0.011) (0.011) (0.011)

2002 0.154 0.138 0.176 0.192 0.105 0.163 0.167 0.163 0.144 (0.007) (0.010) (0.011) (0.014) (0.014) (0.014) (0.010) (0.011) (0.009)

2004 0.170 0.160 0.201 0.192 0.137 0.183 0.175 0.177 0.161 (0.008) (0.012) (0.011) (0.015) (0.015) (0.016) (0.011) (0.012) (0.011)

2006 0.156 0.131 0.213 0.225 0.116 0.149 0.180 0.161 0.151 (0.007) (0.010) (0.012) (0.014) (0.014) (0.013) (0.011) (0.010) (0.010)

2008 0.170 0.155 0.201 0.209 0.106 0.195 0.183 0.180 0.159 (0.008) (0.012) (0.015) (0.012) (0.013) (0.014) (0.013) (0.011) (0.012)

2010 0.169 0.141 0.242 0.212 0.125 0.177 0.186 0.188 0.149 (0.008) (0.012) (0.017) (0.012) (0.013) (0.013) (0.013) (0.011) (0.010)

2012 0.169 0.141 0.202 0.224 0.083 0.181 0.204 0.167 0.172 (0.009) (0.015) (0.013) (0.014) (0.011) (0.013) (0.016) (0.012) (0.013)

2014 0.172 0.153 0.188 0.213 0.094 0.175 0.206 0.172 0.171 (0.008) (0.013) (0.014) (0.012) (0.011) (0.013) (0.014) (0.011) (0.012)

2016 0.185 0.141 0.220 0.258 0.139 0.184 0.207 0.191 0.178 (0.008) (0.013) (0.016) (0.014) (0.014) (0.012) (0.014) (0.011) (0.011)

Observations 53,958 19,226 14,636 16,979 17,245 15,575 21,138 27,238 26,720

Note: Standard errors in parenthesis.

Table 2 Trends in Childhood Obesity Over Time: NHANES.

(1) (2) (3) (4) (5) (6) (7) (8) (9) All White Black Hispanic Age 2–5 Age 6–11 Age 12–19 Male Female

post-1998 �0.057 �0.233 0.236 0.029 0.064 �0.083 �0.057 �0.171 0.059 (0.124) (0.178) (0.187) (0.237) (0.242) (0.243) (0.171) (0.181) (0.170)

post-2002 0.128 0.344* �0.111 �0.034 0.037 0.169 0.102 0.264 �0.012 (0.126) (0.181) (0.191) (0.240) (0.246) (0.246) (0.175) (0.184) (0.173)

Time trend 0.388*** 0.340*** 0.360*** 0.285** 0.198*** 0.409*** 0.501*** 0.393*** 0.383***

(coefficient*100) (0.046) (0.060) (0.090) (0.143) (0.058) (0.091) (0.076) (0.067) (0.062) Time trend*post-1998 0.257 0.988 �0.943 �0.068 �0.229 0.327 0.226 0.770 �0.263 (coefficient *100) (0.538) (0.770) (0.812) (1.034) (1.047) (1.052) (0.743) (0.785) (0.736) Time trend*post-2002 �0.510 �1.382* 0.600 0.137 �0.080 �0.666 �0.423 �1.076 0.067 (coefficient *100) (0.542) (0.776) (0.817) (1.032) (1.053) (1.056) (0.749) (0.789) (0.740) Constant 0.056*** 0.052*** 0.080*** 0.111*** 0.048*** 0.069*** 0.049*** 0.055*** 0.056***

(0.003) (0.003) (0.010) (0.017) (0.005) (0.007) (0.005) (0.005) (0.005)

Observations 53,958 19,226 14,636 16,979 17,245 15,575 21,138 27,238 26,720 R-squared 0.009 0.009 0.011 0.005 0.004 0.007 0.014 0.010 0.008

Note: Robust standard errors in parentheses. *** p < 0.01. ** p < 0.05. * p < 0.1.

P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25 19

Table 2 presents results from regression Eq. (1). The estimated time trend (multiplied by 100) in the overall sample is 0.388 from 1978 to 2000, meaning that the prevalence of obesity increased by almost 0.4 percentage points each year. The trend increased by 0.257 (implying annual growth of 0.65 percentage points) from 2000 to 2004, then decreased by 0.51 (implying a flattened annual

growth rate of 0.14 percentage points) from 2004 through 2016, though neither change in trend is statistically significant at conventional levels. To put these slopes into context, from 1978 to 2002 there was growth in child obesity equal to approximately 1 percentage point every 2.5 years; starting in 2004, the growth rate fell to a rate of about 1 percentage point every 8 years. The overall

20 P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25

prevalence of obesity has been statistically unchanged from 2004 to 2016.

The levels and the trends in obesity vary by race and ethnic group.6 Columns (2) through (4) of Table 1 display the prevalence of obesity among children age 2–19 separately by race and ethnic group.7 All groups show an upward trend since 1978, with a flattening of the time trends around the early 2000’s. Obesity prevalence among Hispanics and blacks are significantly higher than among whites, with 2016 rates of 26, 22, and 14 percent, respectively. As shown in Table 2, the 1978–1998 time-trends for whites and blacks are somewhat higher than for Hispanics, though the difference is not statistically significant. Among whites (only), there has been no trend growth between 2004 and 2016.

The prevalence of obesity also follows a different pattern across age groups. As shown in columns (5) through (7) of Tables 1 and 2, all age groups saw a large increase in obesity between 1978 and 2004, and an apparent trend break thereafter.8 Children ages 2 to 5 experienced a slower increase in obesity in recent years, and there was a statistically significant decline in the prevalence of obesity from 2004 to 2014 before the obesity point estimate among the youngest children spiked up in the latest wave, statistically significantly increasing from 9.4 percent in 2014 to 13.9 percent in 2016. Because the sample size here is rather small (average N = 885), we will need more data to know whether this trend will continue. Obesity prevalence for 2 to 4-year-olds in the WIC data, shown below, do not show a similar increase between 2014 and 2016.

Children aged 6–11 and 12–19 have higher prevalence of obesity than the youngest children. While there is no difference in the prevalence of obesity among 6 to 11-year-olds between 2004 and 2016, those aged 12–19 have seen a significant 3 percentage point increase over that time period. We find that the prevalence of obesity is statistically indistinguishable between middle-group children age 6–11 and older children age 12–19, but both groups have significantly higher prevalence than children age 2-5. We explore the relationship between age and obesity in more detail below. As shown in columns (8) and (9), boys and girls had essentially the same prevalence of obesity until 2000, then boys’ obesity increased relative to girls’ in the early 20000s but have converged in recent years.9

Over this time period there has been a sizeable change in the demographic characteristics of the population. For example, the share of the population that is white declined from 75 to 52 percent, while the share that is Hispanic grew from 7.5 to 24 percent. Since Hispanics have a higher prevalence of obesity than whites, the compositional change of the population likely explains some of the increase in the overall prevalence of obesity. To formally test this, in Table 3 we present results from a series of Blinder-Oaxaca decompositions. To calculate the decomposition, we first run separate regressions predicting obesity in a “base” year and a “final” year, for example 1978 and 2016, as a function of racial/ethnic categories, age groups, and gender. That is, for a given year we estimate:

Obeseit= β0 + β1blackit+ β2Hispanicit + β3otherit+ β4age2to5i + β5age12to19it + β6femaleit + eit (3)

6 When analyzing obesity separately by gender in the NHANES, we find that boys and girls follow similar patterns to the overall prevalence. The point estimates tend to be higher for boys, but differences across gender are not statistically significant.

7 These are displayed graphically in Appendix Fig. 1. 8 The prevalence of obesity by age group in the NHANES is displayed graphically

in Appendix Fig. 2. 9 See Appendix Fig. 3 for obesity displayed graphically.

Based on the estimated coefficients from these regressions and the variable means in each year, the overall change from 1978 to 2016 can be decomposed as follows10:

Obese2016� Obese1978 = [β12,016 (black2016� black1978 + β22,016 (his- panic2016� hispanic1978) + β32,016 (other2016– other1978) + β42,016 (age2to52016� age2to51978) + β52,016 (age12to192016� age12to191978) + β62,016 (female2016� female1978)] + [(β02,016� β01,978) + (β12,016� β11,978)black1978 + (β22,016� β21,978) hispanic1978 + (β32,016� β31,978)other1978 + (β42,016� β41,978)age2to51978 + (β52,016� β51,978)age12to191978 + (β62,016� β61,978)female1978] (4)

In the first bracketed expression we multiply the change in mean values of the control variables by the estimated coefficients from the final year regression to calculate the change in obesity predicted by the change in the demographic composition of the population over the time frame. The ratio of this prediction to the total change in obesity across the two years gives the share of the change that is explained by the shift in demographics.11

Columns (1) and (2) display the results from the Blinder-Oaxaca decomposition over the entire time period,1978 to 2016. Of the 13.0 percentage-point increase in obesity over this time period, 2.1 percentage points (16% of the total change) are explainable by changes in the demographic composition of the population—the vast majority of which is the increase in Hispanics. The remaining columns show decompositions for subsets of years. Only 11 percent of the 8.5 percentage-point increase between 1978 and 2000 can be explained by covariates, and only 1.4 percent of the 2.9 percentage- point increase between 2000 and 2004 can be explained. Of the 1.5 percentage point increase in obesity between 2004 and 2016, over half can be explained by changes in demographic characteristics, again mainly the increase in Hispanics.

In summary, the data show a large increase then flattening of obesity trends, and overall relatively little of this can be explained by observable characteristics. As noted previously, though, there is too much imprecision in the NHANES due to its small sample size to say with statistical certainty whether trends are continuing to increase, or have flattened out, and how this varies across subgroups. Because of this statistical imprecision, we turn to additional datasets to augment the analysis.

6. Obesity by age group

Because there is uncertainty about whether the rate of increase in obesity has stopped growing within age group, we turn to additional (larger) datasets to assess the time trends. To assess older children, we analyzeYRBSSandNHIS,bothofwhichsurveymanymoreadolescents than the NHANES but only have self-reported data on height and weight. Fig. 2 shows time trends in the prevalence of obesity in the NHANES (ages 12–19), YRBSS (ages 12–18), and NHIS (ages 12–17). Although the years are different from the NHANES, the methodology hereisthesameasinEq. (1):wefit lineartimetrendsonthemicrodata, allowing for breaks in the slopes at optimal time points.

It is immediately evident that the self-reported data indicate a lower level of obesity, as people misreport their height and weight in predictable ways.12 However, the trends over time appear to be more reliable than the levels; statistically the time trends are

10 In the table we decompose not only the full time period change, but also the changes over each of the three sub-periods with different trends. The process is identical to that described for the full period. 11 The remainder, which can also be calculated as the change in estimated coefficients times the mean values of the x variables in the initial year (as shown in the second bracketed expression), is the portion explained by changes in the coefficients, which is typically considered to be the unexplained portion. 12 Females tend to understate their weight, and males tend to overstate their height.

Table 3 Blinder-Oaxaca Decomposition of Change in Child Obesity Over Time in NHANES.

Total Change in Obesity

Year 2016 – Year 1978 Year 2000 – Year 1978 Year 2004 – Year 2000 Year 2016 – Year 2004

0.130 0.085 0.029 0.015

Explained by change in covar. mean

Explained by change in covar. coeff.

Explained by change in covar. mean

Explained by change in covar. coeff.

Explained by change in covar. mean

Explained by change in covar. coeff.

Explained by change in covar. mean

Explained by change in covar. coeff.

(1) (2) (3) (4) (5) (6) (7) (8)

Black �0.001 0.007 0.000 0.006 0.000 �0.005 �0.001 0.006 Hispanic 0.020 0.004 0.009 0.001 �0.001 �0.008 0.008 0.015 Other Race 0.004 0.002 0.002 0.002 0.000 �0.007 0.002 0.006 Age 2 to 5 �0.001 �0.005 �0.002 �0.004 0.000 �0.001 0.000 0.000 Age 12 to 19 �0.001 0.022 0.000 0.011 0.000 �0.005 0.000 0.016 Female 0.000 �0.007 0.000 0.000 0.000 �0.008 0.000 0.002 Constant 0.000 0.086 0.000 0.061 0.000 0.063 0.000 �0.038 All Covariates 0.021 0.109 0.009 0.076 0.000 0.028 0.009 0.007 Percent Explained by change

in Covariate Means

15.9% 10.9% 1.4% 56.4%

P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25 21

identical across all 3 sources. There is a significant increase in reported obesity between 2001 and 2003, but between 2003 and 2015 there has been no change in the prevalence of obesity. The trends by gender and race/ethnicity in these data (not shown) are similar to the overall trend, with no significant change in obesity between 2003 and 2015. Obesity levels are different across groups in the YRBSS and NHIS data in ways that align with the overall NHANES trends, with boys reporting higher obesity prevalence than girls, and blacks and Hispanics reporting higher obesity than whites, but no statistical change in obesity prevalence within any race or ethnic group during this time period (see Appendix Fig. 4).

Turning to the other end of the age distribution, we analyze a very large sample of low-income children ages 2 to 4 who participate in the WIC nutrition program. Height and weight are measured directly in these data. Among the WIC sample there was an increase in measured obesity between 1996 and 2004. Between 2004 and 2010, the level remained statistically unchanged, and then we see a decline in obesity in 2012–2016 (compared to 2010, see Appendix Fig. 5). The decline in recent years is driven by a sharp decrease in obesity among boys. These results are consistent with the less precise results from the NHANES. Note that the difference in levels between data sources is driven by socio-economic status: WIC serves low-income children, and low-income children tend to have higher obesity prevalence. A fundamental concern with WIC data is that we only observe individuals who participate in WIC,

Fig. 2. Obesity Trends among Adolescents.

and that WIC participation rates have declined between 2010 and 2016—exactly the same period when we see the decline in obesity. To investigate whether sample composition is driving the change in obesity, we compare state-level changes in take-up rates to state-level changes in obesity. We conclude that changing caseload composition does not explain the observed decline in obesity, and that there has been a real decline in obesity over this timeframe among those in the WIC sample. The timing of this decrease generally lines up with the 2007 revision to the WIC food package, which reduced the amounts of some items (e.g. fruit juices, milk, cheese) and introduced fruits and vegetables, whole-grain foods, and low-fat milk.

In the WIC data, Blacks, whites and Asians have had similar prevalence of obesity since 2004, and for all 3 groups those levels are lower in 2014 than they were in 2004 (see Appendix Fig. 5).13 In these data blacks and whites have similar obesity prevalence, which is somewhat surprising since on average black children have higher obesity levels than whites. This reflects the fact that only low-income children are eligible to participate in WIC. Among whites, wealthy families have the lowest obesity, and low-income families have the highest prevalence. Hispanics, shown in blue, have also seen a decline in obesity between 2010 and 2014, though the level of obesity among Hispanics is substantially higher than among blacks and whites in the WIC data. The major disadvantage of the WIC data is that it is a selected sample, as mentioned above. The advantage of it is that it is very large. This is the only dataset where there are substantial numbers of Native Americans that allow us to observe obesity for this group. Native Americans have not experienced a decline in obesity like the other racial/ethnic groups, and their obesity levels are high: one in five Native children aged 2–4 who participate in WIC are obese.

7. Obesity as a cohort ages

Another important way to analyze obesity is by following cohorts of children as they age (Komlos et al., 2009). This allows us to better understand the timing of increases in obesity, both across cohorts and within cohorts. Identifying periods of particular increases in obesity growth can potentially allow us to target more effective interventions. We start with adolescents, investigating the YRBSS which surveys a large cross-section of adolescents every

13 Note, microdata are only available to us through 2014. We impute 2016 data using growth rates overall and by race and ethnicity from the published reports (Thorn et al., 2018,2015).

Fig. 3. Obesity by Birth Cohort: Adolescents in the YRBSS.

22 P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25

other year. Although we cannot follow individual children over time, we can calculate the prevalence of obesity for individual birth cohorts as they age. Taking data from different survey years, we calculate obesity prevalence for each birth cohort for each age 14 through 18. We show in Fig. 3 results from two birth cohorts: those born in 1985-86, and those born in 1997-98. We find statistically stable obesity prevalence across each age. In other words, the prevalence of obesity for a birth cohort when they are age 14 is the same as it is for the same cohort when they are age 18. This suggests that the increase in cohort obesity prevalence occurs at an earlier age.

To measure obesity growth at younger ages, we turn to two longitudinal datasets, the Early Childhood Longitudinal Studies, which follow two separate cohorts of kindergarten students as they progress through school. Height and weight are measured directly by the survey collectors in these datasets. With these data we can follow individual children from two birth cohorts (born around 1992 and 2005) and observe how the prevalence of obesity varies as they age. Fig. 4 shows the age-trend line for the 1992 and the 2005 birth cohorts, for the ages available in each dataset to date. Obesity increases with age for both cohorts at about the rate of 1.3 percentage points per year of age, then level off around age 11. The 2005 cohort entered kindergarten with a higher obesity prevalence than the 1992 birth cohort, and it has a slightly flatter growth line; as a result, obesity across the two cohorts converge between ages 8–10.

Fig. 4. Obesity by Birth Cohort: Children in the ECLS-K.

While the YRBSS and ECLS-K results are from large datasets, they are limited by the fact that they only follow cohorts across a relatively narrow range of ages. To follow a cohort from ages 2 to 18, we can only use the NHANES—recognizing the limitations of small sample sizes. Looking at the 1992 and 2005 birth cohorts that roughly match the ECLS-K, results (shown in Appendix Fig. 7) are consistent with the patterns shown in Figs. 3 and 4. Obesity prevalence appears to steadily increase until about age 10, and then remain stable through age 18.14 These results, coupled with those from the other data sources, suggest that reducing the rate of increase in obesity between ages 2 and 10 is particularly important, and may alter the long-run level of obesity among cohorts.

8. Heterogeneity in obesity among recent cohorts of children

The ECLS data allow us to disaggregate obesity levels and growth over time and across cohorts, and the patterns allow us to form hypotheses about the contributors to childhood obesity that can inform a future research agenda. Table 4 displays obesity prevalence overall, and by race/ethnic groups and gender, for 5 and 9-year-olds for both ECLS-K cohorts. Table 4 allows us to compare obesity prevalence across cohorts at age 5 and age 9 (columns labeled “cross-cohort differences”), and also to compare differ- ences in growth rates in obesity between age 5 and 9 for the two cohorts. As described in the literature review, one strand of the literature examines differences in obesity levels by groups and looks for patterns of differences in the environment facing children that are consistent with these patterns in the data. Differences in obesity prevalence prior to age 5 are likely to reflect differences in family, neighborhood, and pre-school environments. Differences in growth rates in obesity after school-entry age may additionally reflect differences in the school environment. In Table 4 we examine these differences in levels and growth rates overall, and for male and female children, by race and ethnicity, and analyze these changes with an eye toward the importance of school and non-school environments, and other differences in environments as children age, in determining obesity trends.

Considering the cross-cohort differences first, overall, obesity prevalence at age 5 increased by about 2 percentage points between 1997 and 2010, to 14.2 percent. Among whites, obesity levels at age 5—essentially at kindergarten entry—have remained statistically constant over this time, at 10–11 percent. On the other hand, age-5 obesity prevalence among blacks and Hispanics has increased by 4.4 and 2.2 percentage points, respectively, over this time frame—in other words, both of these groups enter kindergar- ten with higher prevalence of obesity in more recent years. This pattern suggests researchers should look for changes in the pre- kindergarten environment for these children that leads to increased body weights—for example, changes in preschool attendance, screen time, or maternal employment. At age 9, on the other hand, there has been no increase in obesity between cohorts overall, or for black or Hispanic children. However, white 9-year-olds experienced a statistically significant 1.6 percentage point decline in obesity over this timeframe.

Overall, growth in the prevalence of obesity between ages 5 and 9—from approximately kindergarten entry to third grade— declined by 34 percent (from 7.7 points to 5.1 points) across the cohorts. Changes in growth rates between ages 5 and 9 over time

14 When using the NHANES for this age analysis we allow the data to choose the optimal age for each cohort at which a break should be modeled. We then get predictions (using model (2) shown earlier) for each cohort, using that optimal age break. Because of the small sample sizes, we then average the predictions for cohorts born in 1991 and 1993 to create a “1992” cohort prediction, and then we average those for the 2004 and 2006 cohorts to create a “2005” cohort prediction.

Table 4 Obesity Prevalence at Ages 5 and 9 in Two ECLS-K Cohorts, by Race/Ethnicity and Gender.

Born~1992 Born~2005 Cross-cohort difference

Age 5 Age 9 Growth Age 5 Age 9 Growth Age 5 Age 9

Panel A: Overall Overall 0.121 0.198 0.077 0.142 0.193 0.051 0.021 �0.004 [SE] [0.002] [0.003] [0.004] [0.003] [0.004] [0.005] [0.004] [0.005] N 17104 13777 12821 12423 White 0.103 0.170 0.066 0.112 0.154 0.042 0.009 �0.016 [SE] [0.003] [0.004] [0.005] [0.004] [0.005] [0.006] [0.005] [0.006] N 9150 7858 6059 5894 Black 0.118 0.233 0.114 0.163 0.229 0.066 0.044 �0.004 [SE] [0.006] [0.010] [0.012] [0.009] [0.012] [0.014] [0.011] [0.015] N 2617 1745 1741 1332 Hispanic 0.171 0.261 0.090 0.193 0.266 0.073 0.022 0.004 [SE] [0.007] [0.009] [0.011] [0.007] [0.008] [0.010] [0.010] [0.012] N 3171 2476 3209 3420

Panel B: Males Overall 0.134 0.212 0.079 0.153 0.209 0.056 0.019 �0.003 [SE] [0.004] [0.005] [0.006] [0.004] [0.005] [0.007] [0.006] [0.007] N 8501 6984 6445 6335 White 0.116 0.178 0.062 0.118 0.162 0.044 0.002 �0.017 [SE] [0.005] [0.006] [0.008] [0.006] [0.007] [0.009] [0.008] [0.009] N 4545 4014 3036 3049 Black 0.116 0.209 0.093 0.164 0.216 0.052 0.048 0.007 [SE] [0.009] [0.014] [0.017] [0.013] [0.016] [0.020] [0.015] [0.021] N 1280 860 876 684 Hispanic 0.187 0.302 0.115 0.220 0.301 0.081 0.033 �0.001 [SE] [0.010] [0.013] [0.016] [0.010] [0.011] [0.015] [0.014] [0.017] N 1590 1255 1641 1742

Panel C: Females Overall 0.108 0.182 0.074 0.131 0.177 0.046 0.023 �0.006 [SE] [0.003] [0.005] [0.005] [0.004] [0.005] [0.006] [0.004] [0.007] N 8603 6793 6376 6088 White 0.091 0.160 0.069 0.106 0.146 0.040 0.015 �0.014 [SE] [0.004] [0.006] [0.007] [0.006] [0.007] [0.009] [0.007] [0.009] N 4605 3844 3023 2845 Black 0.120 0.255 0.135 0.161 0.242 0.082 0.040 �0.013 [SE] [0.009] [0.015] [0.017] [0.012] [0.017] [0.021] [0.015] [0.023] N 1337 885 865 648 Hispanic 0.155 0.219 0.065 0.165 0.229 0.064 0.010 0.009 [SE] [0.009] [0.012] [0.015] [0.009] [0.010] [0.014] [0.013] [0.016] N 1581 1221 1568 1678

P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25 23

may reflect differences in school policies, after-school environ- ments, or other interventions aimed at children in this age range. White children’s increase in obesity between these ages experi- enced a statistically significant drop from 6.6 points to 4.2 points. Black children’s obesity prevalence grew by 11.4 points in the earlier cohort, but statistically significantly declined to 6.6 in the most recent cohort. The prevalence of obesity among Hispanic children from age 5 to 9 increased by 9.0 points in the earlier cohort but grew only by 7.3 points in the more recent cohort, though the cross-cohort difference was not statistically significant for Hispanic children. In general, both the across-cohort comparisons at age 9, and the within cohort age 5-to-9 increases in obesity prevalence suggest that growth rates in obesity among school-age children are slowing, and in some cases the obesity prevalence has declined.

Turning to differences by gender and race/ethnicity can help shed light on whether researchers should focus on changes in the environment that may contribute to changes in obesity over time, or should examine whether there are differential impacts of changes in the environment by gender. Given that blacks and Hispanics often live in different neighborhoods and attend different schools than white children, differences in obesity by race/ethnicity may be driven by differences in these environments. However, male and female children of a given race/ethnicity live in the same families and neighborhoods, and attend the same

schools, and thus face similar environments. Differences by gender may indicate differential impacts of environmental factors, and thus differential effects of potential interventions.

Panels B and C of Table 4 present the same analyses for males and females, respectively. Males enter kindergarten with higher obesity prevalence than females, and although the cross-cohort differences are larger among girls than among boys (2.3 points vs. 1.9 points), this difference is not statistically significant. There is important variation by race, however. White boys have experi- enced very little increase in obesity at age 5 over time (0.2 points) where prevalence among white girls has increased by 1.5 percentage points, such that in the later cohort white boys and girls have statistically the same obesity prevalence at age 5. For black and Hispanic boys and girls, this cross-cohort pattern is reversed: age-5 obesity has gone up more across cohorts for boys than for girls. These patterns suggest both that the pre-school environment for black and Hispanic children has continued to change in ways that increase obesity, and that there are differential effects in the pre-school environment for boys and girls.

The patterns of growth over time across cohorts also differ by the interaction of race and gender. Overall, age-9 obesity is unchanged across the two cohorts even though age-5 prevalence increased. Although the magnitudes vary, the same pattern holds for boys and girls across all races. Statistically, no group has a different obesity prevalence at age 9 in the later cohort compared

24 P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25

with the earlier cohort, even though white, black and Hispanic girls, and black boys, all entered kindergarten with higher prevalence of obesity.

For both boys and girls and for most gender-by-race cells, the age 5 to 9 growth in obesity statistically significantly declined across the cohorts. For white boys and girls, this decline was by 1.9 and 2.9 points, respectively, or about 30% for white boys and 42% for white girls. For black boys, the decline was 4.1 points (44 percent) and for black girls it was 5.3 points (40 percent). Hispanic boys had a decline in obesity growth from age 5 to 9 of about 3.3 points (29 percent), with Hispanic girls showing no change in the 5 to 9 growth in obesity across cohorts. These patterns are consistent with there being improvements in the school-age environment across all racial groups, but the magnitude of these improvements have some heterogeneous effects by gender.

Since obesity is calculated by examining whether BMI is above the age-sex specific cutoff for obesity, it may be the case that obesity may stabilize even when average BMI increases, depending on whether that increase in BMI pushes more children above the threshold for obesity. In separate analyses (shown in Appendix Table 2), we find that average BMI has increased across race and gender groups for 5-year-olds, though the age 5 to 9 growth in BMI is lower for all groups. Research has also shown that the distribution of BMI has become more unequal, with BMI among the obese pulling away from the mean (Butcher and Park, 2008). We find (not shown, available on request) average BMI among the obese is high. For example, among obese 9-year-olds overall in the later cohort, BMI is 25, which is the threshold for overweight in the adult BMI distribution. The patterns of changes within and between cohorts is similar across gender and race groups, with the notable exception of black boys. For all groups except black boys, BMI among the obese is higher at age 5 and age 9 across the cohorts. However, BMI growth between age 5 and age 9 within a cohort fell from the first wave to the second. For black boys, BMI growth among the obese between ages 5 and 9 was about 16 percent higher in the second cohort. The patterns of BMI overall and among the obese are broadly consistent with the patterns of obesity. Taken together, the evidence indicates there have been recent improvements in the rate of body weight increase during the school-age years.

9. Summary and discussion

In this paper we examine multiple data sets to study the latest trends, both over time and by age cohort, in childhood obesity by race and gender in the United States. While childhood obesity continues to be prevalent, there is some evidence that the rate of increase has leveled off in recent years, especially among school- age children. Our analyses by age cohort indicate a general pattern of obesity prevalence that increases throughout early childhood and levels off by around age 10. Our analysis of the ECLS-K indicates that obesity in early childhood rose between the 1992 and the 2005 birth cohort, as indicated by obesity at age 5 when children enter kindergarten. On the other hand, there is moderately good news on obesity for school aged children: Obesity prevalence at age 9 is stable or has fallen slightly across the two birth cohorts, and the growth in obesity between ages 5 and 9 was slower in the later cohort. These findings mask heterogeneity by gender within race. White boys have experienced the smallest gains in age-5 obesity and have gains that are smaller than those of white girls. Black and Hispanic boys have experienced the largest age-5 gains in obesity, and their gains were larger than those among black and Hispanic girls. All groups have experienced a decline in the rate of growth in obesity from age 5 to age 9.

These patterns are intriguing and suggest a path for future research. It appears that environments faced by children younger

than Kindergarten have become increasingly obesogenic, particu- larly for black and Hispanic boys. However, the school-aged outcomes indicate that the rate of growth in obesity has slowed. Some of these changes differ by gender, suggesting that it is not the environment that faces school-aged children per se that has changed—since boys and girls are in the same schools, families, and neighborhoods—but instead whatever changes may have hap- pened in the environment have had heterogenous effects. Understanding what effective interventions are during early childhood, and understanding the potential heterogenous effects of interventions during both early childhood and for school-age children remain of paramount importance.

Acknowledgement

Support for this article was provided by the Robert Wood Johnson Foundation. The views expressed here do not necessarily reflect the views of the Foundation.

Appendix A. Supplementary data

Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.ehb.2019.02.002.

References

Alviola, Pedro A., Nayga Jr., Rodolfo M., Thomsen, Michael R., Danforth, Diana, Smartt, James, 2014. The effect of fast-food restaurants and childhood obesity: a school level analysis. Econ. Hum. Biol. 12 (January), 110–119.

Anderson, Patricia, 2012. Parental employment, family routines and childhood obesity. Econ. Hum. Biol. 10 (December (4)), 340–351.

Anderson, Patricia M., Butcher, Kristin F., 2006a. Childhood obesity: trends and potential causes. Future Child. 16 (Spring (1)), 19–45.

Anderson, Patricia M., Butcher, Kristin F., 2006b. Reading, writing, and refreshments are school finances contributing to children’s obesity? J. Hum. Resour. 41 (3), 467–494.

Anderson, Patricia M., Butcher, Kristin F., Levine, Phillip B., 2003. Maternal employment and overweight children. J. Health Econ. 22 (3), 477–504.

Anderson, Patricia, Butcher, Kristin, Schanzenbach, Diane W., 2009. Child disadvantage and obesity: is nurture trumping nature? In: Gruber, Jonathan (Ed.), The Problems of Disadvantaged Youth: An Economic Perspective. University of Chicago Press, Chicago.

Belfield, Clive R., Kelly, Inas Rashad, 2013. Early education and health outcomes of a 2001 US birth cohort. Econ. Hum. Biol. 11 (3), 310–325.

Bowling, April, et al., 2017. ADHD medication, dietary patterns, physical activity, and BMI in children: a longitudinal analysis of the ECLS-K study. Obesity 25 (10), 1802–1808.

Butcher, Kristin F., Park, Kyung H., 2008. Obesity, disability, and the labor force. Econ. Perspect. (Federal Reserve Bank of Chicago) 2–16 First Quarter.

Cascio, Elizabeth U., Schanzenbach, Diane Whitmore, 2013. The impacts of expanding access to high-quality preschool education. Brookings Papers on Economic Activity, , pp. 127.

Cawley, John, 2010. The economics of childhood obesity. Health Aff. 29 (March (3)), 364–371.

Cawley, John, Frisvold, David, Meyerhoefer, Chad, 2013. The impact of physical education on obesity among elementary school children. J. Health Econ. 32 (4), 743–755.

Classen, Timothy, Hokayem, Charles, 2005. Childhood influences on youth obesity. Econ. Hum. Biol. 3 (July (2)), 165–187.

Cutler, David M., Glaeser, Edward L., Shapiro, Jesse M., 2003. Why have americans become more obese? J. Econ. Perspect. v17 (3, Summer), 93–118.

Frisvold, David E., Lumeng, Julie C., 2011. Expanding exposure: can increasing the daily duration of Head Start reduce childhood obesity? J. Hum. Resour. 46 (2), 373–402.

Green, Tiffany L., 2014. Examining the temporal relationship between childhood obesity and asthma. Econ. Hum. Biol. 14 (July), 92–102.

Hales, C.M., Carroll, M.D., Fryar, C.D., Ogden, C.L., 2017. Prevalence of Obesity Among Adults and Youth: United States, 2015–2016. NCHS Data Brief, No 288. National Center for Health Statistics, Hyattsville, MD.

Herbst, Chris M., Tekin, Erdal, 2011. Child care subsidies and childhood obesity. Rev. Econ. Househ. 9 (3), 349–378.

Jo, Young, 2014. What money can buy: family income and childhood obesity. Econ. Hum. Biol. 15 (December), 1–12.

Komlos, John, Breitfelder, Ariane, Sunder, Marco, 2009. The transition to post- industrial BMI values among U.S. children. Am. J. Hum. Biol. 21, 151–160.

Lakdawalla, Darius, Philipson, Tomas, 2009. The growth of obesity and technological change. Econ. Hum. Biol. 7 (December (3)), 283–293 Elsevier.

NCD Risk Factor Collaboration, 2017. Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of

P.M. Anderson et al. / Economics and Human Biology 34 (2019) 16–25 25

2416 population-based measurement studies in 128.9 million children, adolescents, and adults. Lancet 390, 2627–2642.

Sallis, James F., Glanz, Karen, 2006. The role of built environments in physical activity, eating, and obesity in childhood. Future Child. 89–108.

Sandy, Robert, et al., 2013. Effects of the built environment on childhood obesity: the case of urban recreational trails and crime. Econ. Hum. Biol. 11 (1), 18–29.

Schanzenbach, Diane Whitmore, 2009. Do school lunches contribute to childhood obesity? J. Hum. Resour. 44 (3), 684–709.

Thorn, Betsy, Tadler, Chrystine, Huret, Nicole, Ayo, Elaine, Trippe, Carole, Mendelson, Michele, 2015. Kelly Lawrence Patlan, Gabriel Schwartz and Vinh

Tran. “WIC Participant and Program Characteristics 2015. Prepared by Insight Policy Research Under Contrace No. AG-3198-C-11-0010. U.S. Department of Agriculture, Food and Nutrition Service, Alexandria, VA November.

Thorn, Betsy, Kline, Nicole, Tadler, Chrystine, Budge, Eric, Wilcox-Cook, Elaine, Michaels, Jason, Mendelson, Michele, Patlan, Kelly, Tran, Vinh, 2018. WIC Participant and Program Characteristics 2016. U.S. Department of Agriculture, Food and Nutrition Service, Alexandria, VA April.

U.S. Department of Health and Human Services, 2001. The Surgeon General’s Call to Action to Prevent and Decrease Overweight and Obesity. U.S. Department of Health and Human Services, Public Health Service, Office of the Surgeon General, Rockville, MD Available from: U.S. GPO, Washington.