The Effect of School Lunch on Early Teenagers’ Body Weight
Shiko Maruyama a,*, Sayaka Nakamura b a
Economics Discipline Group, University of Technology Sydney, PO Box 123, Broadway, NSW 2007
Australia, Email:
[email protected] b
School of Economics, Nagoya University, Furocho, Chikusa, Nagoya, 464-8601 Japan, Email:
[email protected]
THIS VERSION: March 27, 2017
JEL codes: H51, I18, I28, J13
1
The Effect of School Lunch on Early Teenagers’ Body Weight Abstract Strengthening nutritional standards for school lunch recently attracts attention as a measure against obesity, but its effectiveness is not fully understood. We examine the population causal effect of Japanese school lunch programs with strict nutritional standards on junior high school students’ weight, exploiting municipal variation in school lunch provision. Unlike in previously studies, individual selection into school lunch participation and stigma induced underreporting of participation are not issues in this study. We use individual level data drawn from the 1975-1994 National Nutrition Survey. To account for possible endogeneity of municipal provision of school lunch, we employ difference in differences framework and compare differences between junior high school students and elementary school students between areas with and without school lunch at junior high schools. We find no evidence that school lunch affects body weight in the full sample analysis. However, in subsample analysis of children with low socioeconomic background, we find significant negative effect of school lunch on BMI and obesity. These findings are robust to propensity score trimming. Additionally, municipal school lunch provision at junior high schools is not significantly associated with growth or weight problems of local preschool and elementary school children, inconsistent with the reverse causality explanation. (199 words)
2
1. Introduction Child obesity is growing rapidly around the world, just like adult obesity (Ng et al. 2014), and its adverse health consequences are well established (e.g., Biro and Wien 2010; Reilly and Kelly 2011; Pulgarón 2013). As a part of public health intervention against obesity, school lunch reforms are recently enacted in the US and the UK: stricter nutritional standards were introduced in the UK and US, and eligibility for free school lunch was significantly expanded in UK (Scottish Government 2014; Woo-Baidal and Taveras 2014; Long 2015). These reforms are, however, under heated debate due to their high costs and the lack of agreement regarding the effectiveness (e.g., McSmith 2010; Tickle 2014; Woo Baidal and Taveras 2014). For instance, a recent study finds no evidence that changes in school lunch quality due to vendor contract turnover cause changes in obesity prevalence in California (Anderson et al. 2017). Additionally, while school lunch has been scrutinized as a risk factor for obesity in the UK and US, previous findings are mixed, with some studies finding support for this possibility for the US (Whitmore-Schanzenbach 2009; Li and Hooker 2010; Millimet et al 2010; Hernandez et al. 2011), and others not finding support for the US (Hofferth and Cuntin 2005; Gleason and Dodd 2009; Hinrichs 2010; Gundersen et al. 2012; Mirtcheva and Powell 2013; Peterson 2014) and the UK (von Hinke Kessler Scholder 2013). School lunch could affect body weight in ambiguous ways. First, school lunch directly affects children’s food intake at lunch, which might increase or decrease their weight, depending on the relative contents of school lunch and counterfactual home-prepared lunch they would have eaten in the absence of school lunch. That is, if school lunch contains more (less) energy or larger (smaller) amount of weight-increasing macronutrients than the home-prepared lunch, then that would lead to an increase (decrease) in body weight, other things being constant. Second, school lunch might have indirect effects on body weight through behavioral responses of children and their parents. In particular, school lunch
3
might have a long-lasting impact on weight by altering children’s food preference and eating habits. At the same time, school lunch might affect children’s food intake at other occasions, which might offset the direct impact of school lunch on weight. In sum, school lunch could either increase, decrease, or have no effect on body weight. Despite this theoretical ambiguity the literature has mostly focused on obesity-increasing effect of school lunch, and no studies that we know of have found weight-reducing effect of school lunch. This study examines causal effects of school lunch on the weight of Japanese junior high school students using individual level data drawn from the 1975-1994 National Nutrition Survey (NNS), a nationally representative household survey with measured height and weight data. Studying Japan fosters understanding the effects of the school lunch reform policies under debate in the UK and US, such as strict nutritional requirements and universal targeting without income-based eligibility restrictions, because they were introduced much earlier in post-WWII Japan. Unlike existing studies that mostly focus on obesity-inducing effects of low-quality school lunch, we exploit this unique feature of Japan to examine the effects of higher-quality school lunch than previously studied ones. Additionally, as in UK and US, recently school lunch attracts public attention in Japan as a measure of promoting healthy eating and combatting rising metabolic syndrome, and Japanese government introduced an agenda to increase school lunch provision at public junior high schools to promote healthy eating (The Third Basic Program for Shokuiku Promotion, 2016 1). Thus, this study aims to facilitate policy debate and development in Japan as well as abroad. The assessment of the causal effect of school lunch participation on obesity has been difficult for several reasons. First, in many countries including the US and the UK school lunch participation is
1
www.mhlw.go.jp/file/06-Seisakujouhou-10900000-Kenkoukyoku/0000129496.pdf. 4
optional, and thus individual selection into participation is possibly endogenous. Previous studies account for this issue using dynamic panel analysis that compares weight changes before and after elementary school entry by school lunch participation status (Whitmore-Schanzenbach 2009; Millimet et al 2010), regression discontinuity based on eligibility cutoff for school lunch subsidies (WhitmoreSchanzenbach 2009), and difference-in-differences (DID) analysis based on policy changes resulting in increased out-of-pocket costs of school lunch for households in certain income brackets (von Hinke Kessler Scholder 2013). Second, school lunch participation is significantly underreported in the US, which has been overlooked in the literature with the exception of Gundersen et al. (2012). Gundersen et al. (2012) conduct nonparametric bound analysis to account for the combination of endogeneity and underreporting of participation, but possibility of large and systematic underreporting prevents definitive conclusion. We assess the population effect of school lunch provision by exploiting unique characteristics of Japanese school lunch programs, whereas prior studies mostly examine marginal effects of increased school lunch participation. Unlike previous studies, this study does not suffer from either endogenous individual selection into school lunch participation or stigma-induced underreporting of participation, because Japanese school lunch programs neither allow individual choice over participation nor have income-based eligibility restrictions. Japanese compulsory education consists of six years of elementary school and three years of junior high school (Ministry of Education 1980), and the majority attends municipal schools for compulsory education (Ministry of Education 1994). Each municipality decides whether or not to provide school lunch at municipal elementary schools and junior high schools, and at schools where school lunch is provided, in principle all students must eat school lunch. While almost all municipalities provide school lunch for elementary schools, there is a large variation in municipal school lunch provision for junior high schools.
5
Municipal provision of school could be endogenous if unobserved municipal characteristics affect both children’s body weight and municipal decision to provide school lunch. The most common approach in this setting would be DID analysis comparing changes in children’s weight by changes in municipal provision of school lunch, but that is infeasible due to cross-sectional nature of our data. Instead, we use a modified DID framework and compare differences between junior high school students and children in higher grades of elementary school between areas with and without school lunch for junior high schools, controlling for municipality-specific effects to account for unobservable heterogeneity in municipal characteristics. We find no evidence that school lunch affects body weight in full sample analysis. In subsamples of children with low socioeconomic background, however, we find significant negative effect of school lunch on body mass index (BMI) and obesity 2. These findings are robust to propensity score trimming. Additionally, municipal school lunch provision at junior high school is not significantly associated with growth or weight problems of local preschool and elementary school children, implying that growth and weight problems among local children has little effects on municipal provision of school lunch for junior high school students. Our findings provide support to the recent movement toward stricter nutritional standard for school lunch and school lunch expansion, while it might question the efficiency of providing school lunch to all students regardless of family income.
2. Background 2.1 Obesity in Japan In Japan, despite the low average BMI and the low prevalence of obesity with BMI 30 or over (less than 3% among adults), the prevalence of obesity-related diseases such as diabetes is close to that in other
2
BMI is defined as [weight in kilograms] / [height in meters]2. 6
developed countries (Guariguata et al. 2014), pausing a serious public health issue (McCurry 2007). Because the rise in obesity-related health risks starts at lower BMI among Asians than among Caucasians, since 2000 the Japan Society for the Study of Obesity advocates defining obesity as BMI of 25 or over, as opposed to the WHO benchmark of BMI 30. Under this criterion about 20% of adult Japanese are obese since the 1950s (Kanazawa et al. 2002; Kodama et al. 2013). Both obesity prevalence and mean BMI have significantly increased among Japanese children from the late 1970s to around 2000 in Japan (Matsushita et al. 2004; Yoshinaga et al. 2010; Maruyama and Nakamura 2015). Recent Japanese studies find that child obesity is a strong predictor of adult obesity (Togashi et al. 2002; Ge et al. 2011) and is significantly associated with child metabolic syndrome (Yoshinaga et al. 2005). Until the 2000s, however, few health professionals were aware of child obesity issues in Japan (Yoshinaga 2012). This suggests that during our study period from 1975 to 1994 child obesity was growing but attracted little attention in Japan, which reduces concerns for reverse causality from local obesity prevalence to municipal school lunch provision. 2.2. The School Lunch Programs In the Japanese compulsory education system 6 to 12-year-olds attend elementary school and 12 to 15year-olds attend junior high school under the strict age-grade system (Ministry of Education 1980), and all but few elementary school students and the majority of junior high school students attend municipal schools (Ministry of Education 1994). In Japan government subsidized school lunch program for elementary school children with low-income background started in 1932, but the program was interrupted due the deterioration in the war situation in 1944. Large-scale provision of school lunch without eligibility restrictions started right after the WWII under American occupation as a measure against child malnutrition resulting from severe food shortage. School lunch for municipal elementary schools was resumed in Tokyo in 1946, and was gradually expanded to nationwide by 1951. The
7
nutritional and sanitary guidelines became the School Lunch Law (henceforth SLL) in 1954, and the SLL was revised to include municipal junior high schools in 1956 (NIEPR 2013). In Japan there are no other large-scale public policy that involves food provision, unlike in the US with various food assistance programs for low-income households, including the National School Lunch Program, the School Breakfast Program and the Food Stamp Program (e.g., Hofferth and Cuntin 2005). The SLL obliges municipalities only to “make effort” to provide school lunch at its municipal elementary and junior-high schools and allows municipalities’ discretion. The school lunch coverage rate for municipal elementary school students has been above 98% since the late 1970s, but school lunch provision for junior high school students increased much more slowly, as detailed in Appendix 1. The school lunch coverage rate for municipal junior high school students gradually increased from about 58% in 1978 to about 75% in 2004, and remained almost constant in the late 2000s (Figure A1). Under rising concerns and awareness over child nutrition (e.g., Mainichi 2016), recently the rate started to rise sharply, from about 77% in 2012 to about 83% in 2015. In 2016 the Japanese government introduced an agenda to increase school lunch provision at municipal junior high schools to promote healthy eating (The Third Basic Program for Shokuiku Promotion 2016). Prefectures with high and low school lunch coverage are almost evenly distributed across Japan, as detailed in Appendix 1 (Figure A2). Despite this municipal variation in school lunch provision, Ministry of Education in principle requires that at schools that receive government funding for school lunch all students eat school lunch without bringing own food to school, except for students with special dietary needs such as food allergy (Ono 2007). 3 An increasing number of municipalities started optional school lunch programs where
3
Ministry of Education was consolidated into MEXT, Ministry of Education, Culture, Sports, Science
and Technology, in 2001. 8
students can choose between school lunch and home-prepared lunch since the 1990s, but that was rare during our study period, as detailed in Appendix 1. Under the SLL municipalities pay the labor costs and the cost of construction and maintenance of facilities, guardians pay ingredient costs and energy bills, and the national treasury subsidizes facility construction (NIEPR, 2013). As a result, households’ payment for school lunch has been kept to a relatively low level. In 2013, for example, the average monthly school lunch fee is about 4,200 yen for elementary schools and about 4,800 yen for junior high schools (MEXT 2015). Households’ ability to pay is unlikely to affect individual children’s school lunch participation because payments are exempted for children from low-income households, and because all but few municipalities provide school lunch to all students regardless of fee payment (Fujisawa 2008). Ministry of Education has set nutritional standards for school lunch, including target values for energy, protein, total fat, calcium, and vitamin since 1954 (Nozue 2011). The standards are significantly stricter than their contemporary counterparts in the UK and US. In particular, in the US even current federal requirements lack target values for protein, total fat, calcium, and vitamin, and there had been no maximum requirements for energy supply until 2012 (Woo-Baidal and Taveras 2014). Similarly, in England and Scotland there have been no legally binding nutrition requirements for school lunch until the mid-2000s (Scottish Government 2008; Dimbleby and Vincent 2013). A survey conducted by Japanese Ministry of Education from 1978 to 1988 finds that the actual intake from school lunch complies well with the nutritional standards (National School Health Center of Japan 1990, 1991). The standards were revised several times, but during our study period there were only minor changes, as detailed in Appendix 1. The SLL lists educational goals of the school lunch program, such as to encourage good eating habits, foster sociability, promote health, and gain knowledge on foods. Since 1958 Ministry of
9
Education placed the school lunch program as a part of the curriculum, and instructed teachers to guide table manners and discourage picky eating, for instance. Since 1970, Ministry of Education encourages each municipal school to hire a licensed nutritionist for nutrition planning, supervision over food preparation and hygiene control, and guidance to students on desirable diet (NIEPR 2013). While some speculate that the school lunch program has contributed to the lower prevalence of obesity in Japan (Fisher 2013; Kaneda and Yamamoto 2015), we know of no study that examines the effect of Japanese school lunch program on child obesity. Previous studies find significant positive association between school lunch and intake of vegetables and dairy products by comparing food intake in weekdays and weekends among students attending schools with school lunch (Takahashi et al. 1983; Nozue et al. 2010) and comparing junior high school students attending municipal schools with and without school lunch (Kawaraya and Mori 2009). Additionally, a survey shows poor nutritional contents of junior high school students’ weekday lunch who attend schools without school lunch: In 1978, only 76.7% of these students bring home-cooked lunch to school every day, and 87.3% of students who do not bring home-cooked lunch only have bread and beverages for lunch (National School Health Center of Japan 1990). 3. Data 3.1. The National Nutrition Survey We construct a sample of 9- to 15-year-old children in elementary school or junior high school using individual-level data drawn from the 1975-1994 National Nutrition Survey (NNS), excluding 12year-olds for years 1975-1985 due to data restriction. The NNS is a nationally representative, annual
10
cross-sectional survey conducted by the Ministry of Welfare. 4 Census districts, which are subdivisions of a municipality, are cluster sampled from all 47 Japanese prefectures, and all households within the sampled districts are asked to participate in the survey. This sampling scheme enables us to exploit municipality-level variation in school lunch program. The response rate is not reported for earlier years, but in 2002, for instance, among about 5,000 households invited to the survey 4,160 participated in the survey (MHLW 2003). Height and weight data are measured without shoes and with adjustment for the weight of clothes by health professionals, and thus are accurate and free from the reporting bias associated with self-reports (Connor Gorber et al. 2007). The NNS also collects information on diet and demographic and socioeconomic characteristics of the household and its members. A certified nutritionist visits each participating household to provide further guidance and correct for misreporting. Details of the data construction, including the reasons for restricting the time frame, are provided in Appendix 2. Our final sample consists of 18,252 children from 2,255 census districts. 3.2. Municipal provision of school lunch We assess the district-level school lunch provision status as follows. Until 1994, each household participating in the NNS survey chooses the survey period covering three consecutive days excluding Sundays and holidays, and reports type of meals (e.g., school lunch) its members had during the survey period. School lunch is not served on Saturdays, and the NNS does not record either the days of the week covered by each household’s survey period. Nevertheless, each household’s survey period covers at least two weekdays. The NNS includes a masked identifier for census district, and each census district
4
Katanoda et al. (2005) confirm the national representativeness of the NNS. The NNS was renamed the
National Health and Nutrition Survey in 2003, and Ministry of Welfare was consolidated into Ministry of Health, Labour, and Welfare (MHLW) in 2001. 11
belongs to only one municipality. Municipalities have one or more elementary schools and junior high schools, and the school district assigns each resident child to one of them. We categorize whether the municipality of each census district provides school lunch at its municipal junior high schools based on the majority rule; if a half or more of junior high school students with valid school lunch information report having school lunch, then we regard municipal junior high school students in the district as having school lunch. We exclude census districts with only one report, and census districts with exactly two conflicting reports. We also complement the data with official statistics, as detailed in Appendix 2. School lunch provision at municipal elementary schools is categorized analogously. Individual reports might not accurately reflect municipal school lunch provision for the following reasons. First, a non-negligible number of junior high school students attend private and national schools and few of these schools provide school lunch, while few Japanese children attend non-municipal elementary schools. Because information on school ownership type is not available in the NNS, we obtain the number of junior high school students by prefecture and ownership type from School Basic Survey conducted by MEXT, 5 and exclude prefectures with high proportion of junior-high school students attending non-municipal schools from sample. The proportion of junior high school students attending non-municipal schools has increased over time, so we exclude prefectures where the proportion is 5% or higher in 1994 (Tokyo, Kochi, Nara, Kanagawa, Kyoto, Hyogo, Hiroshima, Osaka, Chiba, and Mie, in the descending order of the proportion), dropping about 35% of the observations. Second, some children might miss school lunch due to sickness or extra-curricular activities such as excursion, although the NNS instructs households to choose the survey period when the household members have daily life as usual and do not have special events. Third, children might attend municipal
5
http://www.mext.go.jp/b_menu/toukei/chousa01/kihon/1267995.htm 12
schools outside of the municipality. It is permitted under special reasons, such as geographic difficulties in commuting to the designated school (Nakamura 2000). Additionally, although illegal, parents might make a false resident registration, so that their children can attend municipal schools outside of the school district (e.g., Mainichi 2015). Our categorization based on the majority rule and exclusion of census districts with insufficient information would take account of the second and third types of noise. We compare the reports on school lunch and our categorization with the official statistics and confirm the reliability of the reports and our method. Additionally, the census district-level percentage of junior high school students who had school lunch equals either 0 or 100 in the vast majority of cases and is rarely greater than 0 and smaller than 50, which supports our categorization. Details are provided in Appendix 2. 3.3. Other variables and summary statistics As outcome measures we use BMI, obesity, and underweight. We use two different definitions for obesity and underweight. First, for comparability with other studies we use “extended International Obesity Task Force definition (henceforth, IOTF definition)” (Cole and Lobstein 2012). Based on the estimated gender-specific BMI distribution for the Japanese by Kato et al. (2011) we use the proportion corresponding to BMI 25 and 17.5 at age 17.5 as the cutoff value for obesity and underweight, respectively. By construction the rate of obesity varies little by age under the IOTF definition, which increases comparability between younger and older children and fits our DID framework. Second, we also use a modified version of weight for height called “percentage of overweight” (POW), which is widely used in Japan. Under POW criteria, children whose measured weight exceeds (falls below) the standard weight for height by more than 20% are categorized as obese (underweight). We use the standard weight for height by age and gender estimated by Murata and Ito (2003). POW has an advantage over BMI-based measures in that BMI increases with height for children in puberty (Sugiura
13
and Murata 2011). The first point is particularly beneficial for this study because we compare 12- to 15year-olds with 9- to 12-year-olds. 6 In the regression analysis we control for various individual characteristics. To control for gender specific age effects we use age dummies interacted with the female dummy, treating 12-year-olds attending elementary and junior high schools as separate age groups. We also control for various parental and household characteristics, including presence of father, grandfather, and grandmother in household, parental age, parental height and BMI (in z-scores by gender, age, and five-year cohort), parental occupations (laborer, white collar worker, self-employed, agriculture/fisheries/forestry, other/not in employment), the number of children (17 years old or younger) in household, and the percentage ranking of total monthly household expenditure per capita. 7 The household expenditure is reported by eight categories and the categorization varies by survey year, so we calculate the percentage rank of each category by survey year over all households. A small number of children without mothers in household are excluded from sample. Summary statistics for the sample are shown in Table 1. About 21.1% of children live in census districts without school lunch at municipal junior high schools. Comparing the treatment and control
6
“WHO Reference” defines obesity and underweight based on the distribution of American children’s
BMI (de Onis et al. 2007). We do not use WHO reference because it does not account for racial differences in height and growth patterns (Cole and Lobstein 2012) and because the IOTF criteria better reflects weight-related health risks for Asian children than WHO reference (de Wilde et al. 2013). 7
The NNS does not record kinship among household members. We regard 27- to 59-year-olds in a
child’s household as parents and those aged 60 or older as grandparents and use mean values in the case of multiple “fathers” or “mothers”. 14
areas, obesity among elementary school students is more prevalent in the control areas than in the treatment areas but obesity among junior high school students is more prevalent in the treatment areas than in the control areas for both POW and the IOTF measures. The differences in height and underweight prevalence between treatment and control areas are small for both elementary school children and junior high school children. 3.4. Subsamples The effect of school lunch on children with lower socioeconomic status (SES) is of particular policy interests for the following reasons. First, in many countries including UK, US, and Canada, the main target of school lunch programs are children in low income families, in contrast to the Japanese school lunch programs that target all students regardless of household income (Harper et al. 2008). Second, low SES is a strong predictor of child obesity in developed countries (Shrewsbury and Wardle 2008). In particular, Japanese studies find negative association of adolescent obesity with household income and per capita household expenditure (Kachi et al. 2015) and positive association of adulthood obesity with low paternal SES and paternal death during childhood (Lee 2013). Additionally, Japanese studies find positive association between higher SES and healthier eating habits, such as higher frequency of eating vegetables (Nakamura et al. 2016) and greater conformity to government dietary guidelines (Fukuda and Hiyoshi 2012). We conduct subsample analysis of children with below-median per capita household expenditure (henceforth, children with low household expenditure) and children who do not have a father in white collar occupations (i.e., children without fathers in household and children with fathers in non-white collar occupations). Table 2 shows the summary statistics of outcome measures for the subsamples. For both elementary school students and junior high school students there are little differences in the means of any of the outcome variables between the subsamples and the full sample.
15
3.5. Determinants of municipal school lunch provision We explore determinants of municipal school lunch provision to examine the reverse causality from child obesity and other growth problems to school lunch provision. We logistically regress No School Lunch Area dummy on means of height, BMI, and rates of obesity and underweight among 1-to 11-yearold elementary and preschool children in the census district, controlling for other area characteristics. The unit of observation is census district, and we exclude districts with less than five elementary school or preschool children. To reduce the effect of age-composition among these young children, for means of height and weight we use z-scores standardized by gender, age, and five-year cohort, and for obesity and underweight we use the IOTF cutoff values for BMI 25 and 17.5, respectively. We also control for geographic characteristics and NNS-based aggregate values. Geographic characteristics include annual prefectural population density obtained from Statistics Bureau (2012), dummies for municipal size (11 largest cities, cities with more than 150 thousand population other than 11 largest cities, cities with 50-150 thousand population, cities with less than 50 thousand population, towns and villages), and prefecture specific effects or regional block specific effects (Hokkaido and Tohoku, Kanto, Chubu, Kinki, Chugoku and Shikoku, Kyushu and Okinawa). NNS-based aggregate values include the number of participants, age composition (proportions of age 1-19, 20-39, 40-59, 60 and older), median percentage ranking of per-capita household expenditure, mean household size, and occupational composition among 23- to 54-year-olds (proportions of laborers, white collar workers, selfemployed, agriculture/fisheries/forestry workers, other/not in employment). Summary statistics are shown in Table 3. Over half of census districts in No school lunch Area are in the 11 largest cities or the other cities with 150,000 population or more, whereas over half of districts with school lunch for municipal junior high schools are in villages, towns, or cities with less than 50,000 population.
16
Estimated coefficients from the logistic regression of No School Lunch Area dummy, i.e., lack of school lunch provision at municipal junior high schools at the census district, are shown in Table 4. Model 1 controls for prefecture fixed effects and Mode 2 controls for regional block fixed effects instead of prefecture fixed effects. The number of observations is smaller for Model 1 because prefectures without variation in No School Lunch Area dummy are omitted from the estimation. In all specifications means of height, BMI, rates of obesity and underweight among elementary school and preschool children are both independently and jointly insignificant, implying that municipal provision of school lunch at junior high school is not influenced by prevalence of obesity or other growth problems among these younger children. The results suggest strong influence of urbanicity: the likelihood of school lunch provision sharply decreases with population density and municipal population size. Economies of scale imply decreasing per capita cost of school lunch with school size and thus increasing municipal school lunch provision with urbanicity, and do not consist with this finding. Possible explanations include national subsidy for school lunch provision in underpopulated areas (Ministry of Education 1980) and higher land and labor costs in urban areas. The estimated coefficients of year dummies indicate an increase in school lunch provision over time, consistent with the nationwide and sample trends (Figure A1). 4. Estimation Strategy 4.1. DID analysis We examine the causal effect of school lunch on junior high school students’ body weight. Cross sectional comparison between junior high school students who have and do not have school lunch would be problematic if there are systematic differences in local obesogenic environment between municipalities that provide and do not provide school lunch. Community characteristics, such as availability of healthy and unhealthy food, urban sprawl, access to parks and sports facilities, and
17
transportation systems are significantly associated with local obesity prevalence (e.g., Lake and Townshend 2006). Although such association is overall weak for children and adolescents (Dunton et al. 2009), American studies find causal effects of access to recreational trails (Sandy et al. 2013) and fast food restaurants (Aviola et al. 2014) on child obesity. To account for possible unobserved heterogeneity, we use DID framework where the differences in outcome measures between elementary school students and junior high school students are compared between areas with and without school lunch for municipal junior high schools, as illustrated in Figure 1. We assess the effect of not having school lunch at municipal junior high school, because our base group is elementary school children who have school lunch. In our analytical framework, all children have school lunch at elementary schools and the majority also have school lunch at junior high schools, but school lunch is removed from some children once they enter junior high school. In our sample municipal provision of school lunch and individual participation are almost identical, because school lunch was not optional, and because we limit our sample to children in prefectures with low presence of national and private junior high schools. Thus, this study assess the population effect of imposing school lunch on all junior-high school students in a district. In contrast, prior studies mostly examine marginal effects of increased school lunch participation. Specifically, the estimated treatment effects in the regression discontinuity design based on eligibility cutoff for school lunch subsidies are for children whose household income are close to the cutoff (WhitmoreSchanzenbach 2009), and those in the DID analysis based on policy changes only affecting households in specific income brackets are for children in those households (von Hinke Kessler Scholder 2013). We estimate the following equation: 𝑌𝑌𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝑋𝑋𝑖𝑖 𝛽𝛽 + 𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾𝛾 𝐻𝐻𝐻𝐻𝐻𝐻ℎ𝑖𝑖 × 𝑁𝑁𝑁𝑁 𝑆𝑆𝑆𝑆ℎ𝑜𝑜𝑜𝑜𝑜𝑜 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿ℎ 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑑𝑑 + 𝜇𝜇𝑑𝑑 + 𝜖𝜖𝑖𝑖𝑖𝑖
18
(1)
where i indexes the child and d indexes the census district. The dependent variable, Y, is one of the outcome measures of body weight, and X is a vector of individual characteristics, including the interaction terms of age dummies and female dummy, as described in subsection 3.6. Junior High dummy indicates that a child is a junior high school student, and No School Lunch Area dummy indicates no school lunch for municipal junior high schools in the census district. Thus, γ, the coefficient
of the interaction term between Junior High and No School Lunch Area, is the average treatment effect. The term µ is a census district specific effect, and ε is the usual error term. The equation does not
include Junior High and No School Lunch Area as regressors, because the effect of attending junior high school is absorbed in the gender-age specific effects, and because the effect of living in No School Lunch Area is absorbed in census district specific effects. Year effects and area characteristics are also absorbed in census district specific effects. We cluster standard errors at the census district level to allow for correlation of the error term within each census district. 4.2. Placebo test: DID analysis of height Our DID framework relies on the assumption that unobserved heterogeneity in area characteristics between the target and control areas has similar effects on both elementary school students and junior high school students. This assumption would be violated if there are systematic differences in the growth pattern, especially in the timing of puberty onset, between the target and control areas, although the use of POW-based obesity definitions would mitigate this potential issue. To examine this possibility we include height as a regressand in DID analysis as a placebo test. Because height is determined primarily by genetic factors and early-life environment (Beard and Blaser 2002), school lunch is unlikely to have a strong, immediate effect on height. Thus, the coefficient of the interaction of Junior High dummy and No School Lunch Area dummy would reflect differences in children’s growth pattern between the target and control areas.
19
4.3. Propensity score trimming We also conduct propensity-score trimming in the DID analysis for a robustness check. To ensure sufficient overlap in characteristics between the target and control areas we trim the sample based on propensity scores of lack of school lunch at municipal junior high schools in the district. Using a logistic regression model we regress No School Lunch Area dummy on area characteristics to obtain propensity scores, where the unit of observation is a child. The explanatory variables closely overlap with those in the census-district level regression in Subsection 3.5, but following Imbens (2015) we use numerical variables instead of dummy variables when possible, omit insignificant variables, and add the interaction terms of significant variables to increase the fitness of the model. Specifically, we omit occupational composition, substitute a linear trend for year dummies and the mean age for the age composition, and add interaction terms of the municipal population size dummies with the annual prefectural population density and with the linear time trend. For each of the target and control groups we define the support as the interval between the first and 99th percentiles of the estimated propensity scores, and exclude observations outside of the common support (Stuart 2010). We also exclude observations whose estimated propensity score is smaller than 0.1 or larger than 0.9 (Imbens 2015). This trimming reduces the absolute standardized difference of all the explanatory variables used in the DID analysis and the Logit analysis between the target and control areas to less than 0.25 for both the full and sub-samples, where the values above 0.25 are considered as problematic (Rubin 2001). 5. Results To examine the common trend assumption between the treatment and control groups Figure 2 plots means, 20th percentiles, and 80th percentiles of height and BMI in the full sample against age by gender for the treatment and control groups. The red vertical line indicates age 12.125, the mean threshold age between elementary school and junior high school when the survey was conducted in
20
November each year. We omit the values for 12-year-olds because our sample excludes 12-year-olds for years 1975-1985 due to data limitations, as described in Appendix 1. For both genders the pseudo height growth curves closely overlap between the treatment and control groups, implying little difference in growth patterns not only in means but also in upper and lower tails of the distribution. Similarly, for both genders there are little differences between the treatment and control groups for the pseudo BMI growth curves in means and upper and lower percentiles, especially for elementary school children. Additionally, for both height and BMI and for both genders the gap between the 80th and 20th percentiles remains almost constant over age, suggesting little difference in growth patterns between taller and shorter children and between heavier and lighter children. These findings support our common trend assumption. The estimated average treatment effects (ATEs) from simple DID regression analysis are shown in panel (a) of Table 5. 8 For the full sample the estimated ATEs are insignificant for height and all of the weight measures. In contrast, for children with low household expenditure, the estimated ATEs are all significantly positive for BMI and both of the obesity measures, while they are insignificant for height and both of the underweight measures. The estimated ATEs for children without fathers in white-collar occupations are highly similar to those for children with low household expenditure in signs and significance: they are significantly positive for BMI and both of the obesity measures and insignificant for height and both of the underweight measures, although the significance is weak for obesity under the POW definition. The results from DID regression analysis with propensity score trimming, shown in panel (b) of Table 5, are highly similar to those from the simple DID analysis.
8
Full regression results are available from the authors upon request. 21
These results imply that the effect of school lunch for junior high school students’ weight is insignificant on average, but that school lunch significantly reduces BMI and obesity for children with lower socioeconomic background. The point-estimates of ATEs are large for children with low household expenditure, although the standard errors are also large: the lack of school lunch increases BMI by 0.316 without trimming and by 0.422 with propensity score trimming, and increases obesity prevalence by around 5% without trimming and by around 4% with propensity score trimming under both IOTF and POW definitions. On the other hand, we find no evidence that school lunch affects underweight prevalence. Finally, the insignificance of height in the full and subsamples implies little systematic differences in the growth pattern between children in areas with and without school lunch at junior high schools, consistent with our assumption. We examined robustness of our findings in several ways. We estimated the same models using zscores by gender, age, and five-year cohort for height and BMI, and confirmed that the results are not affected by this modification. 9 Additionally, prefectures with 5% or more of junior high school students attending non-municipal schools as of 1994 are omitted from our sample, and we relax this exclusion criteria and increase the minimum proportion from 5% to 10%. This modification does not change our findings, as detailed in Appendix 3. We further examine whether the weight reduction effect of school lunch accumulates during junior high school years, and whether it remains after children graduate from junior high school and stop having school lunch, using the full sample and the subsample of children with low household expenditure. To estimate the effect of not having school lunch by age group, we interact the interaction term of No School Lunch Area dummy and Junior High dummy with age group dummies for age 12-13
9
The results are available from the authors upon request. 22
and age 14-15. 10 Additionally, to estimate the aftereffect of school lunch we include 15- to 16-year-old children that are not in compulsory education in the sample and add the interaction term of No School Lunch Area dummy and a dummy for the post junior high school children. The estimated ATEs for BMI and obesity measures from the simple DID analysis are shown in panel (a) of Table 6. In the full sample the estimated ATEs for junior high school students and post junior high school children are all insignificant for BMI and both of the obesity measures. In the subsample of children with low household expenditure, the estimated ATEs are significantly positive for BMI and both of the obesity measures and are similar for 12- to 13-year-olds and 14- to 15-year-olds, but are all insignificant for post junior high school children. The results from DID regression analysis with propensity score trimming, shown in panel (b) of Table 6, are overall similar to those from the simple DID analysis. These results are all consistent with Table 5. These results suggest that the weight-reducing effect of having school lunch during three years of junior high school in addition to six years of elementary school is short-lived and does not accumulate or remain aftermath. This insignificant aftereffect is consistent with our finding of significant negative effect of school lunch on junior high school students’ weight: If school lunch has long-lasting weightloss effect by affecting children’s food preference and eating habits, then that would reduce the difference in body weight between junior high school students with and without school lunch, because they all had school lunch at elementary schools. 6. Conclusion and Discussion We examine the causal effect of school lunch on Japanese junior high school students using individual level data drawn from the 1975-1994 NNS. To account for possible endogeneity of municipal
10
The NNS does not collect information on children’s grade in school. 23
provision of school lunch, we employ DID framework and compare differences between junior high school students and elementary school students in higher grades between areas with and without school lunch at junior high schools. We find no evidence that school lunch affects body weight in the full sample analysis. In subsample analysis of children with low socioeconomic background, however, we find significant negative effect of school lunch on BMI and obesity. Our further analysis suggests that the weight-reducing effect of school lunch is short-lived and does not accumulate or remain aftermath. This study has the following policy implications. First, our findings imply that high-quality school lunch could reduce child obesity and thus might provide empirical support for the recent movement toward stricter nutritional requirements for school lunch in US (Woo-Baidal and Taveras 2014) and UK (Scottish Government 2014; Long 2015), eligibility expansion for free school lunch in UK (Scottish Government 2014; Long 2015), and expansion of school lunch provision at public junior high schools in Japan (Mainichi 2016; The Third Basic Program for Shokuiku Promotion 2016). Second, we find health benefits from school lunch only for children with low SES, which might question the efficiency of Japan’s universal school lunch programs providing school lunch to all students regardless of their families’ economic conditions. Third, despite the second point, the compulsory nature of Japanese school lunch programs might play a key role in reducing obesity, because children with strong preference for fattening food might avoid school lunch if left to own choices. Fourth, the obesityreducing effects of school lunch for low-SES children might have contributed to the small income gradient in child health in Japan compared to other developed countries with similar or lower levels of relative child poverty rates (Nakamura 2014). Fifth, we find no evidence that school lunch reduces underweight, and a possible explanation would be that children suffering from poverty-induced hunger are rare in Japan. Consistent with this possibility, according to an international survey in 2002 only 2% of the Japanese reported that they could not afford to buy food their family needed during the previous
24
year, compared to 15% of Americans and 11% of the British (Pew Research Center 2013). Finally, data limitation prevents us from using more recent data for this study, and future studies would require accumulation of microdata of children with accurate information on both body measurements and school lunch situations. Additionally, previous studies find positive impact of school lunch quality on academic performance in London (Belot and James 2011) and California (Anderson et al. 2017). Future research also includes examining the effect of school lunch provision on cognitive ability in Japan, which is infeasible with our data.
Acknowledgments: The authors are grateful for a Grant-in-Aid for Young Scientists (B) No. 25780187 from the Japan Society for the Promotion of Science. We also wish to thank Mayu Fujii, Jui-fen Rachel Lu, Eiji Mangyo, Ryuichi Tanaka, and seminar participants at Hitotsubashi, Nagoya, Osaka, and Toyama Universities for their helpful comments. Both authors declare that they have no conflicts of interest. An official permission to utilize the National Nutrition Survey and the National Health and Nutrition Survey was obtained from the Statistics and Information Department of the Ministry of Health, Labour and Welfare on October 2, 2015 (Permission Number 1002-6) and November 20, 2015 (Permission Number 1120-1), respectively.
References Anderson ML, Gallagher J, Ritchie ER, 2017. School lunch quality and academic performance. NBER Working Paper, No.23218. Aviola, PA, Nayga RM, Thomsen MR, Danforth D, Smartt J, 2014. The effect of fast-food restaurants on childhood obesity: A school level analysis. Economics and Human Biology, 12:110-119.
25
Beard AS, Blaser MJ, 2002. The ecology of height: The effect of microbial transmission on human height. Perspectives in Biology and Medicine, 45(4): 475-498. Belot M, James J, 2011. Healthy school meals and educational outcomes. Journal of Health Economics, 30(3): 489-504. Biro FM, Wien M, 2010. Childhood obesity and adult morbidities. American Journal of Clinical Nutrition, 91(suppl):1499S-1505S. Cole TJ, Lobstein T, 2012. Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity. Pediatric Obesity, 7(4):284-294 Connor Gorber S, Tremblay M, Moher D, Gorber B, 2007. A comparison of direct vs. self-report measures for assessing height, weight and body mass index: a systematic review. Obesity Reviews, 8(4):307-326. de Onis M, Onyango AW, Borghi E, Siyam A, Nishida C, Siekmann J, 2007. Development of a WHO growth reference for school-aged children and adolescents. Bulletin of the World Health Organization, 85(9):660-667. de Wilde JA, van Dommelen P, Middelkoop BJ, 2013. Appropriate body mass index cut-offs to determine thinness, overweight and obesity in South Asian children in the Netherlands. PLoS One, 8(12):e82822. doi: 10.1371/journal.pone.0082822. Dimbleby H, Vincent J, 2013. The School Food Plan. Department of Education. https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/251020/The_Schoo l_Food_Plan.pdf Dunton, GF, Kaplan J, Wolch J, Jerrett M, Reynolds KD, 2009. Physical environmental correlates of childhood obesity: a systematic review. Obesity Reviews, 10(4):393-402.
26
Fisher M, 2013, January 28. How Japan’s revolutionary school lunches slow the rise of child obesity. The Washington Post. Fujisawa H, 2008. Disputes of school lunch issues. Osaka Keidai ronshu 59(2):199-214 (in Japanese). Fukuda Y, Hiyoshi A, 2012. High quality nutrient intake is associated with higher household expenditures by Japanese adults. BioScience Trends, 6(4):176-182. Ge S, Kubota M, Nagai A, 2011. Retrospective individual tracking of body mass index in obese and thin adolescents back to childhood. Asia Pacific Journal of Clinical Nutrition, 20 (3):432-437. Gleason PM, Dodd AH, 2009. School breakfast program but not school lunch program participation is associated with lower body mass index. Journal of the American Dietetic Association, 109(2):S118-S128. Guariguata L, Whiting DR, Hambleton I, et al. 2014, IDF diabetes atlas: Global estimates of diabetes prevalence for 2013 and projections for 2035, Diabetes Research and Clinical Practice, 103:137149. Gundersen C, Kreider B, Pepper J, 2002. The impact of the National School Lunch Program on child health: a nonparametric bounds analysis. Journal of Econometrics, 166:79-91. Harper C, Wood L, Mitchell C, 2008. The Provision of School Food in 18 Countries. School Food Trust. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.654.9233&rep=rep1&type=pdf Hernandez DC, Francis LA, Doyle EA, 2011. National School Lunch Program participation and sex differences in Body Mass Index trajectories of children from low-income families. Archives of Pediatric Adolescent Medicine, 165(4): 346-353. Hinrichs P, 2010. The effects of the National School Lunch Program on education and health. Journal of Policy Analysis and Management, 29(3): 479-505.
27
Hofferth SL, Curtin S, 2005. Poverty, food programs, and childhood obesity. Journal of Policy Analysis and Management, 24(4): 703-726. Imbens GW, 2015. Matching methods in practice: Three examples. Journal of Human Resources, 50(2):373-419. Kachi Y, Otsuka T, Kawada T, 2015. Socioeconomic status and overweight: A population-based crosssectional study of Japanese children and adolescents. Journal of Epidemiology, 25(7): 463-469. Kanazawa M, Yoshiike N, Osaka T, 2002. Criteria and classification of obesity in Japan and AsiaOceania. Asia Pacific Journal of Clinical Nutrition, 11(Suppl): S732-S737. Kaneda M, Yamamoto S, 2015. The Japanese school lunch and its contribution to health. Nutrition Today, 50(6):268-272. Katanoda K, Nitta H, Hayashi K, Matsumura Y, 2005. Is the national nutrition survey in Japan representative of the entire Japanese population? Nutrition, 21(9):964-966. Kato N, Takimoto H, Sudo N, 2011. The cubic functions for spline smoothed L, S and M values for BMI reference data of Japanese children. Clinical Pediatric Endocrinology, 20(2):47-49. Kawaraya C, Mori H, 2009. The dietary habits of junior high school students: Eating vs. not eating school lunches. Osaka Shoin Women's University Faculty of Liberal Arts collected essays. 46:7790. Kodama K, Tojjar D, Yamada S, et al., 2013. Ethnic differences in the relationship between insulin sensitivity and insulin response: A systematic review and meta-analysis. Diabetes Care, 36:17891796. Lake A, Townshend T, 2006. Obesogenic environments: exploring the built and food environments. Journal of the Royal Society for the Promotion of Health, 126(6): 262-267.
28
Lee Q, 2013. Childhood socioeconomic status and obesity in Japan. Quarterly of Social Security Research, 49(2):217-229 (in Japanese). Li J, Hooker NH, 2010. Childhood obesity and schools: evidence from the national survey of children's health. Journal of School Health, 80(2):96-103. Long R, 2015. School Meals and Nutritional Standards. House of Commons Library, Briefing Paper #04195. researchbriefings.parliament.uk/ResearchBriefing/Summary/SN04195#fullreport Mainichi, 2015, December 10. Himeji City Board of Education to Warn 18 Children Suspected for Attending a School out of School District within This Year. Retrieved from mainichi.jp. (in Japanese) Mainichi, 2016, July 9. Yokosuka City to Introduce Complete School Lunch at Junior High Schools. Retrieved from mainichi.jp. (in Japanese) Maruyama S, Nakamura S, 2015. The decline in BMI among Japanese women after World War II. Economics & Human Biology, 18:125-138. Matsushita Y, Yoshiike N, Kaneda F et al., 2004. Trends in childhood obesity in Japan over the last 25 years from the National Nutrition Survey. Obesity Research, 12:205-214. McCurry J, 2007. Japan battles with obesity. Lancet, 369:451-452. McSmith A, 2010, June 20. Jamie Oliver health approach 'doesn't work', says Health Secretary. Independent, Retrieved from www.independent.co.uk/. MEXT, 2015. Results of Current Status Survey on School Meals (in Japanese). http://www.mext.go.jp/b_menu/houdou/27/02/1354919.htm MHLW, 2003. Report on the 2002 National Health and Nutrition Survey (in Japanese). http://www.mhlw.go.jp/bunya/kenkou/eiyou-chosa2-01/index.html
29
Millimet DL, Tchernis R, Husain M, 2010. School nutrition programs and the incidence of childhood obesity. Journal of Human Resources, 45 (3):640-654. Ministry of Education, 1980. Japan's Modern Educational System: A History of the First Hundred Years. http://www.mext.go.jp/b_menu/hakusho/html/others/detail/1317220.htm Ministry of Education, 1994. Statistical Abstract of Education (in Japanese). http://www.mext.go.jp/b_menu/toukei/002/002b/koumoku.html Mirtcheva DM, Powell LM, 2013. National School Lunch Program participation and child body weight. Eastern Economic Journal, 39:328-345. Murata M, Itoh K, 2003. On the optimal weight of school age children. In: Murata M (ed.). Health Labour Sciences Research Grant Report for FY 2002 on Nutrition, Exercise and Recreation from the Viewpoint of Health and QOL in Children (in Japanese). Nakamura M, 2000. School choice in Japan: Learning from the experiences of other countries. Nagano Technical College Research Reports, 34:107-116. Nakamura S, 2014. Parental income and child health in Japan. Journal of the Japanese and International Economies, 32:42-55. Nakamura S, Inayama T, Hata K, Matsushita M, Takahashi M, Harada K, Arao T, 2016. Association of household income and education with eating behaviors in Japanese adults: a cross-sectional study. BMC Public Health, 16:61. DOI 10.1186/s12889-016-2748-z National School Health Center of Japan, 1979. The School Lunch Data Book: Fiscal Year 1979. Tokyo: Daiichi Hoki Shuppan (in Japanese). National School Health Center of Japan, 1990. The School Lunch Data Book: Fiscal Year 1990. Tokyo: Daiichi Hoki Shuppan (in Japanese).
30
National School Health Center of Japan, 1991. The School Lunch Data Book: Fiscal Year 1991. Tokyo: Daiichi Hoki Shuppan (in Japanese). Ng M, Fleming T, Robinson M, et al., 2014. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet, 384(9945): 766-781. NIEPR (National Institute for Educational Policy Research), 2013. School Lunch Program in Japan. https://www.nier.go.jp/English/educationjapan/pdf/201303SLP.pdf Nozue M, 2011. The contribution of school lunch to dietary intakes in school children and application of dietary reference intakes: A case study of fifth-grade school children. Ph.D. Thesis, Kagawa Nutrition University (in Japanese). Nozue M, Jun K, Ishihara Y, 2010. Differences in food consumption and distribution of meals between the days with or without school lunches among 5th grade elementary school students. Japanese Journal of Nutrition and Dietetics, 68(5):298-308 (in Japanese). Ono N, 2007. Introduction to the review of Article 2 of the School Lunch Implementation Standard. Journal of Humanities and Social Sciences, 23:50-61 (in Japanese). Peterson C, 2014. Investigating the historic long-term population health impact of the US National School Lunch Program. Public Health Nutrition, 17(12), 2783-2789. Pew Research Center, 2013. Economies of Emerging Markets Better Rated During Difficult Times: Global Downturn Takes Heavy Toll; Inequality Seen as Rising. http://www.pewglobal.org/files/2013/05/Pew-Global-Attitudes-Economic-Report-FINAL-May23-20131.pdf Pulgarón, ER, Childhood obesity: A review of increased risk for physical and psychological comorbidities. Clinical Therapeutics, 35(1): A18-A32.
31
Reilly JJ, Kelly J, 2011. Long-term impact of overweight and obesity in childhood and adolescence on morbidity and premature mortality in adulthood: systematic review. International Journal of Obesity, 35: 891-898. Rubin DB, 2001. Using propensity scores to help design observational studies: application to the tobacco litigation. Health Services & Outcomes Research Methodology, 2:169-188. Sandy R, Tchernis R, Wilson J, Liu G, Zhou X, 2013. Effects of the built environment on childhood obesity: The case of urban recreational trails and crime. Economics and Human Biology, 11:18-29. Scottish Government, 2008. Healthy Eating in Schools: A guide to implementing the nutritional requirements for food and drink in schools (Scotland) regulations 2008. http://www.gov.scot/Resource/Doc/238187/0065394.pdf Scottish Government, 2014. Better Eating, Better Learning: A New Context for School Food. http://www.gov.scot/Resource/0044/00445740.pdf Shrewsbury V, Wardle J, 2008. Socioeconomic status and adiposity in childhood: a systematic review of cross-sectional studies 1990-2005. Obesity (Silver Spring), 16:275-84. Statistics Bureau, 2012. Historical Statistics of Japan. http://www.stat.go.jp/english/data/chouki/02.htm Stuart EA, 2010. Matching methods for causal inference: A review and a look forward. Statistical Science, 25(1): 1-21. Sugiura R, Murata M, 2011. Problems with body mass index as an index to evaluate physical status of children in puberty. Pediatrics International, 53:634-642. Takahashi T, MatsuuraY, Osawa S, et al., 1983. Survey on food intake of elementary and junior high school children in Ibaraki prefecture: With special reference to milk, dairy products, and green and yellow vegetables. Japanese Journal of Nutrition and Dietetics, 41(1): 21-29.
32
Tickle L, 2014, May 20. Free school meals: '£56k's being spent to give children a sandwich.' Guardian. Retrieved from www.theguardian.com. Togashi K, Masuda H, Rankinen T, 2002. A 12-year follow-up study of treated obese children in Japan. International Journal of Obesity, 26:770-777. von Hinke Kessler Scholder S, 2013. School meal crowd out in the 1980s. Journal of Health Economics, 32: 538-545. Whitmore-Schanzenbach D, 2009. Do school lunches contribute to childhood obesity? Journal of Human Resources. 44(3):684-709. Woo-Baidal JA, Taveras EM, 2014. Protecting progress against childhood obesity: The National School Lunch Program. New England Journal of Medicine, 371(20): 862-1865. Yoshinaga M, 2012. Current status and strategies for obesity and metabolic syndrome in children and adolescents. Pediatric Cardiology and Cardiac Surgery. 28(2):103-109 (in Japanese). Yoshinaga M, Tanaka S, Shimago A, et al., 2005. Metabolic syndrome in overweight and obese Japanese children. Obesity Research, 13:1135-1140. Yoshinaga M, Ichiki T, Tanaka Y, et al., 2010. Prevalence of childhood obesity from 1978 to 2007 in Japan. Pediatrics International, 52:213-217.
33
Figure 1. Illustration of the DID estimation.
Areas with school lunch for junior high schools
Junior high school students
Areas without school lunch for junior high schools
Treatment: No school lunch
Elementary school students (All have school lunch.)
34
Figure 2. Plots of means and percentiles of height and BMI against age.
Notes: The red vertical line indicates age 12.125, the mean threshold age between elementary school and junior high school. The values for 12-year-olds are omitted due to data limitations.
35
Table 1. Summary statistics for the full sample. All
Variable Number of children Number of areas No school lunch Junior high school Height: elementary Height: junior high BMI: elementary BMI: junior high Obesity: IOTF BMI 25+: elementary Obesity: IOTF BMI 25+: junior high Obesity: POW 20%+: elementary Obesity: POW 20%+: junior high Obesity: IOTF BMI 30+: elementary Obesity: IOTF BMI 30+: junior high Obesity: POW 30%+: elementary Obesity: POW 30%+: junior high Underweight: IOTF BMI 18.5-: elementary Underweight: IOTF BMI 18.5-: junior high Underweight: POW 20%-: elementary Underweight: POW 20%-: junior high Underweight: IOTF BMI 17-: elementary Underweight: IOTF BMI 17-: junior high Underweight: POW 30%-: elementary Underweight: POW 30%-: junior high Year Male Age Prefectural population density Region block: Hokkaido & Tohoku Region block: Kanto Region block: Chubu Region block: Kinki Region block: Chugoku & Shikoku Region block: Kyushu & Okinawa Municipal size: 11 largest cities Municipal size: cities with 150k+ population Municipal size: cities with 50-150k population Municipal size: cities with 50k- population Municipal size: towns & villages (reference)
Mean SD 18,289 . 2,264 . 0.211 0.408 0.464 0.499 138.472 8.555 158.064 7.705 17.279 2.482 19.625 2.637 0.076 0.265 0.059 0.235 0.078 0.267 0.067 0.249 0.005 0.068 0.007 0.083 0.034 0.182 0.030 0.171 0.154 0.361 0.156 0.363 0.020 0.140 0.019 0.136 0.026 0.158 0.023 0.150 0.002 0.046 0.000 0.019 1983.909 5.473 0.516 0.500 11.802 2.007 0.473 0.438 0.174 0.379 0.165 0.371 0.030 0.170 0.114 0.317 0.205 0.404 0.518 0.500 0.057 0.232 0.271 0.444
Treatment (No School Lunch Areas) Mean SD 3,854 . 497 . 1 1 0.461 0.499 138.539 8.383 158.274 7.640 17.170 2.401 19.628 2.653 0.069 0.254 0.061 0.239 0.069 0.254 0.069 0.254 0.004 0.062 0.006 0.075 0.030 0.171 0.034 0.182 0.159 0.366 0.159 0.366 0.019 0.137 0.020 0.139 0.028 0.165 0.021 0.143 0.002 0.049 0.000 0.000 1982.947 5.556 0.521 0.500 11.763 1.998 0.494 0.420 0.170 0.375 0.075 0.263 0.068 0.251 0.087 0.283 0.257 0.437 0.600 0.490 0.173 0.378 0.351 0.477
Control (School Lunch Areas) Mean SD 14,435 . 1,767 . 0 0 0.465 0.499 138.454 8.602 158.009 7.721 17.308 2.502 19.624 2.633 0.078 0.268 0.058 0.234 0.080 0.271 0.066 0.248 0.005 0.069 0.007 0.085 0.035 0.184 0.029 0.168 0.152 0.359 0.155 0.362 0.020 0.141 0.018 0.135 0.025 0.156 0.024 0.152 0.002 0.045 0.000 0.021 1984.166 5.423 0.514 0.500 11.813 2.010 0.467 0.443 0.176 0.381 0.189 0.391 0.019 0.138 0.120 0.326 0.191 0.393 0.496 0.500 0.026 0.159 0.249 0.433
0.219
0.414
0.237
0.425
0.214
0.410
0.106 0.348
0.307 0.476
0.107 0.132
0.310 0.339
0.105 0.406
0.307 0.491
36
Table 1. Summary statistics for the full sample (cont.). All
Variable Father's age Father's height (in z score) Father's BMI (in z score) Father's BMI missing Father: laborer Father: self-employed Father: agriculture Father: other occupation Without father in HH Father: white collar worker Mother's age Mother's height (in z score) Mother's BMI (in z score) Mother's BMI missing Mother: laborer Mother: white collar worker Mother: self-employed Mother: agriculture Mother: other occupation Grandfather in HH Grandmother in HH #child in HH %ranking of HH expenditure
Mean 38.092 -0.025 0.007 0.252 0.292 0.186 0.073 0.009 0.093 0.346 39.557 -0.062 0.073 0.036 0.243 0.163 0.132 0.086 0.376 0.196 0.307 2.303 0.635
37
SD 12.973 0.846 0.851 0.434 0.453 0.388 0.258 0.091 0.291 0.475 4.386 0.976 0.983 0.187 0.426 0.366 0.336 0.277 0.479 0.397 0.461 0.785 0.241
Treatment (No School Lunch Areas) Mean SD 38.147 13.049 0.024 0.852 -0.017 0.865 0.255 0.436 0.297 0.456 0.191 0.393 0.053 0.223 0.009 0.088 0.094 0.292 0.356 0.478 39.553 4.319 -0.018 0.981 -0.001 0.927 0.036 0.185 0.236 0.421 0.170 0.372 0.136 0.342 0.056 0.227 0.402 0.485 0.165 0.371 0.272 0.445 2.277 0.813 0.607 0.239
Control (School Lunch Areas) Mean SD 38.077 12.953 -0.038 0.844 0.014 0.847 0.252 0.434 0.291 0.453 0.185 0.387 0.078 0.267 0.009 0.092 0.093 0.291 0.343 0.474 39.558 4.404 -0.074 0.975 0.093 0.997 0.036 0.187 0.245 0.427 0.161 0.364 0.130 0.335 0.095 0.288 0.369 0.477 0.204 0.403 0.317 0.465 2.310 0.777 0.643 0.241
Table 2. Summary statistics of outcome measures for the subsamples. All
Mean Children with below median household expenditure Number of children 8733 Number of census districts 1571 No school lunch 0.188 Height (cm): elementary 138.012 Height (cm): junior high 157.690 BMI: elementary 17.238 BMI: junior high 19.669 Obesity: IOTF BMI 25+: elementary 0.076 Obesity: IOTF BMI 25+: junior high 0.061 Obesity: POW 20%+: elementary 0.077 Obesity: POW 20%+: junior high 0.069 Underweight: IOTF BMI 18.5-:elementary 0.156 Underweight: IOTF BMI 18.5-: junior high 0.151 Underweight: POW 20%-: elementary 0.019 Underweight: POW 20%-: junior high 0.015 Children who do not have white-collar fathers Number of children 11426 Number of census districts 1882 No school lunch 0.206 Height (cm): elementary 138.233 Height (cm): junior high 157.754 BMI: elementary 17.296 BMI: junior high 19.668 Obesity: IOTF BMI 25+: elementary 0.078 Obesity: IOTF BMI 25+: junior high 0.060 Obesity: POW 20%+: elementary 0.080 Obesity: POW 20%+: junior high 0.068 Underweight: IOTF BMI 18.5-: elementary 0.152 Underweight: IOTF BMI 18.5-: junior high 0.154 Underweight: POW 20%-: elementary 0.021 Underweight: POW 20%-: junior high 0.017
38
Treatment (No School Lunch Areas) Mean SD
Control (School Lunch Areas) Mean SD
. . 0.390 8.572 7.774 2.479 2.700 0.265 0.239 0.266 0.253 0.363 0.358 0.136 0.123
1639 314 1 138.138 157.812 17.072 19.717 0.060 0.077 0.056 0.088 0.165 0.155 0.017 0.016
. . 1 8.455 7.691 2.252 2.806 0.237 0.267 0.231 0.283 0.371 0.363 0.129 0.125
7094 1257 0 137.983 157.662 17.275 19.658 0.080 0.057 0.081 0.064 0.154 0.150 0.019 0.015
. 0 8.599 7.795 2.526 2.674 0.271 0.232 0.273 0.245 0.361 0.357 0.138 0.122
. . 0.404 8.491 7.701 2.506 2.686 0.269 0.237 0.271 0.252 0.360 0.361 0.142 0.130
2352 405 1 138.264 157.950 17.144 19.749 0.067 0.069 0.072 0.078 0.157 0.147 0.016 0.019
. . 1 8.340 7.525 2.377 2.683 0.250 0.254 0.258 0.268 0.364 0.354 0.125 0.137
9074 1477 0 138.225 157.703 17.336 19.647 0.081 0.057 0.082 0.066 0.151 0.156 0.022 0.017
. . 0 8.531 7.746 2.537 2.687 0.273 0.233 0.275 0.248 0.358 0.363 0.146 0.129
SD
Table 3. Summary statistics for the census district level data. All
Variable N No school lunch Year Prefectural population density (1,000 person/km2) Municipal size: 11 largest cities Municipal size: cities with 150k+ population Municipal size: cities with 50-150k population Municipal size: cities with 50k- population Municipal size: towns & villages (reference) Region block: Hokkaido & Tohoku Region block: Kanto Region block: Chubu (reference) Region block: Kinki Region block: Chugoku & Shikoku Region block: Kyushu & Okinawa Area: #participants Area: proportion of age 1-19 Area: proportion of age 20-39 (reference) Area: proportion of age 40-59 Area: proportion of age 60+ Area: HH median % ranking of HH expenditure Area: mean HH size Area: proportion of white collar worker (reference) Area: proportion of laborer Area: proportion of self-employed Area: proportion of agriculture/fisheries/forestry Area: proportion of working women Area: mean child height (z score) Area: mean child BMI (z score) Area: child obesity rate (IOTF BMI 25+) Area: child underweight rate (IOTF BMI 18.5-)
Treatment (No School Lunch Areas) Mean SD 1648 1 1 1982.74 5.64 0.509 0.433 0.182 0.386 0.363 0.481 0.224 0.418 0.096 0.295 0.135 0.342 0.171 0.377 0.081 0.273 0.344 0.476 0.064 0.245 0.090 0.286 0.250 0.433
Mean 2116 0.221 1983.84 0.473 0.064 0.273 0.219 0.101 0.343 0.181 0.166 0.303 0.028 0.116 0.206
SD 468 0.415 5.59 0.443 0.245 0.446 0.414 0.302 0.475 0.385 0.372 0.460 0.165 0.320 0.405
84.49
27.06
81.29
0.311 0.260 0.275 0.153 0.556 4.261 0.374 0.345 0.192 0.089 0.601 -0.004 0.008 0.160 0.114
0.070 0.076 0.071 0.087 0.190 0.623 0.208 0.188 0.153 0.167 0.196 0.365 0.375 0.136 0.110
0.315 0.263 0.277 0.145 0.503 4.152 0.386 0.345 0.209 0.059 0.574 0.015 0.000 0.162 0.118
Control (School Lunch Areas) Mean SD 0 1984.16 0.462 0.031 0.248 0.217 0.103 0.402 0.184 0.190 0.292 0.018 0.123 0.194
0 5.54 0.446 0.173 0.432 0.412 0.303 0.490 0.387 0.392 0.455 0.132 0.329 0.395
27.52
85.40
26.87
0.067 0.075 0.071 0.079 0.184 0.582 0.207 0.192 0.152 0.142 0.188 0.360 0.395 0.142 0.115
0.309 0.260 0.275 0.156 0.571 4.291 0.370 0.345 0.187 0.098 0.608 -0.010 0.010 0.159 0.113
0.070 0.076 0.071 0.089 0.189 0.630 0.208 0.188 0.153 0.173 0.197 0.366 0.370 0.134 0.108
Notes: The number of observation is smaller than the number of districts in the sample of children because 148 districts with less than five preschool and elementary school children are removed from this sample.
39
Table 4. Estimated coefficients from Logit regression of No School Lunch Area dummy. Variable Prefecture dummies Region block dummies Year 1976 Year 1977 Year 1978 Year 1979 Year 1980 Year 1981 Year 1982 Year 1983 Year 1984 Year 1985 Year 1986 Year 1987 Year 1988 Year 1989 Year 1990 Year 1991 Year 1992 Year 1993 Year 1994 Prefectural population density (1,000 person/km2) Municipal size: 11 largest cities Municipal size: cities with 150k+ population Municipal size: cities with 50-150k population Municipal size: cities with 50k- population Area: # participants Area: proportion of age 1-19 Area: proportion of age 40-59 Area: proportion of age 60+ Area: HH median % ranking of HH expenditure Area: mean HH size Area: proportion of laborer Area: proportion of self-employed Area: proportion of agriculture Area: proportion of working women Area: mean child height (z score) Area: mean child BMI (z score) Area: child obesity rate (IOTF BMI 25+) Area: child underweight rate (IOTF BMI 18.5-)
Model1 (N=2074) Yes No 0.088 (0.335) 0.199 (0.377) -0.139 (0.366) -0.181 (0.364) -0.36 (0.364) -0.335 (0.385) -0.344 (0.373) -1.126** (0.439) -0.746* (0.391) -0.641 (0.4) -1.088*** (0.395) -0.754* (0.408) -0.685* (0.397) -1.364*** (0.465) -0.970** (0.44) -0.415 (0.408) -0.729* (0.431) -1.155** (0.466) -1.051** (0.522) -3.901** (1.613) 3.421*** (0.328) 1.684*** (0.212) 1.402*** (0.221) 1.109*** (0.256) -0.001 (0.003) 1.504 (1.54) 1.296 (1.059) 2.120* (1.264) -0.447 (0.46) -0.125 (0.158) 0.071 (0.401) 0.231 (0.487) -0.759 (0.67) 0.192 (0.449) 0.05 (0.181) -0.111 (0.232) 0.235 (0.528) 0.026 (0.687)
Model2 (N=2,116) No Yes 0.055 (0.299) 0.088 (0.319) -0.248 (0.318) -0.253 (0.333) -0.424 (0.318) -0.318 (0.335) -0.536 (0.333) -1.135*** (0.383) -0.849** (0.346) -0.709** (0.344) -1.161*** (0.345) -1.000*** (0.356) -0.788** (0.338) -1.395*** (0.413) -0.958*** (0.368) -0.718** (0.356) -0.845** (0.371) -1.314*** (0.418) -1.206*** (0.447) 0.09 (0.159) 2.912*** (0.273) 1.602*** (0.192) 1.375*** (0.2) 1.090*** (0.235) -0.001 (0.003) 1.278 (1.409) 1.09 (0.962) 1.930* (1.121) -0.645 (0.404) -0.032 (0.136) 0.191 (0.372) 0.088 (0.441) -0.298 (0.615) 0.592 (0.389) 0.196 (0.16) -0.074 (0.202) 0.411 (0.491) -0.278 (0.623)
Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The standard errors are in parenthesis. The constant is included in the model but omitted from the table. 40
Table 5. The estimated ATEs from the DID regression analysis. Panel (a): Simple DID analysis Sample
# children
# districts
Full sample
18,289
2,264
Children with low household expenditure Children without white-collar fathers
8,733
1,571
11,426
1,882
Height 0.016 (0.216) -0.068 (0.331) 0.13 (0.27)
BMI 0.068 (0.088) 0.316** (0.141) 0.289** (0.116)
Obesity (IOTF) 0.006 (0.009) 0.045*** (0.014) 0.026** (0.012)
Obesity (POW) 0.006 (0.01) 0.052*** (0.015) 0.022* (0.013)
Underweigh (IOTF) 0.002 (0.014) -0.01 (0.021) -0.014 (0.017)
Obesity (IOTF) 0.007 (0.011) 0.041** (0.018) 0.025* (0.015)
Obesity (POW) 0.008 (0.012) 0.043** (0.018) 0.025 (0.016)
Underweigh (IOTF) 0.007 (0.016) -0.005 (0.025) -0.015 (0.021)
Panel (b): DID analysis with propensity score trimming Sample
# children
# districts
Full sample
11,215
1,418
Children with low household expenditure Children without white-collar fathers
4,936
951
6,547
1,114
Height 0.123 (0.255) 0.07 (0.385) 0.161 (0.333)
BMI 0.094 (0.109) 0.422** (0.175) 0.367** (0.145)
Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The standard errors are in parenthesis.
41
Table 6. The estimated ATEs by age group. Panel (a): Simple DID analysis Sample Full sample
#children 21,514
#districts 2,264
ATE by age group Age 12-13 Age 14-15 Post junior high school
Children with low household expenditure
10,187
1,623
Age 12-13 Age 14-15 Post junior high school
BMI 0.092 (0.114) 0.042 (0.108) 0.032 (0.123) 0.315* (0.183) 0.355** (0.171) 0.230 (0.22)
Obesit (IOTF) 0.00 (0.01 0.01 (0.01 0.00 (0.01 0.046 (0.02 0.051 (0.01 0.02 (0.02
Panel (b): DID analysis with propensity score trimming Sample Full sample
#children 13,332
#districts 1,433
ATE by age group Age 12-13 Age 14-15 Post junior high school
Children with low household expenditure
5,875
995
Age 12-13 Age 14-15 Post junior high school
BMI 0.118 (0.137) 0.057 (0.134) 0.193 (0.147) 0.407* (0.222) 0.432** (0.212) 0.327 (0.274)
Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The standard errors are in parenthesis.
42
Obesit (IOTF) 0.01 (0.01 0.00 (0.01 0.00 (0.01 0.06 (0.02 0.03 (0.02 0.01 (0.02
Appendices Appendix 1. Details of the School Lunch Program
Trends in school lunch provision We examine trends in school lunch provision using official statistics from the School Lunch Data Book (SLDB) published by National School Health Center of Japan and Current Status Survey on School Meals conducted by MEXT. 1 We calculate annual nationwide and prefectural school lunch coverage rates for municipal school students based on SLDB for years 1978-1980, 1982, 1985, 1987, 1989-1997, and 2000-2004, and School Lunch Implementation Survey for years 2006-2015. Municipalities provide either “complete school lunch”, “complementary school lunch” where dishes except for staple food are provided and students bring steamed rice or bread from home, or “milk-only school lunch” where only milk is provided and students bring food from home. We calculate the school lunch coverage rate as the sum of the proportion of students with “whole-meal school lunch” and that with “complementary school lunch”, but the latter has always been less than 1% for both elementary school students and junior high school students. The school lunch coverage rate for municipal elementary school students has always been above 98% since 1978 and was about 100% in 2015. Figure A1 shows trends in the nationwide school lunch coverage rate for municipal junior high school students and in the sample proportion of junior high school students in School Lunch Area (defined in Subsection 4.1). Both share similar increasing trends, although the latter is always significantly higher than the former. This difference between our sample and the nationwide statistics can be understood by comparing prefectures included and excluded in our sample in Figure A2. Geographic variation in school lunch provision
1
www.mext.go.jp/b_menu/toukei/chousa05/kyuushoku/1267027.htm 1
Figure A2 shows geographic variation in the prefecture-level school lunch coverage rate for municipal junior high school students in 1985, the mid-year of our study period, by gradation of pink color. Prefectures with small and large rates are almost evenly distributed across Japan, while there appears to be some spatial correlation. The dotted areas represent prefectures excluded from our sample due to their high proportion of junior high school students attending private or national schools (Chiba, Hiroshima, Hyogo, Kanagawa, Kochi, Kyoto, Mie, Nara, Osaka, and Tokyo). In the vast majority of these excluded prefectures the rates are low, which can account for the higher school lunch coverage rate for junior high school students in our sample than in the nationwide statistics. Optional school lunch programs Since the 1990s Ministry of Education allows optional school lunch programs where students can choose between school lunch and home-prepared lunch as a temporary measure for schools that newly start school lunch programs (Asahi Shimbun 1996). Among municipalities included in our data, Nagoya city started such optional school lunch programs on trial at seven of its municipal junior high schools in 1993 (Asahi Shimbun, 1993). Thus, we exclude children living in Nagoya in 1993 and 1994 from our sample based on prefecture and municipal population size information. Two cities in Chiba prefecture, Funabashi and Matsudo, also started optional school lunch programs before 1994 (Asahi Shimbun 1993), but our sample does not include children in Chiba prefecture. Revisions in the nutritional standards Since 1954 the nutritional standards for school lunch have been revised eight times, mostly for minor changes. During our study period the revision occurred only once in 1986. In the 1986 revision, target values for fat was replaced with the maximum percentage of energy intake from fat of 30 percent, and the target amounts for energy and protein were slightly reduced (Nozue 2011). Nevertheless, the
2
revision did not cause significant changes in the energy or fat contained in school lunch (Narusaka 1996). Appendix 2. Details of the Data Construction
Identifying elementary school and junior high school students The NNS questionnaire asks if children are in compulsory education, but does not distinguish between elementary school and junior high school. Because the school grade is strictly determined by the child’s age on April 1st in Japan, among children in compulsory education we categorize 6- to 11year-olds as elementary school students and 13- to 15-year-olds as junior high school students. Additionally, for years 1986-1994 the birth month is available from the Comprehensive Survey of Living Conditions (CSLC), another annual survey conducted by the MHLW since 1986, because the NNS participants are the subpopulation of the participants in the CSLC. Because the NNS is conducted in November, we categorize 12-year-olds born between April and October as elementary school students and those born between December and March as junior high school students. We exclude 12-year-olds for years 1975-1985 and 12-year-olds born in November for years 1986-1994 from sample, because we cannot determine whether they attend elementary school or junior high school. Exclusion criteria Our sample consists of 9- to 15-year-old children in elementary school or junior high school, excluding 12-year-olds for years 1975-1985 and 12-year-olds born in November for years 1986-1994 as described above. We exclude children in prefectures with high proportion of junior high school students attending non-municipal schools as described in subsection 3.2, and children in Nagoya city in 1993 and 1994 as described in Appendix 1. We limit our sample to children with valid information on height and weight, and exclude a small number of children whose height is less than 100cm and/or whose weight is 3
larger than 100kg as deviants. A small number of children without mother in household or without valid report on household expenditure are also excluded from sample. We further limit our sample to children in census districts with at least one elementary school student and one junior high school student because our identification is based on DID. Finally, as described in subsection 3.2 we exclude children in census districts with unreliable school lunch information due to too few and/or conflicting reports and those in a small number of census districts where less than half of elementary school children report having school lunch. Figure A3 presents the sample size remaining after each exclusion criterion imposition. Reasons for the time frame restriction The individual-level data from the NNS is available for both genders from 1975. We do not use the NNS data collected after 1994 for the following reasons. First, since 1995 the food diary covers only one day of each household’s choice, excluding Sundays and holidays, and the day of week is not reported. Because school lunch is not served on Saturdays, if Saturday is chosen for food diary, no information is collected regarding whether children regularly have school lunch. Second, as described in Appendix 1 the increase in optional school lunch programs since the late 1990s makes it difficult to examine the causal effect of school lunch. Municipal Provision of School Lunch We complement our categorization regarding municipal provision of school lunch with official statistics from the SLDB, and we linearly impute for the missing years in the SLDB. We regard children in prefectures with 99% or higher school lunch coverage as having school lunch. During our study period about 0.5-0.7% of Japanese municipal junior high school students have “complementary school lunch”, and about 19.4-26.5% have “milk only school lunch”. This heterogeneity might lead to
4
ambiguity about the definition of “having school lunch” and to reporting heterogeneity. To examine this possibility and the reliability of the respondents’ reports on school lunch in the NNS, we compare the reports with the official statistics. First, we construct prefecture-year level data of the proportion of junior high school students who report having school lunch in our sample. For this analysis we include districts with too few and/or conflicting reports and districts where less than half of elementary school students report having school lunch. Second, we merge them with the SLDB data on the proportion of municipal junior high school students who have complete, complementary, and milk-only school lunch. Finally, we regress the former variable constructed from our sample on the latter variables obtained from the SLDB. The estimation results are shown in Table A1. The coefficients of the proportions of students with complete and complementary school lunch are both significantly positive and close to one, implying that the vast majority of students having complete or complementary school lunch report having school lunch. The coefficient of the proportion of students with milk-only school lunch is insignificant and small in magnitude, implying that few students having milk-only school lunch report having school lunch. The constant is also insignificant and small in magnitude, suggesting little systematic reporting bias. These findings support the reliability of the NNS data and our methods. We also examine the distribution of the census-district level percentage of students who report having school lunch in our sample. For this analysis we include districts with too few and/or conflicting reports and districts where less than half of elementary school students report having school lunch. Table A2 shows a histogram. We first examine whether the majority of students report having school lunch in districts with 98% or higher prefectural rate of school lunch coverage according to the SLDB. For elementary school students the percentage equals 100 in 81.4% districts and 50 or more and less than 100 in 17.4% districts, and for junior high school students the percentage equals 100 in 90.6% and 50 or more and less than 100 in 9.0% districts, suggesting that the major cause of the within-district 5
reporting discrepancies is students’ nonattendance. The histogram for the full sample of elementary school students is highly similar to that for the subsample. In the full sample of junior high school students, within-district reporting discrepancy arises in 21.3% of the students, but in 20.3% the percentage of positive reporting is 50 or more and less than 100, suggesting that the district provides school lunch at municipal junior high schools but some students do not eat school lunch during the survey period due to nonattendance. Appendix 3. Robustness Check: Relaxing the Exclusion Criteria
Children in prefectures with 5% or more of junior high school students attending private and/or national schools as of 1994 are omitted from our sample. As a robustness check, we relax this exclusion criterion and include children in prefectures where the proportion was 5% or more but less than 10%. The majority of the excluded prefectures meet this criterion (Chiba, Hiroshima, Hyogo, Kanagawa, Kyoto, Mie, Nara, and Osaka), with the exception of Kochi (15.6%) and Tokyo (22.9%). The estimated average treatment effects (ATEs) from the DID regression analysis are shown in Table A3. The estimated ATEs are highly similar to those with our preferred sample shown in Table 5. We do not conduct propensity score trimming for this sample because the trimming methods used for our preferred sample (i.e., the logit model and cutoff values specified in Subsection 4.5) fail to reduce the absolute standardized difference of some of the explanatory variables between the target and control areas to a satisfactory level.
References Asahi Shimbun, 1993, October 27. The Optional School lunch programs at junior high Schools in Nagoya Had a 60% Utilization Rate, Falling below Expectations. Retrieved from Asahi Shimbun Kikuzo database. (In Japanese) 6
Asahi Shimbun, 1996, August 27. Shaken by O157: Turning Point for Uniformly Imposed School Lunch. Retrieved from Asahi Shimbun Kikuzo database. (In Japanese) Narusaka M, 1996. Changes in fat supply in school lunch menus in Okayama City. Japanese Journal of Nutrition and Dietetics, 54(2):121-128 (in Japanese). Nozue M, 2011. The contribution of school lunch to dietary intakes in school children and application of dietary reference intakes: A case study of fifth-grade school children. Ph.D. Thesis, Kagawa Nutrition University (in Japanese).
7
Figure A1. Trends in school lunch coverage rate among municipal junior high school students.
8
Figure A2. Map with school lunch coverage rate for municipal junior high school students in 1985
less than 20% 20-40% 40-60% 60-80% 80-100%
Notes: Dotted areas represent prefectures excluded from our sample.
9
Figure A3. Sample exclusion criteria and sample sizes. 9- to 15-year-old children in compulsory education, excluding 12year-olds for years 1975-1985 and 12-year-olds born in November for years 1986-1994
5,580 census districts, 37,990 children
Exclude children in prefectures with high proportion of junior high school students in non-municipal schools and children in Nagoya city in 1993 and 1994
3,280 census districts, 24,530 children
Exclude children without valid information on height and weight
3,228 census districts, 22,241 children
Exclude children with deviant height and weight
3,228 census districts, 22,229 children
Exclude children without mother in household or without valid report on household expenditure
3,206 census districts, 21,349 children
Exclude children in census districts without elementary school students or without junior high school students
2,660 census districts, 20,083 children
Exclude children in census districts with too few and/or conflicting reports on school lunch
2,310 census districts, 18,687 children
Exclude children in census districts without school lunch for elementary school children
2,264 census districts, 18,289 children
10
Table A1. Pooled regression of proportions of junior high school students reporting having school lunch. Variable Proportion of students with complete school lunch Proportion of students with complementary school lunch Proportion of students with milk only Constant N
Coefficients 0.893*** 0.944** 0.032 8.090 427
Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The unit of observation is year-prefecture.
11
Table A2. A histogram of census-district level proportions of students reporting having school lunch. % students who had school lunch during the survey period
0% more than 0% and less than 50% 50% or more and less than 100% 100%
Census districts with 98% or higher prefectural school lunch coverage rate for elementary school or junior high school Elementary Junior high School School
0.6 0.6 17.4 81.4
0.4 0.0 9.0 90.6
Full sample
Elementary School
Junior high School
1.6 1.0 20.3 77.0
20.8 2.7 10.0 66.5
Notes: The unit of observation is year-census district. The sample for this histogram includes districts where 50% or less of elementary school students had school lunch during the survey period, while we exclude them from the sample for the DID regression analysis.
12
Table A3. Regression analysis after relaxing the exclusion criteria. Sample
# children
# districts
Height
Full sample
24,480
3,147
Children with low household expenditure Children without white-collar fathers
10,417
1,969
14,500
2,493
-0.035 (0.17) 0.03 (0.271) 0.071 (0.227)
13
BMI 0.001 (0.066) 0.231** (0.113) 0.206** (0.092)
Obesity (IOTF) 0.003 (0.007) 0.041*** (0.012) 0.023** (0.01)
Obesity (POW) 0.006 (0.007) 0.049*** (0.012) 0.024** (0.01)
Underweight (IOTF) 0.001 (0.011) -0.007 (0.017) -0.011 (0.014)
Underweight (POW) 0.007 (0.004) 0.007 (0.006) 0.012** (0.006)
