Abstract

In this article, we empirically study the role of education attainment on individual body mass index (BMI), eating patterns, and physical activity. We allow for endogeneity of schooling choices for females and males in a mean and quantile instrumental variables framework. We find that completion of lower secondary education has a significant positive impact on reduction of individual BMI, containment of calorie consumption, and promotion of physical activity. Interestingly, these effects are heterogeneous across genders and distributions. In particular, for BMI and calorie expenditure, the effect of education is significant for females and is more pronounced for women with high body mass and low physical activity. On the other hand, the effect of education on eating patterns is significant mainly for males, being more beneficial for men with elevated calorie consumption. We also show that education attainment is likely to foster productive and allocative efficiency of individuals in the context of BMI formation. Given that the literature suggests that education fosters development of cognition, self-control, and a variety of skills and abilities, in our context it is thus likely to promote lifetime preferences and means of individuals, which in turn enable them to achieve better health outcomes. Education also provides exposure to physical education and to school subjects enhancing individual deliberative skills, which are important factors shaping calorie expenditure and intake. Finally, we show that in the presence of strong socioeconomic inequalities in BMI, education is likely to have a pronounced impact on healthy BMI for the disadvantaged groups, represented in our framework by females.

Introduction

Rising body weight has become a widespread concern for health care systems, and the World Health Organization (WHO) places it among top global health issues. Excess body weight contributes to development of diabetes, hypercholesterolemia, hypertension, strokes, and a variety of cancers (e.g., Finkelstein et al. 2003; Must et al. 1999; Sturm 2002). Allison et al. (1999) and Peeters et al. (2003) found a positive association between swelling body weight and mortality. In particular, Fontaine et al. (2003) claimed that reduction in life expectancy of a heavily obese young man may reach up to 22 % of his expected life span, accounting for a 13-year anticipation of death.

Excess body weight accumulates because of overeating and scarce physical activity. According to Mokdad et al. (2004), poor diet and lack of physical activity together are about to become one of the leading causes of avoidable death. Philipson and Posner (1999), Lakdawalla and Philipson (2009), Cutler et al. (2003), and Anderson et al. (2003) have suggested that technological progress has reinforced sedentary lifestyles through reduction of on-the-job physical activity, introduction of labor-saving devices, and increase in the availability of unhealthy processed foods.

These trends are particularly harmful for poor socioeconomic groups. Although economic progress and urbanization have reduced the overall opportunities to exercise and access to healthy fruits and vegetables, low-income neighborhoods are the least safe for outdoor physical activity and are the most exposed to “cheap calories,” refined grains, sugary drinks, and fast food (Basiotis and Lino 2002; Drewnowski and Specter 2004; Ranney and McNamara 2004).

Overall, the present world encourages excess body weight unless one has the necessary knowledge, deliberative skills, social support, and economic means to conduct a healthy lifestyle. Many empirical studies have found a strong negative correlation between individual body weight and education, even when controlling for various measures of socioeconomic status (SES) (Baum and Rhum 2009; Clark and Royer 2010; Currie 2009; Currie and Moretti 2003; Cutler and Lleras-Muney 2006; Loureiro and Nayga 2004; MacInnis 2008; Rashad et al. 2006). Although recent decades have seen an increase in schooling attainment with a contemporaneous rise in body weight statistics, individuals who are more educated continue to be slimmer than those who are less educated. Although it has become clear that education is positively associated with healthy body weight, much less is known about the channels through which the educational gradient operates.

As Mirowsky and Ross (2003) suggested, education provides general and cognitive skills, which are the key drivers of the relationship between health and SES. In fact, the authors argued that schooling may influence health through a direct and an indirect channel. First, they argued that education delivers productive abilities and fosters the sense of control throughout life, developing habits of preventing and solving problems. Education teaches us how to communicate, read, write, inquire, research, interpret, experiment, and synthesize ideas. Ruhm (2012) underlined the role of individual deliberative abilities, rooted in cognitive functioning, which enable individuals to evaluate the long-run implications of their lifestyle choices. Also Kenkel et al. (2006), following the seminal work of Grossman (1972), suggested that education rises individual productive efficiency. He argued that better-educated individuals obtain better health outcomes from a fixed set of inputs because they have the abilities and information to make better choices for their lifestyles.

Second, according to Mirowsky and Ross (1989), education indirectly facilitates individual development and interpersonal relationships, enabling people to pursue personal and professional success, which has a positive impact on health. Rosenzweig and Schultz (1983) argued that better education indirectly increases individual allocative efficiency, enabling individuals to allocate more inputs in health. Also, Ross and Mirowsky (1999) suggested that individuals with high levels of perceived control and social support are likely to invest more in health, and hence drink alcohol and smoke tobacco less while choosing balanced diets and active lifestyles—thereby reducing their chances of being overweight or obese.

In this context, we empirically investigate the importance of education for individual body weight and energy balance formation, using a survey of Italian adult population. According to Sassi (2010), Italians show a particularly pronounced negative correlation between body weight and socioeconomic conditions: poorly educated Italian women are 3 times more likely to be overweight than better-educated women, and poorly educated men are 1.3 times more likely to be overweight than better-educated men. Moreover, a recent expansion of unhealthy lifestyles, especially for poor socioeconomic groups, has resulted in a sharp increase in body weight of Italian children, who show the highest prevalence of being overweight among all European countries. If not counteracted, this situation will inevitably translate to a growing share of adults with substantial excess body weight.

The aim of our analysis is thus to expand upon existing studies. First, we construct multidimensional indicators of calorie intake and expenditure in order to analyze the precise impact of schooling on individual BMI in the context of energy balance components. Although energy balance is the building block of individual body weight, it rarely receives attention in empirical studies. Second, we study the gender heterogeneity of the schooling effect along the whole distribution of BMI and energy balance indicators, using instrumental variables (IV) quantile regression. Our instrument is based on a schooling reform, which not only raised the mandatory education-leaving age but also introduced some major qualitative changes to compulsory schooling. Finally, we propose a framework in which we allow education to play an indirect role on BMI through energy balance components. Although in this last case we are not able to provide causal estimates, we nevertheless disentangle the direct and indirect channels of the impact of education attainment in the context of BMI formation.

Overall, and despite minor data limitations of the study, we find that education plays an important role for individual BMI as well as calorie intake and expenditure. Education fosters preferences toward balanced food consumption and enhanced physical activity, which in turn lead to healthy body weight. Moreover, the returns to schooling are highly heterogeneous across individuals with different genders, BMI levels, and different lifestyles.

Data

We conduct the empirical analysis using the multipurpose survey Aspects of Everyday Life (Aspetti della vita quotidiana (AVQ)), which is collected by the Italian National Institute of Statistics. Each year, the AVQ surveys a nationally representative sample of 50,000 individuals and 20,000 households in terms of various aspects of everyday life. We exploit the waves 2005–2009, which contain data on individuals’ weight and height as well as a set of detailed information on their eating habits and physical activities.

Outcome Variables: BMI and Energy Balance Components

Given the information available in the AVQ, we construct three outcome variables: individual BMI, calorie intake, and calorie expenditure.

BMI is the ratio between weight in kilograms and the square height in meters (kg/m2). Its construction is straightforward: the data set provides self-reported information on both height and weight of adult individuals.

A more complex aspect of our analysis lies in the construction of calorie intake and calorie expenditure indicators. In particular, for the calorie intake, we exploit the information on daily and weekly consumption of foods and beverages. All respondents report how frequently and how many portions, on average, they consume of particular nutritional items: the possible answers include “more than once per day,” “once per day,” “more than once per week,” “less than once per week” and “never.”1 We then assign to each item the amount of kilocalories it contains, following information obtained from nutritional charts.2 For each individual, we weigh the kilocaloric contents of each nutritional product with the relative frequency reported. Finally, we tally all the kilocalories to obtain the individual daily calorie intake (CI). The nutritional items listed in the survey are not comprehensive of all foods and beverages, which is likely to cause an overall underestimation of individual daily calorie intakes. Moreover, the frequencies reported by the respondents can be seen as proxies for the latent actual quantities of each nutrient. Nevertheless, in the present analysis, we are not interested in the absolute levels of intakes but rather in the differences in terms of calories consumed across individuals. In this context, the generic and unspecific nature of the nutritional survey comes as an advantage: questions asked to the respondents are very basic and do not involve food diaries, calculations, or major memory demands. In fact, the measure of daily caloric intake obtained within this framework delivers very realistic descriptive statistics.

Second, we construct a measure of calorie expenditure (CE). The AVQ reports various aspects of individual physical activity and sedentariety habits. Individuals rate the intensities of their physical effort made during on-the-job activities and household chores, classifying them as light, medium, or heavy. Furthermore, they indicate the daily amount of time in minutes/hours they devote to each physical effort. We then assess the amounts of time by respective average kilocalorie amounts required for each light/medium/heavy physical effort type.3 Moreover, the respondents deliver precise information on their commuting physical activity: they indicate both the means of transportation used (walk, bike, public transportation and its type, car as driver, or car as passenger) and the amount of time they devote to it daily. Also in this case, we quantify the amount of kilocalories burned daily by each individual according to the type and time of commuting style. In terms of sports, the AVQ supplies detailed information concerning types and frequencies of sports practiced. Each individual provides a rating of sports activity regularity and type. Moreover, the data report information on other types of physical activities not classified as sport. Finally, individuals indicate the daily amount of time in minutes/hours they spend viewing television. To synthesize the multidimensionality of various physical activity dimensions and obtain a comprehensive measure of individual calorie expenditure, we use factor analysis.

Table 1 describes the factor loading of each physical activity dimension. The factor analysis indicates that all variables related to physical activity are combined into two dimensions (two factors with the highest eigenvalues), amounting to 62.74 % of variation. The variables loaded heavily into the first factor relate to on-the-job and household physical activity as well as television watching. On the other hand, sports practice and commuting style feed mainly in the second component. In fact, this variance structure and factor loadings suggest that the first factor extracted refers to physical activity resulting from professional and socioeconomic characteristics. On the other hand, the second factor represents purely the propensity toward sports and leisure physical activity.4 The two physical activity dimensions are constructed as continuous normalized variables with a mean of 0 and a standard deviation of 1. Our final measure of the overall calorie expenditure is the sum of the two dimensions for each individual.5

Educational Attainment

In terms of education, the AVQ data contain information on individual schooling attainment expressed in terms of the highest degree completed. We construct discrete dummy variable indicators pertaining to lower secondary, upper secondary, and tertiary education attainment degrees.

Descriptive Statistics

Tables 2 and 3 present the descriptive statistics from two samples, which we use in two distinct parts of the empirical analysis. For our IV analysis, we use a subsample of individuals born in a six-year window around the pivotal cohort, which was the first to be affected by the compulsory schooling reform. This subsample includes 9,964 females and 9,436 males. When we investigate the indirect role that education has on BMI by means of a simple ordinary least squares (OLS) analysis, we use the larger sample, which consists of pooled data from the five waves, for a total of 65,628 males and 68,964 females who are aged 30–70 and are no longer enrolled in full-time education.

Under many characteristics, the large and the restricted samples are fairly similar, although some reasonable differences emerge. The restricted samples are older, with lower average levels of educational attainment and higher BMI (because of their higher average age). Moreover, in both samples, females have lower levels of kilocalorie intake and job-related physical activity than do males; males devote less time to viewing television and perform less domestic physical activity.

Methods

The main goal of this analysis is to investigate the causal impact of education on BMI, eating patterns (CI), and physical activity (CE) of males and females. First, we account for the endogeneity of education within a mean IV framework, using as the instrument a compulsory schooling reform that occurred in Italy in 1962. Subsequently, we study the heterogeneity of the impact of education on various points of our outcomes distribution by using an IV quantile regression technique (IVQTE). Finally, we propose a framework in which BMI formation is a function of the energy balance components (CI and CE) and education status.

Schooling Reform and IV Approach

Because schooling is voluntary, individuals are likely to self-select into subsequent educational qualifications. In this setting, it is impossible to infer the causal impact of schooling on BMI and healthy lifestyles without accounting for the endogeneity of education. Unobservable traits or abilities of individuals are likely to determine their choices in terms of schooling as well as lifestyles and body weight. Thus, exploring the effect of better education on healthy BMI provides only a correlation, with the causal impact lying in a third, unobservable factor. We therefore adopt an IV technique, which allows us to isolate the variation in individual schooling that is not correlated with the error term containing the unobserved factor, and obtain consistent estimates of the impact of education on BMI and energy balance.

Our IV approach is based on a 1962 schooling reform in the Italian educational system (Law 31.12.1962, art.8 n.1859) (Brandolini and Cipollone 2002; Brunello and Checchi 2007; Oreopoulos 2006). The reform introduced numerous important changes to secondary education and is considered to be the most important schooling reform ever implemented in Italy. First, it raised the compulsory schooling age from 11 to 14. As a result, children were obliged to stay in full-time education at least until graduation from lower secondary school. Second, the policy reformulated the role of schooling in general. Prior to the reform, the Italian schooling system followed from the fascist ideology, and it focused on optimizing individual labor market entry. For this purpose, lower secondary education was organized in two optional school types, which selected and tracked students to the immediate labor market entry or to further schooling. The vocational track prepared lower-ability students to enter the labor market directly after compulsory schooling and concentrated on honing manual skills and abilities at the expense of humanistic disciplines. Moreover, vocational-track graduates could not continue schooling. On the other hand, the general track prepared students for upper secondary education and included more weekly hours and advanced levels of algebra and linguistic subjects. Because the general track was open to students who passed a complex entry exam, lower secondary education acted as a filter that reduced the vertical social mobility and restricted the number of upper secondary and tertiary education graduates. Furthermore, it offered a significantly inferior schooling within the vocational track. The 1962 reform abandoned the dual tracking system and rendered the new general track compulsory for all students. The new unique track featured major qualitative changes; in particular, it officially introduced physical education into the lower secondary schooling curriculum. Overall, the reform established a new role for the lower secondary education, which was no longer seen as a mere preparation for further schooling or profession, but instead was designed to offer general education, deliver knowledge, and encourage personal development.

Following Oreopoulos (2006), we expect that the reform-obliged individuals, who would normally drop out from school earlier or graduate from the vocational track, will obtain the new lower secondary degree. The treatment group consists of the first cohorts affected by the reform—hence, students born after 1949—and the control group includes older cohorts who were not affected by the reform (Brandolini and Cipollone 2002). To isolate potential confounding effects of other policies that came after that date, we limit the analysis to the six subsequent years around the pivotal cohort (1946–1951) (Brunello et al. 2013). As structured, the reform is considered a valid strategy to instrument for education attainment in Italy (Brandolini and Cipollone 2002; Brunello and Checchi 2007; Brunello and Miniaci 1999; Brunello et al. 2013).6 In fact, Fig. 1 shows that the sharp increase in lower secondary school attainment is particularly pronounced for females, and given its timing and entity, it points to an exogenous causal impact of the extended compulsory education policy.

Note that IV estimates of the returns to schooling refer to the local average treatment effect (LATE). Our estimates capture changes in schooling choices of those individuals who would not have remained enrolled in full-time education in the absence of the reform (compliers). These groups of the population are usually restricted7 and may generate estimates that are considerably greater than those of their OLS counterparts.

IV Mean Approach

We estimate the returns to lower secondary education in terms of BMI, CE, and CI. Given that official OECD statistics show major differences in the association between SES and body weight between males and females, and given findings of strong gender differences in the causal effects of schooling (Brunello et al. 2013), we perform the empirical analysis for males and females separately.8 Formally, we specify our empirical model as follows:
formula
(1)
for the first stage and
formula
(2)
for the second stage. Yi denotes, alternatively, individual BMI, eating pattern (CI), or physical activity (CE); Ei is the endogenous variable indicating lower secondary schooling attainment; Ri is the reform dummy variable; and is the first-stage predicted value of Ei. Xi is a vector of exogenous controls, including region fixed effects, as well as a polynomial of first and second degree in age of the individuals.9

As frequently done in IV analysis, we benchmark the two-stage least squares (2SLS) results with OLS estimates. Because the IV specification relies on classifying the subsequent cohorts into those who were affected vis-à-vis those who were not affected by the reform, we compute cohort-clustered bootstrapped standard errors using 1,000 replications.

IV Quantile Approach

The education treatment effect is likely to be heterogeneous across individuals, and standard mean IV estimates cannot account for that. Therefore, we use quantile regression (QR), as originally introduced by Koenker and Bassett (1978), to analyze the effect of education across conditional quantiles of the dependent variables’ distribution. This approach is particularly useful in our setting and delivers important insights from policy perspectives.

BMI, CI, and CE within the normal range are desirable values, and both the right and the left tail of these outcome distributions are associated with high health risks. Moreover, the effects of determinants of BMI, CI, and CE change significantly across these outcome distributions. Thus, the results based on the conditional mean of BMI, CI, and CE will not be necessarily indicative of the size and nature of the effect of covariates on the lower or upper tail of the outcome distributions. In particular, we hypothesize that lower secondary school degree will not have a uniform effect over the whole range of BMI and thus will not result in a pure location shift on its conditional distribution. We expect to find significant BMI disparities between more-educated and less-educated individuals, particularly at the right tail of the distribution. Thus, conditional mean estimates refer to average treatment effects, which are usually not the most important from a policy perspective. Conversely, using the conditional quantiles method, we can provide inferences on values located at the extremes of the distributions, especially at the right of the mean, which potentially entail more health risks for individuals and are more informative regarding the impact of the schooling reform. The same is likely to be true for CI and CE.

In addition, we address the endogeneity of education by adopting the IVQTE estimator, proposed by Abadie et al. (2002). Similar to the mean IV framework, this approach allows us to estimate the causal parameter of the reform for a subsample of individuals whose treatment status is affected by the intervention (local treatment for compliers). The identification strategy allows us to estimate the effect of a binary treatment on a continuous dependent variable along its entire distribution. Compliers are selected via a weighting strategy that assigns individuals to the group of compliers in an average sense, based on means of potential outcomes.10

We thus apply the IVQTE estimator to our reduced-form equation expressed in the 2SLS setting for BMI, CI, and CE. We expect to find evidence of heterogeneity of the impact of secondary school attainment along various points of each outcome distribution, with a relatively greater effect on the tails. In order to highlight the importance of accounting for the endogeneity of schooling choices, we also report the estimates obtained using a standard conditional QR framework (equivalent to the comparison between OLS and 2SLS).

Impact of Education on BMI Through Energy Balance Components

In the final estimation strategy, we want to understand whether education has a direct and indirect effect on BMI, acting through the efficiency of energy balance. To our knowledge, no previous research has analyzed the relationship between BMI and education under this perspective, although many authors have investigated the role that education may have on health outcomes in general.

Following the theoretical considerations presented in the Introduction, we want to understand whether better-educated individuals are more likely to obtain healthy BMI because of their higher productivity in terms of their health outcomes, or instead because of a more efficient allocation of their resources in energy balance components. In the first case, higher productivity deriving from better knowledge, self-control, and various abilities and skills acquired during the educational process should enable individuals to obtain healthy BMI through a more informed, aware, and sophisticated approach to achieving a healthy lifestyle. Hence, net of the kilocaloric quantities expressed by CI and CE, more-educated individuals should be more productive and choose the most efficient nutrients and means of calorie expenditure. Given the amount of kilocalories involved, the impact of physical activity on health varies according to its type, suitability, and precision of the effort made. The same holds true for foods and beverages, which net of the kilocaloric content, may vary dramatically in terms the nutritional value, composition of added sugars or complex carbohydrates, quality, and processes involved in their production. On the other hand, better education may indirectly affect BMI and encourage individuals to dedicate more of their resources to achieve energy balance. Under the indirect pathway, better allocative efficiency should be reflected in the quantities of CI and CE, which in turn should lead to a healthy BMI.

The optimal estimation strategy in this context would be to estimate the direct causal impact of education on BMI jointly with the influence that education has on CI and CE. Nevertheless, from a methodological point of view, it would require more exogenous instrumental variables, which are not available in the data. We thus limit this analysis to a simple OLS estimation, without providing precise inference on causal mechanisms of education. We believe, however, that investigating on the pure correlation among these variables could be of interest and may encourage future ad hoc analysis allowing the estimation of the actual causal effects. We estimate the following simplified reduced-form equation:
formula
(3)
where i = 1, . . ., N is the number of individuals, CIi denotes the individual’s kilocalorie intake, and CEi refers to our calorie expenditure measure. Moreover, Ei is a vector that includes lower secondary, upper secondary, and tertiary education attainment dummy variables; Xi denotes a vector of individual characteristics, including region of residence and a first- and a second-order polynomial in age; and εi is a random error term. Because gauging causal inference is no longer our primary focus, we perform the estimation using the full samples, without restricting the analysis to particular cohorts.

Discussion of Empirical Results

Effect of Education on BMI, CI, and CE: Mean IV Approach

According to the first-stage coefficient estimates, the effect of the reform introduction was stronger in magnitude for females than for males, with the estimates equal to 0.107 and 0.074, respectively. Although the parameter estimates are not particularly high in absolute values, they show that the policy indeed increased the rate of lower secondary school completion among the individuals who were affected by the reform. In fact, as discussed earlier, compliance with the policy was not immediate; hence, the estimates provide a lower bound of the effect of the reform. Moreover, the F statistics are large enough to satisfy the Staiger and Stock (1997) rule of thumb, and eliminate the possibility that the instrument is weak.11

The second-stage results are presented in Table 4. First, we find that lower secondary school attainment reduced BMI for females. This result is in line with findings by Brunello et al. (2013) and Grabner (2009), who also focused on compulsory schooling reforms. In fact, numerous international compulsory schooling reforms of the postwar era reduced inequalities in the access to education and fostered schooling participation of disadvantaged socioeconomic groups and females. According to our estimates, attaining a lower secondary school degree reduced BMI of females by an average of 2 units, which is equivalent to a 10 % reduction in body weight for a normal-weight woman. The effect is stronger than the OLS estimate would suggest. In contrast, the IV result for males is not statistically significant, confirming findings of similar studies (see Brunello et al. 2013; Oreopoulos 2006). As explained earlier, IV analysis provides estimates of the average returns to schooling for compliers (local average treatment effect). Compliers were induced by the reform to pursue three additional years of education in the more-challenging general track. In the absence of the policy, they would have obtained less and lower-quality schooling; thus, their actual returns to additional education are higher. In fact, the first-stage results suggest that the compliance rate was higher among females, who were less likely to graduate from lower secondary school prior to the reform.12

Second, the estimated causal impact of education on calorie intake is strong, in terms of both magnitude and statistical significance. The 2SLS average returns to lower secondary education are heterogeneous across genders, amounting to a reduction of 275 kilocalories per day for females and 447 for males. This finding suggests that additional schooling generated a substantially greater reduction in men’s kilocalories consumed, which is likely to result from the nature of the reform. As explained earlier, the policy-makers imposed a new schooling track, whose priority went beyond mere preparation for the labor market to delivering general knowledge, expand deliberative skills, and foster personal development. Following Mirowsky and Ross (2003), the new lower secondary graduates were more likely to obtain necessary skills and preparation, which subsequently enabled them to choose balanced diets. Moreover, it is easy to imagine that conditional on any level of education, females have more general awareness of nutrition with respect to males—especially for older cohorts. An exogenous exposure to better schooling is thus likely to be more beneficial for males in terms of diet comprehension and healthy calorie consumption.

Third, the differences between OLS and IV estimates are pronounced, especially for females. In fact, while females’ OLS estimates are approximately 10 times smaller than their IV estimates, this difference is 2 times smaller for men. Thus, in case of calorie intake, the distortion in the OLS estimates owing to endogeneity of education is stronger for females than for males, which again may result from compliance with the policy.

Finally, education fosters physical activity, although the 2SLS estimates suggest that the causal effect exists only for females. The OLS coefficients are quite low and homogeneous across genders (0.3–0.4), but the IV estimates for females are much higher (2.5), and this result is statistically significant. According to IV estimates, secondary school attainment increased the amount of physical activity of females by almost 1 standard deviation. On the other hand, when we use the IV method to account for endogeneity in the schooling choices of males, their preferences toward physical activity are no longer shaped by education attainment. To some extent, this result is also likely to be influenced by the narrower compliance rate with the reform of males. However, the gender heterogeneity in the impact of schooling on calorie expenditure is in line with the nature of the reform and theoretical considerations. Given that the reform introduced a structured physical education plan for all lower secondary graduates, new graduates likely developed stronger preferences or habits toward physical activity in general. Also, a reasonable conjecture is that the impact of this particular change brought more benefits for females. According to Sassi (2010), conditional on various socioeconomic measures, women tend to be less physically active than men, especially among older cohorts. In this setting, that completion of more years of schooling and being exposed to physical education is indeed likely to have promoted the lifelong awareness of healthy lifestyles and physical activity among females.13

Effect of Education on BMI, CI, and CE: Quantile IV Approach

The estimates presented in the previous section draw causal inference on the effects of educational policy in terms of mean effects. This section refers to the estimates of the impact of education on BMI, CI, and CE (conditional on age and region of residence) in both the standard and the IV quantile regression. Again, the estimates are obtained for males and females separately. First, the evidence presented in Table 5 shows that the effect of education is heterogeneous along the conditional distribution of BMI, with more-pronounced coefficient estimates for the right tail of the distribution. The QR estimates are substantially smaller than their IVQTE counterparts, and this divergence increases with BMI levels. Moreover, we show that the effect is statistically significant only for women, for whom the completion of lower secondary school reduces their BMI up to 2.5 units in the 75th percentile. This result is consistent findings of Brunello et al. (2013) in terms of magnitude, distributional heterogeneity, and significance. For a better understanding of the differences in the estimates across the BMI distribution and between estimators, Fig. 2 shows the results for the entire conditional distribution of BMI of females, with the relative 95 % confidence interval bands.

In terms of calorie intake, consistent with the previous OLS and IV results, we find that schooling has a positive effect on healthy diets only for males (see Table 6). Moreover, IVQTE suggests that the beneficiary impact of education increases monotonically with the amount of kilocalories consumed. Figure 3, which draws the QR and IVQTE estimates for the conditional distribution of CI for males, shows that IV specifications generate a more accentuated curvature, especially for high calorie intakes.

Finally, also in case of CE, the estimates obtained from a regular QR seem to underestimate the role that schooling has on physical activity (see Table 7). Furthermore, the IVQTE estimates of education show a pattern of heterogeneity across different quantiles that is analogous to the one obtained by QR method. Similar to our mean IV results, we find that IV causal parameters are not significant for males at either distribution point. On the other hand, the estimates for females are statistically significant along the entire conditional distribution of physical activity, with the heterogeneity being accentuated at the extremes of the distribution. Figure 4 plots the QR and IVQTE estimates for the conditional distribution of female physical activity.14

Effect of Education on BMI Through Energy Balance Components

In Table 8, we present the OLS and QR estimates of the effect of education as well as CI and CE on individual BMI. The key difference between the specifications lies in controlling for distinct educational attainment. As expected, CI and CE enter the BMI equation with correct signs (positive and negative, respectively), and both are statistically significant across genders. The amount of kilocalorie intake and physical activity is more relevant for females, being on average twice as large as the respective coefficient estimates for males. Interestingly, after we control for the individual level of education, the relative coefficient estimates on energy balance decline by approximately one-half. This decrease represents important evidence of the role of education for eating patterns and physical activity. It is plausible that education has a dual importance for BMI: an indirect role of fostering balanced calorie intake and promoting physical activity, as well as a direct one of increasing the productivity and the individual’s ability to transform and combine fixed amounts of CI and CE into better outcomes. Our results provide empirical insights in line with the literature concerning the effect of education on health in general. In fact, as suggested in the Introduction, there is a wide consensus that education is likely to increase individual efficiency in terms of both higher productivity and optimal allocation of consumption in the production of health outcomes. Individuals with more schooling are likely to have better deliberative skills, self-control, proper knowledge, and comprehension, which may translate to productivity in terms of health—and healthy BMI, in particular. More schooling also entails longer exposure to physical education and promotes personal and professional success in life, hence fostering the allocation of individual resources into healthy levels of CI and CE.

Finally, the sole covariates relative to energy balance and education attainment explain as much as 14 % of individual variation in BMI among females. The same does not hold true for males, for whom the equivalent explanatory set accounts for only 5 % of the respective variation. In fact, the parameter estimates of educational levels are two times larger in magnitude for females. Again, this evidence is in line with the large socioeconomic inequalities in BMI among Italian women. As a result of complex cultural and socioeconomic factors, schooling participation in Italy is still inferior for females with respect to males. Men enroll in education and participate in the labor market; traditionally, women are primarily responsible for childcare and other nonmarket services. This pattern can be counteracted through schooling, which (aside from promoting labor market entry) offers the means and possibility for cultural and personal development. BMI, geographic residence, and SES are strongly associated for females: women in small urban areas or the southern parts of Italy feature show significant rates of excess body weight along with low levels of labor activity and education attainment. These strong inequalities plausibly represent a significant driver of the gender heterogeneity in the educational gradient on BMI. Thus, a thorough analysis of the patterns of endogeneity, with the inclusion of individual socioeconomic indicators, would surely deliver important empirical insights. However, because of the methodological restrictions and the limitations of the data at hand, this analysis cannot be performed in our setting.15

Also of note is that although our IV analysis is based on the postwar era cohorts, for whom the impact of education was more pronounced, the OLS estimates rely on a larger sample that also contains younger cohorts. Thus, the two sets of estimates may not be directly comparable. In fact, the health-educational gradient is dynamic and may evolve over time, especially in presence of growing access to schooling and better overall health conditions. However, inequalities in education undoubtedly still translate into inequalities in health, conditional on considerable enhancements in both schooling and health provision.

In terms of QR estimates, we find similar differences between the specifications accounting solely for energy balance components and the ones including educational attainment covariates. In fact, the coefficient estimates are approximately two times smaller when we include schooling dummy variables. Moreover, our results suggest that after we control for the educational attainment of men, parameter estimates on the energy balance component are no longer significant. This finding is parallel to our IVQTE estimates presented in the previous section, supporting the evidence of the lack of causal effect of education for men.

Limitations

Several issues concerning our study should be discussed. First, the international literature considers BMI to be a universally comparable indicator of individual body weight. Although widely applied across empirical studies, including the present analysis, BMI cannot be considered the most accurate indicator of individual tissue adiposity. As Monteverde et al. (2010) showed, waist-to-hip ratio (WHR) and waist circumference (WC) can be much stronger predictors of individual health status. In fact, BMI is unable to capture body composition because it does not distinguish body fat from muscle mass. This imprecision may potentially introduce important measurement issues, especially in cases of the elderly or muscular individuals (Johnston and Lee 2011). Nonetheless, the data set used in this analysis does not provide any anthropometric information in order to calculate WHR or WC; hence, we are constrained to limit our analysis to the individual BMI. A potential advantage in this setting is that the sample used for the estimation of the causal effect of schooling is homogeneous in terms of ethnicity and age of individuals. Moreover, the inference is gender-specific, which should minimize the distorting effect of BMI.

Furthermore, the data set at hand provides information on weight and height that is self-reported rather than objectively measured. This issue is a common problem of many studies because as Cawley (1999) noted, females tend to underreport weight, whereas men overreport height. Cawley analyzed U.S. data from the NHANES, which contains both self-reported and actual measures, and proposed a procedure of age-/gender-/race-specific corrections that are widely used in the U.S. literature. Although some argue that this solution is not free of flaws (Plankey et al. 1997), several studies have provided evidence that the results computed on the corrected data do not diverge significantly from estimates obtained on self-reported BMI (see, e.g., Chou et al. 2004).

A similar concern may be raised in terms of the self-reported data on individual calorie intake and expenditure. Energy intake may feature measurement errors and can generally be underreported. According to Lichtman et al. (1993) and Nielsen and Adair (2007), this underreporting may be a function of gender and body weight. If one expects that underreporting is more frequent among overweight and obese individuals, our estimates of the effect of schooling on individual calorie intake may be downwardly biased. Downward bias may also occur if the underreporting is correlated with lower schooling attainment. Conversely, one could expect an opposite measurement error—hence, an overreporting pattern—for calorie expenditure. Again, in such a case, the estimates of the effect of schooling on the individual’s energy expenditure could be downwardly biased. In some empirical studies, this severe limitation of the data is addressed using a technique that classifies individual observations as implausible if the self-reported calorie intake is considerably smaller or larger than calorie expenditure (Goldberg et al. 1991; Johnston and Lee 2011). This method is, however, impossible to apply in our case because both the CI and the CE are self-reported, which limits our potential to correct the data. This study, like many other similar analyses, must therefore rely on the self-reported measures. In this context, the unspecific nature of both the nutritional and physical activity survey used for this study may be advantageous. The respondents answer rather generic and basic. In terms of calorie intake, the types of foods and beverages are widely recognized among Italians, which substantially lowers the likelihood of measurement errors as a function of educational attainment or socioeconomic characteristics. For the energy expenditure indicator, we use factor analysis in order to obtain a unified CE measure from individual responses concerning leisure, commuting, sedentariety, and day-to-day domestic activities. In this setting, where the synthesized measure combines all the CE information, the systematic overreporting issues suspected in the context of self-reported data may be less relevant.

Another important aspect of the analysis is the methodological constraint of the IV technique. To comply with the IV strategy and avoid the violation of the exclusion restriction, we must adopt a parsimonious model limiting the analysis to the sole exogenous controls. The IV requires the vector of explanatory variables to be uncorrelated with the error term. Thus, although they are available in the data set, we do not include covariates related to individual SES or lifestyle characteristics, which are undoubtedly correlated with the education attainment and hence are potentially endogenous. This methodological requirement narrows our inference to the analysis of the sole relationship among education, BMI, and energy balance components, where the variation intrinsic to lifestyle patterns or socioeconomic determinants is not accounted for. The same holds true for smoking and health-related individual conditions. Smoking status and bad health are plausibly strongly related to BMI, CI, CE, and schooling outcomes, although exact directions of causality are difficult to disentangle. Many parallel hypotheses are likely to coexist: for instance, worse health may determine schooling choices as well as BMI, CI, and CE levels; individual BMI likely determines CI and CE as well as individual health and schooling; and schooling probably fosters health and patterns of CI and CE, and consequently BMI. The inclusion of exogenous controls of individual health could undoubtedly enrich our evidence. We expect that with these omitted variables, our estimates would be downwardly biased: accounting for individual health and isolating the effect of individual exogenous health status should boost the pure impact of schooling on BMI, CI, and CE.

Despite all these evident data limitations, we believe that the study offers a robust analysis of the interactions among education, BMI, CI, and CE, which cannot be extended or deepened in the present context given our cross-sectional data set. However, the relationships evidenced in this study should be investigated further, possibly with objectively measured information. Moreover, longitudinal settings offering the possibility of accounting for individual unobserved heterogeneity would offer an important improvement over the data set used in this study.

Conclusions

In this study, we investigate the impact of education on individual BMI formation and energy balance components. Our findings show that completion of lower secondary education has a significant positive impact on reduction of individual BMI, containment of calorie consumption, and promotion of physical activity. These effects are heterogeneous across genders and distributions. For BMI and calorie expenditure, the effect of education is significant only for females—particularly for women with high body mass and low physical activity. On the other hand, the effect of education on eating patterns is significant mainly for males, being more beneficial for men with elevated calorie consumption. We also show that education attainment is likely to foster productive and allocative efficiency of individuals in terms of BMI.

Although both education attainment and body weight have seen a considerable rise in recent years, more-educated individuals remain less subject to overweight and obesity. Our study offers insightful evidence regarding channels through which the educational gradient operates. Framing our analysis in the IV approach based on a particular schooling reform in Italy, we show that aside from bridging socioeconomic gaps, education can be a powerful tool, delivering positive externalities for healthy lifestyles, eating habits, and physical activity. Technically, a healthy BMI is obtained if food choices match the caloric expenditure, which can be achieved given the right preferences, motivation, and comprehension of health guidelines as well as nutrition information. By promoting development of cognition, self-control, and a variety of skills and abilities, education fosters lifetime preferences and means, which in turn enable individuals to achieve better health outcomes. In particular, our results suggest that exposure to physical education and to school subjects enhancing individual deliberative skills is an important factor shaping calorie expenditure of females and calorie intake of males.

Finally, we show that in the presence of strong socioeconomic inequalities in BMI, education is likely to have a pronounced impact on healthy BMI for disadvantaged groups, represented in our framework by females. Because socioeconomic inequality is a dynamic phenomenon, referring to different subsets of the population based on cultural, ethnic, or religious contexts, similar analysis could be conducted for different international settings. Further research in this direction is needed, where availability of more suitable data could provide additional insights into the relationships analyzed in the present study.

Acknowledgments

We would like to thank the anonymous referees as well as Davide Dragone and Pedro Mira for their thorough review and valuable comments. Usual disclaimers apply.

Notes

1

The items listed include foods such as bread, rice, pasta, salt-cured meats, poultry, beef, pork, milk, cheeses, eggs, fish, green vegetables, tomatoes and other vegetables, fruit, green salad, legumes, potatoes, salty snacks, sweets and desserts, olive oil, seed oils, butter, and lard; and beverages, such as water, soda, beer, wine, aperitifs, and liquor.

2

The nutritional charts are available online (http://www.inran.it/646/tabelle\_di\_composizione\_degli\_alimenti.html).

3

Following the charts provided by Harvard Medical School (http://www.health.harvard.edu/newsweek/Calories-burned-in-30-minutes-of-leisure-and-routine-activities.htm), we compute the averages of kilocalories burned performing an hour of light, medium, or heavy physical activity in the professional or household environment.

4

We use the varimax rotation. Varimax relies on orthogonal rotation and maximizes the variance of the squared loading for each factor.

5

We also perform the factor analysis by gender. Using the alternative set of results in the estimation, however, provides identical results.

6

Compliance with the reform was not immediate. Although it obliged all students to graduate from the unique track, the change did not bring instant improvements owing to persisting selective mechanisms within school structures and among teaching staff. As a result, full compliance with the reform occurred only for the cohorts born during the 1970s.

7

According to Oreopoulos (2006), such individuals represent less than 10 % of the population exposed to the instrument.

8

An alternative approach entails a simple inclusion of interaction terms between our relevant covariates and the gender dummy variable in the estimation of the model for the pooled sample. However, by splitting the sample, we provide a more flexible specification and, therefore, more precise estimates. The estimates with interaction terms are available upon request.

9

This parsimonious model limiting the analysis to the sole exogenous controls excludes variables related to SES, which are potentially endogenous.

10

The implementation of the estimator was offered by Abadie et al. (2002), following the original work of Angrist and Imbens (1994). The estimation procedure in Stata follows Frolich and Melly (2010). We thus estimate the effect of education (E) on each Y, as instrumented by the schooling reform R (the treatment). We define Y1 as the Y value for individuals with lower secondary school; Y0 is the Y value for other individuals. Moreover, E1 is the education status for individuals subject to the reform (R = 1), and E0 is that for individuals born before the reform implementation (R = 0). The identification strategy is based on assumptions that Y0, Y1, E0, and E1 are jointly independent of R for covariates X. Furthermore, we assume “no defiers” (Pr(E1E0|X) = 1, nontrivial assignment (0 < Pr(R = 1|X) < 1), and first-stage relevance E[E1|X] ≠ E[E0|X]. This set of assumptions ensures that the estimation is again confined to the treatment effect for compliers, who would not have graduated from lower secondary school if the reform had not been implemented. It does not capture the always-takers and never-takers, who make the educational choices independent of the reform regulations; nor does it include defiers, who are excluded from the analysis by assumption. Chernozhukov and Hansen (2005) propose an alternative approach to the estimation of quantile treatment effects that delivers a global identification strategy. It is, however, impossible to implement the strategy here because of its reliance on rank invariance or rank similarity, which is unlikely to hold in our setting.

11

To check the validity of the results, we estimate analogous specifications with a “placebo” IV, in which we artificially place the reform in different periods. However, we cannot reject the hypothesis that the coefficients on the placebo policy placed randomly in the seven years after the actual reform is null. Although this outcome may weaken our inference, it most certainly results from the gradual implementation of the reform that has spread its effect over time.

12

For all our IV estimates, the control group not affected by the reform consists of cohorts born in the proximity of the war era. These individuals, usually referred to in the literature as “survivors,” have better average education and health status, which may point to underestimation of the effect of education in our case.

13

To explore in more detail the nature of the educational gradient for BMI, CI, and CE, we investigate whether the estimated impact of education might be explained by income. Because individual income is very likely to be endogenous in our setting, we stratify our subsamples according to geographical areas of residence, which determine strong income differences in Italy. We thus divide the individuals (in two, three, and four groups) according to the average regional disposable income of families as registered in the 1960s and reestimate our OLS and 2SLS specifications. However, this additional exercise offers almost identical inferences in terms of statistical significance, magnitude, and gender heterogeneity, independent of the income level. In case of BMI, the reform seems to have been marginally stronger in terms of the compliers’ subpopulation for the low-income regions, which is also reflected in slightly stronger education coefficient estimates in 2SLS results. However, in both cases, we cannot reject the null hypothesis of equality of education estimates across income-specific subsamples. These results are available upon request.

14

For each outcome variable, we test whether quantile regression coefficients on education are statistically significant across conditional quantiles. In particular, we test two null hypothesis, the first one of the equality of coefficients across all quantiles ([q10]edu = [q25]edu = [q50]edu = [q75]edu = [q90]edu), and the second one of the equality of coefficient estimates between the 25th and the 75th quantile ([q25]edu = [q75]edu). The null hypothesis of equality is rejected for all subsamples and specifications. The only exception is for quantile estimates on calorie intake for females; however, single coefficients are not statistically significant and do not provide any inference for our study.

15

Toward this end, we ran an alternative specification, based on stratified samples according to macro-area income levels. Nevertheless, we did not obtain any additional inferences from the analysis, where conditional on the average level of disposable income of the macro-area residents, the gender heterogeneity in the effect of education and CI and CE remains unaltered. The macro aggregation of income measures is not likely to capture sufficiently the relevant variation explaining this particular educational gradient. The results are available upon request.

References

Abadie, A., Angrist, J., & Imbens, G. (
2002
).
Instrumental variables estimates of the effect of subsidized training on the quantiles of trainee earnings
.
Econometrica
,
70
,
91
117
. 10.1111/1468-0262.00270
Allison, D., Fontaine, K., Manson, J., Stevens, J., & VanItallie, T. (
1999
).
Annual deaths attributable to obesity in the United States
.
Journal of the American Medical Association
,
282
,
1530
1538
. 10.1001/jama.282.16.1530
Anderson, P. M., Butcher, K. F., & Levine, P. B. (
2003
).
Maternal employment and overweight children
.
Journal of Health Economics
,
22
,
477
504
. 10.1016/S0167-6296(03)00022-5
Angrist, J., & Imbens, G. (
1994
).
Identification and estimation of local average treatment effects
.
Econometrica
,
62
,
467
475
. 10.2307/2951620
Basiotis, P. P., & Lino, M. (
2002
).
Food insufficiency and prevalence of overweight among adult women
(Nutrition Insights 26).
Alexandria, VA
:
USDA Center for Nutrition Policy and Promotion
.
Baum, C., & Rhum, C. (
2009
).
Age, socioeconomic status and obesity growth
.
Journal of Health Economics
,
28
,
635
648
. 10.1016/j.jhealeco.2009.01.004
Brandolini, A., & Cipollone, P. (
2002
).
Return to education in Italy 1992–1997
. Unpublished manuscript, Research Department, Bank of Italy, Rome, Italy.
Brunello, G., & Checchi, D. (
2007
).
Does school tracking affect equality of opportunity? New international evidence
.
Economic Policy
,
22
,
781
861
. 10.1111/j.1468-0327.2007.00189.x
Brunello, G., Fabbri, D., & Fort, M. (
2013
).
The causal effect of education on the body mass: Evidence from Europe
.
Journal of Labor Economics
,
31
,
195
223
. 10.1086/667236
Brunello, G., & Miniaci, R. (
1999
).
The economic returns to schooling for Italian men. An evaluation based on instrumental variables
.
Labour Economics
,
6
,
509
519
. 10.1016/S0927-5371(98)00013-X
Cawley, J. (
1999
).
Rational addiction, the consumption of calories, and body weight
.
Department of Economics, University of Chicago, Chicago, IL
:
Unpublished doctoral dissertation
.
Chernozhukov, V., & Hansen, C. (
2005
).
An IV model of quantile treatment effects
.
Econometrica
,
73
,
245
261
. 10.1111/j.1468-0262.2005.00570.x
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
,
565
587
. 10.1016/j.jhealeco.2003.10.003
Clark, D., & Royer, H. (
2010
).
The effect of education on adult health and mortality: Evidence from Britain
(NBER Working Paper No. 16013).
Cambridge, MA
:
National Bureau of Economic Research
.
Currie, J. (
2009
).
Healthy, wealthy, and wise: Socioeconomic status, poor health in childhood, and human capital development
.
Journal of Economic Literature
,
47
,
81
122
. 10.1257/jel.47.1.87
Currie, J., & Moretti, E. (
2003
).
Mother’s education and the intergenerational transmission of human capital. Evidence from college openings
.
Quarterly Journal of Economics
,
118
,
1495
1532
. 10.1162/003355303322552856
Cutler, D. M., Glaeser, E. L., & Shapiro, J. M. (
2003
).
Why have Americans become more obese?
.
Journal of Economic Perspectives
,
17
(
3
),
93
118
. 10.1257/089533003769204371
Cutler, D. M., & Lleras-Muney, A. (
2006
).
Education and health: Evaluating theories and evidence
(NBER Working Paper No. 12352).
Cambridge, MA
:
National Bureau of Economic Research
.
Drewnowski, A., & Specter, S. E. (
2004
).
Poverty and obesity: The role of energy density and energy costs
.
American Journal of Clinical Nutrition
,
79
,
6
16
.
Finkelstein, E. A., Fiebelkorn, I. C., & Wang, G. (
2003
).
National medical expenditures attributable to overweight and obesity: How much, and who’s paying?
.
Health Affairs
,
W3
(
Suppl
),
219
226
.
Fontaine, K. R., Redden, D. T., Wang, C., Westfall, A. O., & Allison, D. B. (
2003
).
Years of life lost due to obesity
.
Journal of the American Medical Association
,
289
,
187
193
. 10.1001/jama.289.2.187
Frolich, M., & Melly, B. (
2010
).
Estimation of quantile treatment effects with Stata
.
Stata Journal
,
10
,
423
457
.
Goldberg, G. R., Black, A. E., Jegg, S. A., Cole, T. J., Murgatroyd, P. R., Coward, W. A., & Prentice, A. M. (
1991
).
Critical evaluation of energy intake data using fundamental principles of energy physiology: Derivation of cut-off limits to identify under-recording
.
European Journal of Clinical Nutrition
,
45
,
569
581
.
Grabner, M. (
2009
).
The causal effect of education on obesity: Evidence from compulsory schooling laws
(SSRN working paper).
Rochester, NY
:
Social Science Research Network
.
Grossman, M. (
1972
).
On the concept of health capital and the demand for health
.
Journal of Political Economy
,
80
,
223
255
. 10.1086/259880
Johnston, D. W., & Lee, W. S. (
2011
).
Explaining the female black-white obesity gap: A decomposition analysis of proximal causes
.
Demography
,
48
,
1429
1450
. 10.1007/s13524-011-0064-x
Kenkel, D., Lillard, D., & Mathios, A. (
2006
).
The roles of high school completion and GED receipt in smoking and obesity
.
Journal of Labor Economics
,
24
,
635
660
. 10.1086/504277
Koenker, R., & Bassett, G. (
1978
).
Regression quantiles
.
Econometrica
,
46
,
33
50
. 10.2307/1913643
Lakdawalla, D., & Philipson, T. (
2009
).
The growth of obesity and technological change
.
Economics and Human Biology
,
7
,
283
293
. 10.1016/j.ehb.2009.08.001
Lichtman, S., Pisarska, K., Berman, E., Pestone, M., & Dowling, H. (
1993
).
Discrepancy between self-reported and actual caloric intake and exercise in obese subjects
.
New England Journal of Medicine
,
327
,
1893
1898
. 10.1056/NEJM199212313272701
Loureiro, M. L., & Nayga, R. (2004, August).
Analyzing world health differences in obesity rates: Some policy implications
. Paper presented at the AAEA Meeting, Denver, CO.
MacInnis, B. (
2008
).
Does college education impact health? Evidence from the pre-lottery Vietnam draft
. Unpublished manuscript, Department of Agricultural and Resource Economics, University of California, Berkeley, CA.
Mirowsky, J., & Ross, C. E. (
1989
).
Social causes of psychological distress
.
New York, NY
:
Aldine de Gruyter
.
Mirowsky, J., & Ross, C. E. (
2003
).
Education, social status, and health
.
Hawthorne, NY
:
Aldine de Gruyter
.
Mokdad, A. H., Marks, J. S., Stroup, D. F., & Gerberding, J. L. (
2004
).
Actual causes of death in the United States (2000)
.
Journal of the American Medical Association
,
291
,
1238
1245
. 10.1001/jama.291.10.1238
Monteverde, M., Noronha, K., Palloni, A., & Novak, B. (
2010
).
Obesity and excess mortality among the elderly in the United States and Mexico
.
Demography
,
47
,
79
96
. 10.1353/dem.0.0085
Must, A., Spadano, J., Coakley, E. H., Field, A. E., Colditz, G., & Dietz, W. H. (
1999
).
The disease burden associated with overweight and obesity
.
Journal of the American Medical Association
,
282
,
1523
1529
. 10.1001/jama.282.16.1523
Nielsen, S., & Adair, L. (
2007
).
An alternative to dietary data exclusions
.
Journal of the American Dietetic Association
,
107
,
792
799
. 10.1016/j.jada.2007.02.003
Oreopoulos, P. (
2006
).
Estimating average and local average treatment effects of education when compulsory school laws really matter
.
American Economic Review
,
96
,
152
175
. 10.1257/000282806776157641
Peeters, A., Barendregt, J., Willekens, F., Mackenbach, J., Al Mamun, A., & Bonneux, L. (
2003
).
Obesity in adulthood and its consequences for life expectancy: A life-table analysis
.
Annals of Internal Medicine
,
138
,
24
32
. 10.7326/0003-4819-138-1-200301070-00008
Philipson, T. J., & Posner, R. A. (
1999
).
The long-run growth in obesity as a function of technological change
(NBER Working Paper No. 7423).
Cambridge, MA
:
National Bureau of Economic Research
.
Plankey, M. W., Stevens, J., Flegal, K. M., & Rust, P. F. (
1997
).
Prediction equations do not eliminate systematic error in self-reported body mass index
.
Obesity Research
,
5
,
308
314
. 10.1002/j.1550-8528.1997.tb00556.x
Ranney, C. K., & McNamara, P. E. (
2004
).
Do healthier diets cost more?
.
Nutrition Today
,
39
,
161
168
. 10.1097/00017285-200407000-00006
Rashad, I., Grossman, M., & Chou, S. Y. (
2006
).
The super size of America: An economic estimation of body mass index and obesity in adults
.
Eastern Economic Journal
,
32
,
133
148
.
Rosenzweig, M. R., & Schultz, T. P. (
1983
).
Estimating a household production function: Heterogeneity, the demand for health inputs, and their effects on birth weight
.
Journal of Political Economy
,
91
,
723
746
. 10.1086/261179
Ross, C. E., & Mirowsky, J. (
1999
).
Refining the association between education and health: The effects of quantity, credential, and selectivity
.
Demography
,
36
,
445
460
. 10.2307/2648083
Ruhm, C. (
2012
).
Understanding overeating and obesity
.
Journal of Health Economics
,
31
,
781
796
. 10.1016/j.jhealeco.2012.07.004
Sassi, F. (
2010
).
Obesity and the economics of prevention: Fit not fat
.
Paris, France
:
OECD Publishing
.
Staiger, D., & Stock, J. (
1997
).
Instrumental variables regression with weak instruments
.
Econometrica
,
65
,
557
586
. 10.2307/2171753
Sturm, R. (
2002
).
The effects of obesity, smoking, and drinking on medical problems and costs
.
Health Affairs
,
21
,
245
253
. 10.1377/hlthaff.21.2.245