The impact of state-level nutrition-education program funding on BMI: Evidence from the behavioral risk factor surveillance system

Social Science & Medicine 82 (2013) 67e78

Contents lists available at SciVerse ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

The impact of state-level nutrition-education program funding on BMI: Evidence from the behavioral risk factor surveillance system Kerry Anne McGeary a, b, c, * a

Department of Economics, Miller College of Business, Ball State University, Muncie, IN 47306, USA Global Health Institute, Ball State University, Muncie, IN 47306, USA c National Bureau of Economic Research, Cambridge, MA 02138, USA b

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 30 January 2013

Currently, there is insufficient evidence regarding which policies will improve nutrition, reduce BMI levels and the prevalence of obesity and overweight nationwide. This preliminary study investigates the impact of a nutrition-education policy relative to price policy as a means to reduce BMI in the United States (US). Model estimations use pooled cross-sectional data at the individual-level from the Centers for Disease Control’s (CDC), Behavioral Risk Factor Surveillance System (BRFSS), state-level food prices from the American Chamber of Commerce Research Association (ACCRA) and funding for state-specific nutrition-education programs from the United States Department of Agriculture (USDA) from 1992 to 2006. The total number of observations for the study is 2,249,713 over 15 years. During this period, federal funding for state-specific nutrition-education programs rose from approximately $660 thousand for seven states to nearly $248 million for all fifty-two states. In 2011, federal funding for nutritioneducation programs reached $375 million. After controlling for state-fixed effects, year effects and state specific linear and quadratic time trends, we find that nutrition education spending has the intended effect on BMI, obese and overweight in aggregate. However, we find heterogeneity as individuals from certain, but not all, income and education levels respond to nutrition-education funding. The results regarding nutrition-education programs suggest that large scale funding of nutrition-education programs may improve BMI levels and reduce obesity and overweight. However, more study is required to determine if these funds are able make the requisite dietary improvements that may ultimately improve BMI for individuals from low income and education-levels. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: United States Obesity Nutrition education Price policy Heterogeneity Endogenous effects

Introduction The average US Body Mass Index (BMI), defined as weight in kilograms divided by height in meters squared (kg/m2), has increased dramatically since the 1980s. Due to this rise, more Americans are classified as overweight (BMI > 25 kg/m2) and obese (BMI > 30 kg/ m2). This increase has serious implications for the state of national health. A substantial amount of cross-disciplinary research has investigated the potential reasons for the increase. The economics literature attributes the dramatic increase in average BMI to economic changes that alter Americans’ preferences for exercise and high calorie, low-nutrient food and drinks (Cawley, 1999; Chou, Grossman, & Saffer, 2004; Cutler, Glaeser and Shapiro, 2003; Lakdawalla & Philipson, 2002; Philipson, 2001). * Department of Economics, Miller College of Business, Ball, State University, Whitinger Building 129, 2000 W. University Ave., Muncie, IN 47306, USA. Tel.: þ1 765 285 4440; fax: þ1 765 285 4313. E-mail address: [email protected]. 0277-9536/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2013.01.023

To incentivize proper nutrition, economic policies typically suggest altering the price of and/or access to low-priced foods that offer lower nutritional content. However, invoking policies affecting access or price may result in regressive effects that disproportionately and negatively impact those who are from a lower socioeconomic status without resulting in change. For this reason, other strategies that directly target consumer information may be more effective in reducing demand without imposing regressive losses on the most economically vulnerable. This paper offers a preliminary study of the impact of federal spending on nutrition-education programs on three commonly examined weight outcomes: BMI, obesity and overweight. Despite a wealth of research on nutrition-education programs from other disciplines, such as public health and nutrition, the current literature fails to provide an aggregate economic analysis that compares different policies and the policies overall effects on BMI, obesity and overweight. The goal of this paper is to evaluate the impact of using nutrition-education programs to reduce individual-level BMI, obesity, and overweight, and to determine

68

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

this impact relative to price changes. To achieve this goal, we use over 2.2 million individual-level observations from the 1992e 2006 waves of the Behavioral Risk Factor Surveillance System (BRFSS) from the CDC. The individual-level BRFSS data are matched by state and year with: 1) real cumulative funding for nutrition-education programs (per million) and 2) real food prices. While the use of real food prices are not controversial, the use of real cumulative funding warrants some explanation. We argue that the dynamic nature of obesity, e.g., the fact that it results from a cumulative increase in BMI, demands the use of real cumulative funding when compared to contemporaneous real funding e levels. In fact, we find that increases in real cumulative nutritioneducation funding is consistent with improving BMI and reducing obese or overweight among individuals from certain, not all, income and education-levels. Estimating the precise impact of nutrition-education programs at an individual-level is ambitious with these data. However, much can be gained from analyzing these data to understand the effects over time and across groups. The analysis is complicated by concerns regarding structural endogeneity, e.g., the bias introduced by unobserved influences. For example state-level “nutrition awareness”, attitudes toward “health behavior improvement”, or policy response to increasing average weight trends, may confound the estimation. The empirical model employed here explicitly controls for such unobserved, and potentially biasing, effects through the use of year fixed effects, state-fixed effects, state linear-time trends and quadratic time trends. Our strategy will mitigate the impact from attitudes or otherwise unobserved state-level variation, time variation and state-specific variation over time is included in the model. Therefore, this paper contributes to the literature in several substantial ways. First, it provides an aggregate investigation of the impact of nutrition-education policy on individual-level weight outcomes. Second, the study assesses the impact of nutritioneducation policy while controlling for price variations (across states and overtime). Third, the study controls for unobserved variation or attitudes that are specific to each state and year, in addition to all unobserved variation at the state-level that changes over time and at differing rates. To our knowledge, this is the first study in any discipline to investigate the impact of nutrition-education policy across states and time while controlling for other important economic factors such as price and income. Extant studies that investigate price policy have done so using state-fixed effects to control for state-level attitudes and strategies to combat obesity, at best. Again, this paper improves on previous work by holding statelevel policies and expenditures to combat obesity constant in models that use state-fixed effects, year fixed effects and statespecific time trends. Background

Data from the BRFSS provide additional evidence of this nationwide increase in the average BMI. We acknowledge that BMI is not a perfect measure of overall fatness or health of an individual. However, given the current state of the literature we selected BMI as the most widely accepted measure of obesity (Burkhauser & Cawley, 2008). Fig. 1 demonstrates the increase in the average BMI for the US population between 1992 and 2006. The trend and its persistence is a dilemma faced by policy makers given that the average BMI is approaching the obese range. This increase in the average BMI has continued even after the problems associated with increasing BMI and obesity were identified and while expenditures on weight-loss services and products designed to aid weight reduction, including medical procedures and pharmaceuticals has risen dramatically (Reuters, April 21, 2009). Associated medical complications and costs The pool of literature documenting the costs associated with obesity has increased since the mid-1990s. The importance and salience of the research derives from the well-documented adverse health outcomes associated with obesity and overweight (Allison, Fontaine, Manson, Stevens, & VanItallie, 1999; McGinnis & Foege, 1993). Obesity and overweight are currently associated with an increased risk of coronary heart disease, type-2 diabetes, certain cancers (endometrial, breast, and colon), hypertension, high cholesterol, stroke, liver and gallbladder disease, sleep apnea and respiratory problems, osteoarthritis, and gynecological problems (Kahn et al., 1997). Given the well-documented complications associated with obesity, the literature has attempted to determine the costs associated with these complications and obesity in general. An often cited study from 2009 estimates the aggregate medical costs associated with obesity to be $147 billion (2008 dollars) using the Medical Expenditure Panel Survey (MEPS) (Finkelstein, Trogden, Cohen, & Dietz, 2009). This study further estimates the annual medical expenditures for an obese person to be $1429 (2008 dollars) greater than the expenditures for individuals with a BMI in a normal range (Finkelstein et al., 2009). Citing concerns regarding endogeneity, Cawley and Meyerhofer (2012) use same MEPS data and correct for endogeneity using an instrumental variables technique yielding a higher estimate of $2571 per year in additional annual medical costs for an obese individual (Cawley & Meyerhofer, 2012). Table 1 presents the estimates for aggregate spending on obesity from two studies. Finkelstein, Fiebelkorn, and Wang (2003) estimates the aggregate spending attributed to obesity to be between $26.8 and $47.5 billion dollars per year (Finkelstein et al., 2003). Meanwhile, later Cawley and Meyerhofer (2012) estimates total medical expenditures from obesity to be $26.6 without accounting for the Medicare population. A point to note from Table 1 is obesity places a significant burden on publicly-funded health insurance

BMI, obesity and overweight The dramatic increase in prevalence of overweight and obesity among individuals 18 years of age and older has been a concern. The Centers for Disease Control (CDC) monitors changes in obesity and overweight using data from the National Health and Nutrition Examination Survey (NHANES), collected at various time intervals, and the Behavioral Risk Factor Surveillance System (BRFSS), collected annually. The NHANES data estimate a 19% increase in overweight and 55% increase in obesity over the roughly 15 - year period between NHANESII (1976e1980) and NHANESIII (1988e 1994). The obesity prevalence continued to rise an additional 42% over the 15 years from NHANESIII (1988e1994) to NHANES (2003e 2004), while the prevalence of overweight remained fairly constant, at 18%, over that same time period (CDC, 2006).

Fig. 1. National prevalence of overweight and obese.

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

69

Table 1 Aggregate medical spending, in billions of dollars, attributable to obesity, by insurance status and data source, 1996e1998 and 2000e2005. Insurance Category

Total Expenditures Private Medicaid Medicare Total

Obesity

Obesity

Source: Finkelstein et al. (2003)

Source: Cawley and Meyerhofer (2012)

MEPS (1998)a

NHA (1998)

$26.8 $9.5 $2.7 $10.8 $26.8

$47.5 $16.1 $10.7 $13.8 $47.5

a

MEPS (2000e2005)b Third Party

$26.6 23.2 N/A

a Calculations based on data from the 1998 Medical Expenditure Panel Survey (MEPS) and health care expenditures data from National Health Accounts (NHA). Source: Finkelstein et al. (2003). b Calculations based on data from the 2000e2005 Medical Expenditure Panel Survey (MEPS) using and instrumental variables technique on a sample of non-elderly adults with biologic children. Source: Cawley and Meyerhofer (2012).

programs, e.g. Medicaid and Medicare. In 1998, approximately 42% of the obesity-related costs were publicly financed, as described by Table 1. In fact, the public funding of overweight and obesity rivals the amount of public funding associated with smoking (National Center for Tobacco Free Kids, 2002). Policies to combat obesity Given the high costs associated with obesity, many disciplines have focused their respective literature on determining the causes of obesity as they seek to ultimately mitigate the trend, as well as, the costs associated with obesity. As mentioned, the literature from economics suggests evolving technological change is correlated with a sustained increase in calorie consumption relative to calorie expenditure. This finding has lead the economics literature to focus on price policy, or taxation, as a tool to effect individual-level behavior such that individuals would move consumption away from high-calorie food with low levels of nutrition (Auld & Powell, 2008; Beyoun, Powell, Yuan, 2008; Cawley, 1999; Lakdawalla, & Philipson, 2002; Chou et al., 2004; Chou, Rashad, & Grossman, 2008; Fletcher, Frisvold, Tefft, 2010; Goldman, Lakdawalla, Zheng, 2009; Lakdawalla & Philipson, 2006; Rashad, Grossman, Chou, 2006; Philipson, 2001). An under-studied mechanism for change is nutritioneducation policy. Nutrition-education policy could be a successful means to reduce individual-level BMI, particularly among higher income, more highly-educated consumers. Nutritioneducation programs that effectively provide training or educational programs with a targeted purpose should reduce BMI and the probability of an individual being overweight or obese. This outcome is consistent with research that demonstrates how training improves production efficiency (Barron, Black, & Lowenstein, 1989; Mincer, 1984). Nutrition-education is, also, consistent with the ideas surrounding improvements in production efficiency. For instance, being able to understand the cost and benefits of choices that influence the energy balance is essential to reversing the growth in BMI. The well cited Grossman (1972) is the first to show that education may improve health production. Other studies have further tested the theory that education can enhance health production. Lleras-Muney and Lichtenberg (2006) find that more highly-educated people are quicker to respond to new health-related drug information and adopt drugs recently approved by the FDA. In addition, Price and Simon (2009) find more educated women experience the largest change in treatment in response to new medical information. Economists have found that spending on education programs can be a successful means to induce other healthy behaviors. For example, the literature on anti-smoking campaigns has found that cigarette sales and anti-smoking campaign funding are inversely related (Farrelly, Pechacek, & Chaloupka, 2003).

The investigation of nutrition-education programs in the economics literature is sparse. However, clinical studies have found nutrition-education to be effective in reducing the risk of chronic diseases such as heart disease, diabetes and obesity (Dansinger, 2005; Lasker, Evans, & Layman, 2008). In addition, some targeted and specialized research has demonstrated the effectiveness of nutrition-education programs in small clinical trials (Connelly, Duaso, & Butler, 2007). More specifically, other studies have examined the impact of nutrition education on college students, finding a positive effect on nutrition and weight reduction (Byrd-Williams, Storther, Kelly, & Huang, 2009; Clifford, Anderson, Auld, & Champ, 2009). While these studies provide suggestive evidence of an association between nutritioneducation programs and weight reduction, these studies use cross-sectional designs and are not state-level studies. These small scale studies are unable to measure differences across states, over time, and, more importantly, across nutritioneducation programs that could contribute both to the level of nutrition-education funding and BMI levels. In comparison an aggregate study is more comprehensive and less targeted than smaller scale studies. Nonetheless, given the comprehensive nature of an aggregate study it is important and may produce different results. Background on the federal nutrition-education program Funding: Federal and non-federal public nutrition programs Federally, nutrition and obesity education is delivered through the United States Department of Agriculture (USDA). The USDA relies on state-level public entities, public schools (school systems and universities), public health clinics, etc. to extend information to the population. Funding for these programs is distributed to each funded state through grants to each state’s health department in cooperation with the state's land grant university. The grants are capped at a dollar-per-dollar federal match to all non-federal public funds allocated at the state-level for the delivery of nutrition-education programs. Therefore, states devoting more internal funds to nutrition-education receive the largest amount of funds from the federal government. The total amount of state-level funding represents a total of all public resources allocated to improvements in weight outcomes. The USDA started allocating funds in 1988 with a grant to the State of Wisconsin through the University of Wisconsin at Madison. To our knowledge, the first year of available nutrition-education funding data is 1992. By 1992, participation had increased via land grant institutions in seven states: Wisconsin, Minnesota, Ohio, New Hampshire, New York, Oklahoma, and Washington. Program participation rose to all fifty states, by 2004. State-level programs vary substantially but maintain their focus on the following key behavioral outcomes to achieve reductions in

70

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

the average BMI state-wide and reduce the prevalence of obesity at the state-level: 1. Balance caloric intake from food and beverages with calories expended. 2. Be physically active every day. 3. Eat fruits and vegetables, whole grains, and fat-free or low-fat milk products daily. The types of interventions include wide spread social marketing campaigns (indirect intervention), and the provision of a comprehensive nutrition-education curriculum to an individual or group (direct intervention). Despite the fairly uniform use of learnercentered and behavioral-focused interventions overall, the variability in delivery and intensity of delivery may result in somewhat diffuse treatment effects. For example, one representative state, Indiana, offered direct education programs in the following settings: local cooperative extension offices from its land grant institution, emergency food assistance sites, elderly services site, WIC clinics, adult education and training sites and churches/other faith based organization sites. The indirect materials used by programs in the state of Indiana included widely disseminated print material and participation in public events. Given the focus of Indiana’s program, 90% of the participants are adults (18 years old or older). In comparison, Michigan, one of the first recipients of the funds, offers a broader program that targets both direct and indirect education through multiple venues. The direct education sites include all education venues previously listed for Indiana plus the following additional sites: public schools (K-12), food stores or other retail outlets and health care sites. The indirect education includes mass communication (radio, television, newspapers, and posters), print materials and participation in public events. The demographic profile of Michigan’s program is estimated to be 52% school aged children (5e17 years old) and infants (less than 5 years old) and 48% adults (18 years old or older) (USDA, 2004). Unfortunately, at the time of this writing data were restricted to 2004. Data The Behavioral Risk Factor Surveillance System (BRFSS) Three individual-level weight outcomes are investigated: 1) actual BMI, 2) an indicator the individual’s BMI is above the normal range (overweight) (BMI  25 kg/m2), and 3) an indicator that an individual’s BMI level is in the obese range (obese) (BMI  30 kg/ m2). The individual-level variables and all demographic variables used to describe the three weight outcomes derive from repeated cross-sections of the BRFSS, an annual telephone survey of persons 18 years and older. Surveys are conducted by each participating state’s department of health and administered by the Centers for Disease Control and Prevention (CDC). The CDC uses BRFSS to calculate national and state-specific estimates of the prevalence of lifestyles and behaviors that contribute to various health outcomes, including obesity. Fifteen states participated in the first survey in 1984. State participation rose to 49 states by 1992 and all 50 states by 1996 (we do not include Guam or the US Virgin Islands). During this time, the respondent participation in the BRFSS rose from 96,213 in 1992 to 355,216 in 2006. Given the growth in respondent participation in BRFSS, our final analysis sample for the pooled cross-section from 1992 to 2006 includes 2,249,713 individuals nationwide. This study was completed in 2012, using data that are publicly available. We should note that the primary data collection effort was approved by an appropriate institutional review board so no additional ethical approval was necessary.

As demonstrated in Fig. 1, during the same time-frame the US average BMI increased by 2.10 kg/m2, or approximately 7%. Fig. 1 illustrates the need for focus on another concern. Underlying the growth is obesity is growth in the prevalence of overweight within the US. The growth in obesity is fed by those transitioning from a BMI in the overweight range to a BMI in the obese range. Therefore, our consideration of prevalence overweight allows for a more accurate picture of the “fattening” of the US population. Over the time frame of study, the prevalence of overweight in the BRFSS increased from 1992 to 2006 by nearly 35%, see Fig. 1. Table 2 shows the means and standard deviations for the nationally representative control variables and dependent variables that will be used in our analysis. The variables include measures of age, gender, marital status, education-level, employment and income-levels. The income-levels are used to create a real income variable (in 1984 dollars) for consistency with the other variables that measure prices and funding. The data definitions, means and standard deviations of all individual-level are reported in Table 2. The statistics are based on a sample of 2,249,713 individuals. Finally, in all regressions the BRFSS survey weights and the standard errors are corrected for heteroskedasticity and clustering within states. Data on state-level nutrition-education funding Recall, these funds represent the federal match to public nonfederal funds allocated by the states for nutrition-education. To receive the federal match states must take the initiative to secure federal funds. Therefore, the estimated empirical models will identify the effect of funding from the individuals who are treated due to their state’s funding compared to the associated outcomes for individuals who are untreated due to the timing of the funding adoption, and the level of the funding within each state. This will be discussed further in the identification section. Further, the trend in the total state-level real nutritioneducation funding (nutrition-education funding from here forward) by year is depicted in Fig. 2. In 1992, only 7 states (Minnesota, Wisconsin, Ohio, New Hampshire, New York, Oklahoma, and Washington) were approved for a total of $661,076 in federal funding. In 1995, the number of states receiving funding jumped to 18 states, for a total of $10.28 million. Than between 1995 and 1996 the number of participating states nearly doubled from 18 to 35 states with a threefold increase in funding. As the data in Fig. 2 represent both participation and amount of funding, they illustrate the increase in participation in as well as the increase in financial commitment to nutrition-education between 1992 and 2006. In addition, Table 1, reports the average cumulative level funding per million people over the same time span. Food price data An additional policy variable of interest is food price. An analytical model with individual weight as the outcome is, in effect, a reduced-form weight or health production function. Therefore, the model must include a measure of input prices (i.e., food prices). Our models use state-level food price indices, classified by whether food is purchased for preparation and consumption at home (home food price) or purchased away from home and already prepared (away food price). Price data is from the ACCRA Cost of Living Index, which is published quarterly by the American Chamber of Commerce Research Association (ACCRA) for 250e300 urban centers. The index is widely used by economists, researchers and corporations to measure relative cost of living. Quarterly state-specific population-weighted city prices in a given year were averaged to get annual prices.

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

71

Table 2 Definitions, means and standard deviations of the variables. Variable BRFSS BMI Obese Overweight Age Male White Black Hispanic Married Divorced Widowed High School Some College or more Employed Income < $10,000 $10,000 < Inc. <¼ $15,000 < Inc. <¼ $20,000 < Inc. <¼ $25,000 < Inc. <¼ $35,000 < Inc. <¼ $50,000 < Inc. <¼ Inc. > $75,000 Real Income

$15,000 $20,000 $25,000 $35,000 $50,000 $75,000

Definition

Mean

Standard deviation

Weight in kg/height in meters2 Dichotomous Variable ¼ 1 if BMI in the obese Dichotomous Variable ¼ 1 if BMI in the overweight range (obese or overweight) The respondent’s age at the time of the survey. The respondent is a male. Dichotomous Variable ¼ 1 if the respondent is White but not Hispanic Dichotomous Variable ¼ 1 if the respondent is Black but not Hispanic Dichotomous Variable ¼ 1 if the respondent is Hispanic Dichotomous Variable ¼ 1 if the respondent is married Dichotomous Variable ¼ 1 if the respondent is divorced Dichotomous Variable ¼ 1 if the respondent is widowed Dichotomous Variable ¼ 1 if the respondent exactly 12 years of school Dichotomous Variable ¼ 1 if the respondent completed at least 1 year of higher education/vocational school Dichotomous Variable ¼ 1 if the respondent is employed Dichotomous Variable ¼ 1 if the respondent’s income is less than $10,000 Dichotomous Variable ¼ 1 if respondent’s income is $10,000 - $15,000 Dichotomous Variable ¼ 1 if respondent’s income is $15,000 - $20,000 Dichotomous Variable ¼ 1 if respondent’s income is $20,000 - $25,000 Dichotomous Variable ¼ 1 if respondent’s income is $25,000 - $35,000 Dichotomous Variable ¼ 1 if respondent’s income is $35,000 - $50,000 Dichotomous Variable ¼ 1 if respondent’s income is $50,000 - $75,000 Dichotomous Variable ¼ 1 if respondent’s income is greater than $75,000 Real income in 1984 dollars, taken from the midpoint of the income categories in thousands (reported above)

26.646 0.216 0.577 47.845 0.421 0.813 0.079 0.058 0.553 0.142 0.101 0.311 0.585

5.483 0.412 0.494 15.547 0.494 0.390 0.270 0.234 0.497 0.349 0.301 0.463 0.493

0.634 0.070 0.068 0.087 0.106 0.156 0.181 0.161 0.173 24.874

0.482 0.255 0.281 0.308 0.362 0.363 0.385 0.367 0.377 15.040

ACCRA Home Food Price Away Food Price

Real food at home price in respondent’s state in 1984 dollars Real food away price in respondent’s state in 1984 dollars

1.158 2.298

0.108 0.145

Nutrition-Education Funding Real Cumulative Nutrition-Education Funding

Cumulative SNAP-ed funds in 1984 dollars per million (Averaged overall states and years)

7.608

19.064

Fig. 2. Total Nutrition Education Program Funding & Participation by Year ($/million).

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

72

The state-level home food price index is calculated as a population-weighted average of the nominal prices for all items included in the home food index for each urban area in a state. The away food price is created similarly. When calculating the away food price and home food price, only food items that are available in the ACCRA consistently from 1992 to 2006 are included. Food prices are deflated by the CPI to 1984 prices (real home food price and real away food price). The averages for the real home food price and the real away food price are provided in Table 2.

from home consumption (Away Food Price) and food purchased for at home consumption (Home Food Price). l is a vector of state indicators, and s is a vector of year indicators. As described previously, b2, b3, and b4 are the coefficients of interest. Identification of the parameters of interest in Eq. (1) is achieved from within state variation for those states with funding relative to states do not receive funding.

Empirical implementation

Despite having both year and state-fixed effects, estimation of Eq. (1) does not allow us to conclude that the association between state funding of nutrition expenditures and individual weight outcomes reflects a causal relationship. To the extent that structural endogeneity remains, there is a concern that changes in nutritioneducation expenditures are driven by 1) trends in obesity rates, or 2) unmeasured changes in the attitudes toward obesity by policy makers and the public. If this problem exists, we cannot interpret an observed association between expenditures and weight outcomes as evidence of effective policies as modeled by Eq. (1). Rather, the association may reflect confounding by unmeasured a reverse causal relationship or state trends with obesity rates driving policy making. We employ a few strategies to address these concerns. The most straightforward way to deal with potential confounding by unmeasured state trends is to control for trends at the state-level. Given the timeframe under consideration, 15 years, we have an ample number of years to collect any underlying state-level trends. Also, the large number of observations, over 2 million, should assuage any concerns regarding the loss of degrees of freedom. Therefore, the first strategy we utilize is to extend Eq. (1) by including state-specific linear, and quadratic time trends, as described by Eq. (2) below.

Inclusion of state-specific time trends

In order to estimate the effect of food prices and nutritioneducation funding on weight outcomes, we begin by employing a methodology that estimates the effect of prices as well as cumulative real nutrition-education funding per million. The most convincing state-level program evaluations have a well-defined control and treatment group. Typically, they measure the effect of a discrete public policy on a population affected and a population not affected in the treatment state with the same two populations in an appropriate control state. We offer a model that has been used in economic studies to investigate the response of tobacco consumption to changes in price and antismoking campaign funding has been modeled in the past. Examples include Chaloupka and Warner (2000), which provides an extensive review of the cigarette price and tax literature, Farrelly et al. (2003), which examines the anti-smoking campaign literature, and Ciecierski, Chatterji, Chaloupka, & Wechsler, (2011) which examines state expenditures on tobacco control programs. Our models, described below, offer a series of state and year fixed effects. These controls by state and year mimic the treatment and control groups necessary to identify our model. All models discussed in this section are estimated with Huber-White standard errors adjusted for clustering at the state-level. Linear probability models are used to estimate the effects of cumulative real nutrition-education funding on two individuallevel binary indicator variables: overweight and obese.1 It is argued that cumulative funds are a better measure than the funds per year since the funds and weight are both cumulative, dynamic processes. Work by Farrelly et al. (2003) has shown, for tobacco control, variation in cumulative expenditure overtime is the relevant variable of interest rather than contemporaneous measures. To estimate the continuous outcome, BMI, a linear model is used. The following baseline estimating equation is offered,

Yist ¼ ß0 þß1 Xist þß2

T X t¼1992

Yist ¼ b0 þ b1 Xist ! PT t¼1992 Real Nutrition  Education Funding þ b2 million

st

þ b3 ðReal Away Food PriceÞst þ b4 ðReal Home Food PriceÞst þ ls þ st þ ls s þ ls s2 þ εist (2)

!! Real Nutrition Education Funding

! þß3 ðReal Away Food PriceÞst

=Million st

þß4 ðReal Home Food PriceÞst þ ls þ st þεist

where i indexes the individual, s indexes the state, and t indexes the year. Yist is one of two individual-level indicators for the binary outcomes of having a BMI in the overweight or obese range. For the continuous model, Yist is individual-level BMI. Xist is a vector of individual demographic characteristics, including binary variables for male, white, black, Hispanic, high-school graduate, some college or more, and continuous variables for real income, age and age squared. Real Away Food Price and Real Home Food Price are the state and year specific real price indices for food purchased for away

(1)

Therefore, controls for state-specific linear-time trends (lss) and quadratic time trends (lss2), will control for all state-level trends that may be correlated with policy. Adding both linear and quadratic time trends will allow for both linearly (slow) and quadratically (fast) moving unobserved trends. This strategy has been employed by others to successfully control for unmeasured state trends using the BRFSS data (Gruber & Frakes, 2006). Next, in addition to examining the effect of the current cumulative nutrition-education expenditures we re-estimate all models with a one year and two year lags of these expenditures as well

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

as one and two year lags of prices as the independent variables of interest. Using lagged measures helps to further mitigate the problem of structural endogeneity that may occur if we try to link current expenditures to current outcomes. As has been shown with anti-tobacco campaign expenditures, nutrition-education expenditures will take time to work and to reduce the growth of BMI levels and the prevalence of obesity or overweight; therefore, we may see stronger effects of lagged rather than current expenditures (Ciecierski et al., 2011; Farrelly et al., 2003; Tauras, 2005). Despite the fact that these models control for state-level linear and quadratic trends, a possibility remains that reverse causality could confound the results with a bias toward zero. Reverse causality would be present if current weight outcomes at the statelevel are driving the level of future state expenditures on nutrition-education spending. To address any remaining concerns we use the most direct method to determine if current state-level weight outcomes impact state-level funding. We estimate a model of state-level funding. In the state-level funding model, we use state-level data to determine if current average state weight outcomes impact the level of future state-level funding. These models also control for state-fixed effects, year fixed effects and time trends. To be complete, we also account for other variables that may affect the level of funding at the state-level given the nature of the funding allocation. For example, we include the number of land grant institutions in the state, and the number of Ph.D. graduates from land grant institutions per year as a proxy for the ability/desire to apply for external grants. Other state-level variables included are unemployment-level and population. The results of this state-level model and the other models described in this section are presented in the next section. Results We examine the relationship state-level nutrition-education spending and adult weight outcomes in Table 3. The table provides the estimated coefficient and robust standard errors from OLS models for BMI and probit models for obese and overweight. The models are estimated using the full sample of adults in the BRFSS. We show the results from three different specifications described previously: Eqs. (1), Eq. (2), Eq. (2*), which is Eq. (2) with one year lags of cumulative funding and food prices, and Eq. (2**) which is Eq. (2) two year lags of cumulative funding and food prices as described in the previous section. Estimates from each specification can be found in the respective column for each weight outcome measure. Further, Table 3 shows the estimated marginal effect and robust standard error of the cumulative nutrition-education spending variables only (full regression results are available from the author upon request). State-level and other controls for endogeneity Column (1) of Table 3 suggests that the cumulative effect nutrition-education funding negatively and significantly impacts BMI and Overweight on the margin. These effects are encouraging. As described previously, the state-fixed effects and year fixed effects in Eq. (1) account for all time invariant state or year effects that may confound an observed association between state-level antiobesity policies and individual weight outcomes. For instance, states with high-levels of healthy behaviors or sentiment may spend relatively more than other states on obesity reduction and may be more likely to have citizens who are more prone to healthy behaviors, and, thus, those states may have a lower BMI on average. Controlling for the state-level fixed effects removes this confounding variation from the estimation. However, these results should be interpreted with caution. As described previously, the

73

possibility remains that unmeasured state-level trends, or some form of structural endogeneity, may confound our findings. Therefore, we now focus on the columns labeled (2) under each weight outcome in Table 3. These estimations demonstrate a change from the results presented in the columns labeled (1) suggesting underlying state-level trends do, in fact, introduce a bias. Given that all unobserved time variation at the state-level is included in this estimation, including all variation that may be correlated with the trends in individual weight outcomes and trends real cumulative nutrition-education funding, we should have more confidence in the results of the model presented in those columns labeled (2). As presented, these results suggest changes in the cumulative amount of state-level funding for nutrition-education information has an impact an outcome of interest, obesity (overweight is nearly statistically significant), but not BMI. It is plausible, after controlling for state-specific trends, states with larger increases in funding over the 15 years, mitigate the increase in prevalence of the most extreme outcome, obesity, compared to any other weight outcome. However, attempting to estimate a result of current cumulative funding on current weight outcomes may bias the results toward zero due to reverse causality. Therefore, we lag the policy variables to determine if previous levels of cumulative funding contribute to the current weight outcomes in the states. The results of this strategy using one year lags (Column (2*)) and two year lags (Column (2**)) are presented for each weight outcome in Table 3. As shown, we find the one year lag of cumulative funding reduces BMI, obese and overweight. The two year lag of cumulative funding also reduces BMI, obese and overweight and by a slightly larger amount. The strategy of using lagged measures of funding and the results are consistent with other studies of education programs (Ciecierski et al., 2011; Farrelly et al., 2003; Tauras et al., 2005). However, since the current study is the first aggregate study, to our knowledge, of nutritioneducation funding a similar comparison is not possible. In addition, the effects of real prices for food purchased for at home consumption (Home Food Price) and for food purchased for away from home consumption (Away Food Price) are statistically insignificant. However, the direction of influence for both Home Food Price and Away Food Price is consistent with expectations. Reductions in prices for food purchased for at home preparation and consumption relative to prices for purchased for away from home consumption is consistent with reductions in BMI, obesity and overweight. State funding results As described above, concerns remained regarding the structural endogeneity that would occur due to the possibility of reverse causality with average BMI, prevalence of Obesity or prevalence of BMI influencing policy making. We further investigate this issue in the most straightforward way possible and estimate a model of state-level funding. In this model, as described in the previous section, the future funding level is estimated as a function of statelevel characteristics and fixed effects for state and year, and year time trends. These results can be found in Table 4. Table 4 presents the results of three separate estimations. Column (1) is the most naïve model that does not control for any state, year or state by year effects. Column (2) presents the results of the model with state and year fixed effects. Column (3) offers the final most accurate model that controls for any unobserved heterogeneity by state, year and for any unobserved trends in each state over time that could confound the results. In columns (2) the controls for state and year remove any bias and suggesting that state-level BMI, overweight, and obesity are not significant predictors of funding levels. Column (3) adds additional controls for state by year effects, in this model

74

Table 3 Response of weight outcomes to cumulative real education funds. A comparison of models and measures of funding. BMI

Real Away Food Price

(2)

L0.0006b (0.0003) L0.3191c (0.1843) 0.0026 (0.1480)

0.0022 (0.0015) 0.1799 (0.2246) 0.0426 (0.0949)

(2*)

Overweight

(1)

(2)

0.0000 (0.0000) L0.0278b (0.0138) 0.0009 (0.0102)

L0.0003c (0.0001) 0.0211 (0.0167) 0.0005 (0.0086)

L0.3290a (0.1442) 0.2823 (0.2135) 0.0497 (0.1304)

Cumulative Real Nutrition Education Funds One Year Lag (per million) Real Home Food Price One year lag Real Away Food Price One-year lag Cumulative Real Nutrition Education Funds Two Year Lag (per million) Real Home Food Price Two year lag Real Away Food Price Two year lag State-fixed effects Year Fixed Effects State Time Trend-Linear State Time Trend-Quadratic N

(2**)

X X

2,249,713

X X X X 2,249,713

X X X X 2,248,274

(2*)

(2**)

(1)

(2)

L0.0000b (0.0000) 0.0041 (0.0148) 0.0017 (0.0112)

0.0002 (0.0001) 0.0119 (0.0187) 0.0018 (0.0104)

L0.02937a (0.0109) 0.0231 (0.0150) 0.0063 (0.0008) L0.3719a (0.1322) 0.3255 (0.1998) 0.0850 (0.1446) X X X X 2,247,051

(2*)

L0.0259b (0.0120) 0.0287 (0.0184) 0.0033 (0.0135) L0.0239a (0.0104) 0.0106 (0.0137) 0.0072 (0.0099)

X X

2,249,713

X X X X 2,249,713

X X X X 2,248,274

(2**)

X X

2,247,051

2,249,713

X X X X 2,249,713

X X X X 2,248,274

Heteroskedasticity corrected standard errors that are adjusted for state-level clustering appear in the parentheses, c ¼ p < 0.10, b ¼ p < 0.05, a ¼ p < 0.01. Significant results are presented in bold. * One year lag key variables. ** Two year lag key variables.

L0.0322a (0.0083) 0.0278 (0.0194) 0.0044 (0.0114) X X X X 2,247,051

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

Cumulative Real Nutrition Education Funds (per million) Real Home Food Price

Obese

(1)

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

75

Table 4 State-level prediction of future funding levels.

Real Nutrition Education Funding Today (in millions) Number of Land Grant Universities Unemployment Rate West Midwest South Population (in millions) Number of Land Grant PhD Grads Average BMI in State Prevalence of Overweight Prevalence of Obesity State Fixed Effects Year Fixed Effects State Specific Time Trend Adjusted R2 N

(1)

(2)

(3)

1.1198a (0.0118) 0.3659a (0.1071) 0.0527b (0.0244) 0.1546 (0.0975) 0.1070 (0.1130) 0.1345 (0.0986) 0.0297b (0.0121) 0.0002 (0.0012) L1.0608b (0.4382) 7.3213b (3.2878) 14.2108b (6.6137)

1.0765a (0.0149) 0.0820 (0.1947) L0.1131b (0.0465) 5.2336a (1.8453) 3.6904a (1.3937) 4.1079a (1.4775) 0.3100a (0.1007) 0.0000 (0.0026) 0.4847 (0.5078) 3.6745 (3.7591) 6.5587 (7.1204) X X

0.95 645

0.97 645

0.7331a (0.0241) 0.0725 (0.1747) 0.0397 (0.0499) 1.6004a (0.6064) L11.2429a (1.9370) L42.3728a (7.1280) 2.5823a (0.4394) 0.0070a (0.0025) 0.4586 (0.4634) 3.5844 (3.5350) 9.6997 (6.3052) X X X 0.98 645

Heteroskedasticity corrected standard errors that are adjusted for state-level clustering appear in the parentheses, c ¼ p < 0.10, b ¼ p < 0.05, a ¼ p < 0.01.

we find that state-level BMI, prevalence of obesity and overweight remain insignificant predictors of nutrition-education funding. In addition, another interesting result emerges. A significant predictor of future funding is the number of Ph.D. graduates from land grant institutions in the prior year, regional effects, population, and the current funding level. As noted previously, the significance of population in the state-level funding equation further complicates the use of a per capita measure of funding in our estimations of interest. However, it is the lack of significant estimates on the weight outcome variables that is of interest. The lack of significance across weight comes suggests that year to year funding levels are not biased due to endogeneity from policy makers responding to changes in state-level BMI or rates of obesity and overweight.

this model to test for heterogeneity among the subsamples. Next, we determine if the impact of education and price policies vary by education level. We create an interaction between education level and 1) one year lags of real cumulative funding and 2) one year lags of price policy. Further we estimate these models with the requisite controls for state, year and state by year trends to account for unobserved variation. Continuing to demonstrate the importance of adding controls for state-level trends, Table 5 presents the results for a) models with state and year effects only in the columns labeled (1) and b) models with the additional state by year linear and quadratic trends in the columns labeled (2) for each weight outcome. To create the interaction terms, we use the BRFSS categories: less than a high-school education (less than high school); only a high school degree (high school); and some college or more (some college). These three categories were interacted with lagged cumulative real nutrition-education funding and the lagged real food price indices. Given that columns labeled (2) offer the most convincing identification strategy, allowing for all state specific unobservable trends

Heterogeneity across groups Given that we have assuaged our concerns regarding structural endogeneity and reverse causality through the use of state-level trends and lagged policy variables, we now move forward with Table 5 Real cumulative nutrition-education funds lagged by education level. BMI

Lagged Real Cumul. Nutrition-Education Funds * Less Than High School Real Cumul. Nutrition-Education Funds * High School Real Cumul. Nutrition-Education Funds * Some College Lagged Real Home Price * Less Than High School Lagged Real Home Price * High School Lagged Real Home Price * Some College Lagged Real Away Price * Less Than High School Lagged Real Away Price * High School Lagged Real Away Price * Some College State Fixed Effects Year Fixed Effects State Time Trend-Linear State Time Trend-Quadratic N

Obese

Overweight

(1)

(2)

(1)

(2)

(1)

(2)

1.0383c (0.5500)

0.8107 (0.5534)

0.0546 (0.0452)

0.0347 (0.0401)

0.0870 (0.0554)

0.0783 (0.0541)

0.0189 (0.0956)

L0.1602a (0.0518)

0.0052 (0.0082)

L0.0098c (0.0053)

(0.0044) (0.0060)

0.0108 (0.0126)

L0.0917c (0.0489)

L0.2274b (0.1108)

L0.0060c (0.0034)

L0.0206a (0.0072)

0.0067 (0.0041)

0.0136 (0.0103)

0.8794 (0.5387)

L0.9983b (0.4393)

0.0399 (0.0349)

0.0276 (0.0311)

0.0344 (0.0550)

0.0549 (0.0453)

0.5375a (0.1652)

0.5913a (0.1860)

0.0259b (0.0100)

0.0447a (0.0118)

0.0694a (0.0152)

0.0701a (0.0156)

L1.1153a (0.1794)

L1.0276a (0.2513)

L0.0725a (0.0125)

L0.0515a (0.0164)

L0.0471a (0.0165)

L0.0435c (0.0243)

0.4680c (0.2588)

0.5362b (0.2111)

0.0236 (0.0168)

0.0182 (0.0150)

0.0215 (0.0264)

0.0319 (0.0217)

L0.4009a (0.0846)

L0.4227a (0.0924)

L0.0241a (0.0051)

L0.0331a (0.0058)

L0.0385a (0.0076)

L0.0386a (0.0078)

0.2633a (0.0908)

0.2249c (0.1216)

0.0149b (0.0063)

0.0049 (0.0079)

0.0042 (0.0080)

0.0027 (0.0117)

X X

X X X X 2,249,713

X X

X X X X 2,249,713

X X

X X X X 2,249,713

2,249,713

2,249,713

2,249,713

Heteroskedasticity corrected standard errors that are adjusted for state-level clustering appear in the parentheses, c ¼ p < 0.10, b ¼ p < 0.05, a ¼ p < 0.01. Additional variables included are age, age squared, gender, education level, real income, martial and employment status.

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

76

and reducing bias from reverse causality, we will focus our attention here. The influence of nutrition-education expenditures on all three weight outcomes does, in fact, vary by education level. Among the least educated, real nutrition-education funding has a positive effect but insignificant effect. Among the significant results, there is a negative impact of nutrition-education funding on BMI, obese and overweight. In addition, as expected, the size of the impact increases, in absolute value, with education level. This suggests that nutrition-education policies may disproportionately benefit higher educated groups and, may, demonstrate the need for further study. Additionally, the results in Table 5 highlight the differential effects of price policies across education-levels. Among the individuals with a high-school education, increasing Home Food

Prices while holding Away Food Price constant will increase BMI, obese and overweight among this cohort. Meanwhile, individuals with a high-school education will see an improvement in their weight outcomes, BMI, obese and overweight fall, as Away Food Price increases, all else equal. The results for price responses among the other education cohorts are either insignificant or counterintuitive. Next, we investigate the interactive effect of real cumulative nutrition-education funding and price policy by income level. To create the interaction terms, the following income categories provided by the BRFSS were used: less than $10,000/year; between $10,000 and $15,000; between $15,000 and $20,000; between $20,000 and $25,000; between $25,000 and $35,000; between

Table 6 Lagged real cumulative education funds by income level with linear and quadratic time trends. BMI

Lagged Real Cumul. Nutrition-Education Funds* Income < $10 Thous. Lagged Real Cumul. Nutrition-Education Funds * $10 < Income < $15 Thous. Lagged Real Cumul. Nutrition-Education Funds * $15 < Income < $20 Thous. Lagged Real Cumul. Nutrition-Education Funds* $20 < Income < $25 Thous. Lagged Real Cumul. Nutrition-Education Funds * $25 < Income < $35 Thous. Lagged Real Cumul. Nutrition-Education Funds* $35 < Income < $50 Thous. Lagged Real Cumul. Nutrition-Education Funds * $50 < Income < $75 Thous. Lagged Real Cumul. Nutrition-Education Funds * Income < $75 Thous. Lagged Real Home Price* Income < $10 Thous. Lagged Real Home Price * $10 < Income < $15 Thous. Lagged Real Home Price * $15 < Income < $20 Thous. Lagged Real Home Price * $20 < Income < $25 Thous. Lagged Real Home Price * $25 < Income < $35 Thous. Lagged Real Home Price * $35 < Income < $50 Thous. Lagged Real Home Price * $50 < Income < $75 Thous. Lagged Real Home Price * Income < $75 Thous. Lagged Real Away Price* Income < $10 Thous. Lagged Real Away Price * $10 < Income < $15 Thous. Lagged Real Away Price * $15 < Income < $20 Thous. Lagged Real Away Price * $20 < Income < $25 Thous. Lagged Real Away Price * $25 < Income < $35 Thous. Lagged Real Away Price * $35 < Income < $50 Thous. Lagged Real Away Price * $50 < Income < $75 Thous. Lagged Real Away Price * Income < $75 Thous. State Fixed Effects Year Fixed Effects State Time Trend-Linear State Time Trend-Quadratic N

Obese

Overweight

(1)

(2)

(1)

(2)

0.0958 (0.2923)

0.3693 (0.3801)

0.0096 (0.0110)

L0.0377b (0.0161) L0.0144c (0.0072) L0.1608b (0.0605)

(1)

(2)

0.2063 (0.2854)

0.065 (0.3794)

0.0107 (0.0144)

0.0173 (0.0196)

0.0147 (0.0117)

(0.0807) (0.0699)

0.1235 (0.2424)

0.1528 (0.3335)

0.0026 (0.0115)

0.0257 (0.0162)

(0.0032) (0.0138)

0.0065 (0.0704)

0.1104 (0.1155)

0.1652 (0.2067)

0.0130c (0.0077)

0.0155 (0.0138)

0.0059 (0.0064)

(0.0848) (0.0593)

0.0010 (0.1020)

0.2755 (0.2030)

0.0135 (0.0097)

0.0148 (0.0154)

0.0090 (0.0061)

0.1007 (0.1049)

0.0130 (0.0600)

L0.2911c (0.1710) 0.0029 (0.0047)

L0.0257c (0.0128)

(0.0037) (0.0054)

0.0608 (0.1028)

0.0102 (0.0509)

L0.2701c (0.1357) 0.0072b (0.0030)

L0.0216c (0.0112)

(0.0012) (0.0039)

0.0164 (0.0558)

0.2572 (0.1653)

L0.5355a (0.1361) 0.0168 (0.0101)

L0.0454a (0.0134) L0.0139c (0.0081) 0.0769 (0.0534)

0.3989 (0.4224)

0.8107b (0.3777)

0.0124 (0.0232)

0.0294 (0.0210)

0.0095 (0.0241)

0.0216 (0.0212)

0.1760 (0.4314)

0.6083 (0.4089)

0.0079 (0.0232)

0.0354 (0.0228)

0.0171 (0.0267)

0.0305 (0.0258)

0.0157 (0.3061)

0.4534 (0.3054)

0.0072 (0.0218)

0.0368c (0.0205)

0.0201 (0.0191)

0.0327 (0.0211)

0.1073 (0.2502)

0.3415 (0.2698)

0.0136 (0.0174)

0.0311c (0.0182)

0.0298 (0.0214)

0.0445c (0.0233)

0.0469 (0.2582)

0.4031 (0.2778)

0.0084 (0.0147)

0.0362b (0.0175)

0.0432b (0.0182)

0.0589a (0.0216)

0.2443 (0.2409)

0.2212 (0.2532)

0.0212 (0.0151)

0.0241 (0.0173)

0.0136 (0.0182)

0.0311 (0.0232)

0.0261 (0.2329)

0.4960c (0.2557)

0.0074 (0.0157)

0.0382b (0.0189)

0.0189 (0.0160)

0.0375c (0.0191)

L0.7461b (0.3030) 0.2399 (0.3317)

L0.0621a (0.0212) 0.0143 (0.0236)

L0.0370c (0.0213) 0.0134 (0.0262)

0.3672 (0.2502)

0.3004 (0.2781)

0.0332b (0.0144)

0.0227 (0.0160)

0.0079 (0.0139)

0.0090 (0.0179)

0.3514c (0.2032)

0.2786 (0.2263)

0.0245b (0.0121)

0.0135 (0.0139)

0.0031 (0.0123)

0.0037 (0.0157)

0.3021c (0.1770)

0.2300 (0.1991)

0.0172c (0.0097)

0.0061 (0.0105)

0.0006 (0.0113)

0.0005 (0.0155)

0.2420 (0.1689)

0.1682 (0.1798)

0.0127 (0.0095)

0.0015 (0.0100)

0.0089 (0.0105)

(0.0087) (0.0136)

0.0977 (0.1315)

0.0285 (0.1553)

0.0031 (0.0089)

0.0077 (0.0101)

L0.0161c (0.0090) 0.0162 (0.0127)

0.0513 (0.1372)

0.0165 (0.1431)

0.0023 (0.0094)

0.0084 (0.0097)

0.0015 (0.0122)

0.0022 (0.0157)

0.3044 (0.2095)

L0.3596c (0.2040) 0.0143 (0.0142)

L0.0241c (0.0141)

0.0056 (0.0150)

0.0062 (0.0161)

0.3189 (0.3386)

0.3764 (0.3351)

0.0076 (0.0216)

0.0176 (0.0211)

0.0071 (0.0216)

0.0046 (0.0230)

X X

X X X X 2,249,713

X X

X X X X 2,249,713

X X

X X X X 2,249,713

2,249,713

2,249,713

2,249,713

Heteroskedasticity corrected standard errors that are adjusted for state-level clustering appear in the parentheses, c ¼ p < 0.10, b ¼ p < 0.05, a ¼ p < 0.01. Additional variables included are age, age squared, gender, education level, real income, martial and employment status.

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

77

$35,000 and $50,000; between $50,000 and $75,000; and over $75,000/year. The three categories were interacted with lagged real nutrition-education funding and the lagged real food price indices. The results of this model are presented in Table 6. Again, as with previous tables, we present the results for the three weight outcomes, offering models with controls for time invariant state-level effects and time specific year fixed effects in columns labeled (1) and models with time varying state specific controls in the columns label (2). Further, column 2 remains the column of interest given the level of controls employed. The results indicate the impact of cumulative nutrition-education funding and price policy vary by level of income. We find that lagged cumulative real nutritioneducation funding has a significant and negative impact on BMI and obese for those in highest income groups ($35,000 per year or greater) relative to the other income groups. This result is consistent with the result for the highest education-levels in Table 5. Also, the results in Table 6 suggest that nutrition-education funding is helpful in reducing obese and overweight among those with the lowest household income (less than $10 thousand per year). Further, Table 6 also describes the differential effects of real food prices by income group for the weight outcomes. There is very little consistent and significant effect of prices on weight outcomes by income-level. However, for the near highest income group ($50,000 to $75,000 per year), the Away Food Price has a negative and significant impact on the probability of being obese. Given the inconsistency of the impact of price across income groups, we believe this result should be interpreted cautiously.

direct medical cost estimates not currently available. Further estimation of these costs by group is outside of the scope of this study. It is important to note that, although the estimated impacts of the evaluated policies appear to be small this may not be the case when the funding for nutrition education is considered relative to the costs of obesity. While the funding for nutrition-education programs has grown over the timeframe of study and is currently over $300 million, the aggregate direct medical costs for obesity have been estimated to be from $26.6 billion using MEPS from 2000 to 2005 to $47.5 billion using NHA from 1996 to 1998. Given that the costs associated with obesity are not only from direct medical expenditures but also from productivity and other associated losses, this suggests that the benefits of nutrition-education funding could outweigh the costs associated with these types of programs. In summary, this research takes advantage of already existing data to provide a useful analysis. Further research could offer an investigation of the mechanisms by which states obtain funds and implement nutrition-education programs. In addition, it would be useful to investigate which types of programs are the most beneficial. However, at this writing, these data are not publicly or aggregately available. Further study may determine the heterogeneity of the costs associated with obesity by income and educationlevels. This information would allow us to determine the relative benefits of the nutrition-education programs as the impact varies. Nevertheless, the current study informs the literature and provides new important information regarding the relative effectiveness of state-level policies (price vs. education) designed to reduce obesity.

Discussion and conclusions

Acknowledgments

Given the results presented in Table 3e6, we suggest that the goal of reducing obesity may be achieved through nutrition-education funding. Our estimated marginal effects suggest that an additional $1 million (over 15 years) could reduce the level of BMI by 0.003 points and the prevalence of obesity and overweight by 0.03 percentage points for the US based adult sample. This result is encouraging for a variety of reasons. To begin, $1 million dollars over 15 years amounts to less that 0.5% of the current expenditures on nutrition-education spending. Further, in light of the fact that additional direct medical costs associated with obesity are in the range of $1429 to $2571 annually per person, an additional $1 million is likely to be a cost saving measure (Cawley & Meyerhofer, 2012; Finkelstein et al., 2008). For instance, the increased funding for nutrition-education programs could achieve a reasonable benefit in reduced direct medical costs alone as nutrition education improves obesity rates and BMI. Further, the estimates contained in the previously reported tables suggest the nutrition-education funding may be successful in aggregate. However, the success of nutrition-education programs may have differential effects across income and education-levels. These results are consistent with previous work suggesting that lower income, less well-educated individuals are less responsive to education programs and nutrition information than higher income, better educated individuals. For lower income groups, regressive Away Food Price increases for food prepared away from home has the intended but perhaps not the desired regressive impacts. Also, for individuals with incomes in the range of $35,000 to $75,000 þ per year, BMI and probability of being obese appear to respond to changes in nutrition-education funding. Also, individuals from higher income groups and educationlevels are more responsive to the cumulative effects of nutritioneducation funding than others. While it would be of interest, to determine how the nutrition-education programs were able to reduce direct medical costs for the income- and education-levels affected by nutrition-education funding, to our knowledge, these

The author wishes to thank two anonymous referees for their thoughtful suggestions. The manuscript also benefited from comments made by participants in seminars at Penn State University’s Department of Health Policy and Administration, University of Iceland’s Department of Economics, Lafayette College’s Department of Economics, and the International Health Economics Association Meetings in Beijing, China. The author also wishes to acknowledge Tinna Ásgeirsdóttir, Dhaval Dave and Michael Grossman for their assistance in reading earlier versions of the manuscript. Finally, sections of the manuscript would not have been possible without the exceptional research assistance of Ian P. Reid. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.socscimed.2013.01.023. References Allison, D. B., Fontaine, K. R., Manson, J. E., Stevens, J., & VanItallie, T. B. (1999). Annual deaths attributable to obesity in the United States. Journal of the American Medical Association, 282, 1530e1538. American Chamber of Commerce Researchers Association (ACCRA). (1992). ACCRA cost of living index. Arlington, VA: ACCRA. Auld, M., & Powell, L. (2008). Economics of food energy density and adolescent body weight. Economica, 1e22. Barron, J., Black, D., & Lowenstein, M. (1989). Job matching and on-the-job training. Journal of Labor Economics, 7(1), 1e19. Beydoun, M. A., Powell, L. M., & Wang, Y. F. (2008). The association of fast food, fruit and vegetable prices with dietary intakes among US adults: is there modification by family income? Social Science & Medicine, 66(11), 2218e2229. Burkhauser, R. V., & Cawley, J. (2008). Beyond BMI: the value of more accurate measures of fatness and obesity in social science research. Journal of Health Economics, 27(2), 519e529. Byrd-Williams, C., Storther, M., Kelly, L., & Huang, T. (2009). Dietary fiber and associations with adiposity and fasting insulin among college students with plausible dietary reports. Nutrition, 1e9. Cawley, J. (1999). Rational addiction, the consumption of calories, and body weight. Ph.D. dissertation. University of Chicago, Chicago, IL.

78

K.A. McGeary / Social Science & Medicine 82 (2013) 67e78

Cawley, J., & Meyerhofer, C. (2012). The medical care cost of obesity: an instrumental variable approach. Journal of Health Economics, 31(1), 219e230. Centers for Disease Control. (2006). QuickStats: percentage of adults aged >20 years reporting selected adverse health characteristics, by body mass index (BMI) category e United States, 1999e2002. MMWR Weekly, 55(23), 656, June, 16 2006. Chaloupka, F. J., & Warner, K. E. (2000). The economics of smoking. In A. J. Culyer, & J. P. Newhouse (Eds.), Handbook of health economics (pp. 1539e1627). Amsterdam: North-Holland. Chou, S. Y., Grossman, M., & Saffer, H. (2004). An economic analysis of adult obesity: results from the behavioral risk factor Surveillance system. Journal of Health Economics, 23(3), 565e587. Chou, S. Y., Rashad, I., & Grossman, M. (2008). Fast-food restaurant advertising on television and its influence on childhood obesity. Journal of Law and Economics, 51, 599e618. Ciecierski, C. C., Chatterji, P., Chaloupka, F. J., & Wechsler, H. (2011). Do state expenditures on tobacco control programs decrease use of tobacco products among college students? Health Economics, 20, 253e272. Clifford, D., Anderson, J., Auld, G., & Champ, J. (2009). Good Grubbin’: impact of a TV cooking show for college students living off campus. Journal of Nutrition Education and Behavior, 41(3), 194e200. Connelly, J., Duaso, M., & Butler, G. (2007). A systematic review of controlled trials of interventions to prevent childhood obesity and overweight: a realistic synthesis of the evidence. Public Health, 121, 510e517. Cutler, G., Shapiro, D., Glaeser, E., & Shapiro, J. M. (2003). Why have Americans become more obese? Journal of Economic Perspectives, 17, 93e118. Dansinger, M. L., Gleason, J. A., Griffith, J. L., Selker, H. P., & Schaefer, E. J. (2005). Comparison of the Atkins, Ornish, weight e watchers, and zone diets for weight-loss and heart disease risk reduction: a randomized trial. Journal of the American Medical Association, 293(1), 43e53. Farrelly, M., Pechacek, T., & Chaloupka, F. J. (2003). The impact of tobacco control program expenditures on aggregate cigarette sales: 1981e2000. Journal of Health Economics, 22(5), 843e859. Finkelstein, E. A., Trogden, J. G., Cohen, J. W., & Dietz, W. (2009). Annual medical spending attributable to obesity: payer- and service-specific estimates. Health Affairs, 28, w822ew831. Finkelstein, E. A., Fiebelkorn, I. C., & Wang, G. (2003). National medical spending attributable to overweight and obesity: how much, and who’s paying? Health Affairs, W3, 219e226. Fletcher, J., Frisvold, D., & Tefft, N. (2010). Can soft drink taxes reduce populations weight. Working Paper, 28(1), 23e35.

Goldman, D., Lakdawala, D. and YZheng, (2009), Food prices and the dynamics of body weight, Working Paper, Rand. Grossman, M. (1972). On the concept of health capital and the demand for health. Journal of Political Economy, 80(2), 223e255. Gruber, J., & Frakes, M. (2006). Does falling smoking lead to rising obesity? Journal of Health Economics, 25, 183e197. Kahn, H. S., Tatham, L. M., Rodriguez, C., Calle, E. E., Thun, M. J., & Heath, C. W., Jr. (1997). Stable behaviors associated with adults’ 10-year change in body mass index and likelihood of gain at the waist. American Journal of Public Health, 87, 747e754. Lakdawalla, D., & Philipson, T. J. (2002). Technological change and the growth of obesity. NBER Working Paper 8946. Cambridge MA: National Bureau of Economic Research. Lasker, D., Evans, E., & Layman, D. (2008). Moderate carbohydrate, moderate protein weight, loss diet reduces cardiovascular disease risk compared to high carbohydrate, low protein diet in obese adults: a randomized clinical trial. Nutrition and Metabolism, 7, 5e30. Lleras-Muney, A., & Lichtenberg, F. (2006). The effect of education on medical technology adoption: are the more educated more likely to use new drugs? Annales d’Economie et Statistique, 78/80. McGinnis, J. M., & Foege, W. H. (1993). Actual causes of death in the United States. Journal of the American Medical Association, 270, 2207e2212. Mincer, J. (1984). Labor mobility, wages and training, Mimeo. New York: Columbia University. National Center for Tobacco Free Kids. Toll of tobacco in the United States of America. Tobacco free kids.org Web site, 2002, available at: http://www. tobaccofreekids.org/research/factsheets/pdf/0072.pdf. Accessed: 10.11.09. Philipson, T. (2001). The world-wide growth in obesity: an economic research agenda. Health Economics, 10, 1e7. Price, J., & Simon, K. I. (2009). Patient education and the impact of new medical research. Journal of Health Economics, 28(5), 1166e1174. Rashad, I., Grossman, M., & Chou, S.-Y. (2006). The super size of America: an economic estimation of body mass index and obesity in America. Eastern Economic Journal, 32, 133e148. Reuters. (April, 21, 2009). Profiting from a portly Nation (website accessed 01.01.10). Tauras, J. (2005). Can public policy deter smoking escalation among young adults?”. Journal of Policy Analysis and Management, 24, 771e784. Tauras, J. A., Chaloupka, F. J., Farrelly, M. C., Giovino, G. A., Wakefield, M., Johnston, L. D., et al. (2005). State tobacco control spending and youth smoking. American Journal of Public Health, 95(2), 338e344. U.S. Department of Agriculture. (2004). State programming information.