Might extended education decrease inequalities in health?
587
......................................................................................................... European Journal of Public Health, Vol. 25, No. 4, 587–592 ß The Author 2015. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. doi:10.1093/eurpub/cku243 Advance Access published on 24 January 2015
.........................................................................................................
Might extended education decrease inequalities in health?—a meta-analysis Sofia Ljungdahl1, Sven G. Bremberg1,2 1 The Public Health Agency of Sweden, Stockholm, Sweden 2 Department of Public Health Sciences, Karolinska Institute, Stockholm, Sweden Correspondence: Sven Bremberg, Fa¨rgarga˚rdstorget 52, 11643 Stockholm, Sweden. Tel: +46 70 68 99753, Fax: +46 8328330, e-mail: [email protected]
.........................................................................................................
Introduction 1
ess educated people have poorer health. Thus, it has been
Lsuggested that health in these people might be improved by 1–4
extending their education. There are a large number of potential causal chains that might explain educational effects on health.5 Many are unknown, but it is possible to discern two groups of causal chains. In a first group, education improves the ‘absolute’ competences of the individual, independently of other individuals.6 Education might improve individual ability to interpret information and to translate acquired knowledge into practice, including any health-promoting practice.2 Thus, education might increase the individual’s ability to make use of a resource. This effect does not depend on any effect of education on other individuals. Two sociologists, Mirowsky and Ross, describe this effect of education as boosting individual ‘healthefficiency’.6 In the second ‘relative’ causal chain, education provides individuals with abilities that they can use in competition with other individuals. Thus, a long education increases opportunities for privileged employment, with a high salary and high social prestige.7 Such resources contribute to health but are constrained; that is, if one individual has access to more, others get less. Education, understood in this way, contributes to the distribution of resources among individuals but does not increase collective assets. Thus, in order to improve individual health in the least educated group, without deteriorating health in other groups, absolute effects of education are required. It is, however, not clear whether absolute or relative educational effects on health are predominant. Accordingly, the issue has to be clarified.
Methods for differentiating between absolute and relative effects of education The effects of education on health outcomes can be analysed in cohort studies. Such a study design, however, faces two important problems. The first is confounding.8 Statistical methods will control for confounders, but many potential confounders are not known. When a potentially important variable is not included, the effects of education on health might be overestimated. The second refers to the large number of long potential causal chains, which are only partly understood. These two problems make cohort studies less able to determine whether absolute or relative effects of education predominate. Studies of natural experiments may be more helpful in elucidating the significance of the absolute effects of education.9 During the 1900s, the number of years in education increased in most countries, partly due to reforms that extended compulsory education. The reforms affected only students who otherwise would have left education at the end of compulsory education, and did not affect the significant number of young people who, both before and after the reform continued into post-compulsory education. Accordingly, such reforms are useful for analysing the effects of extended education on individuals with the least education. These students still have the least education and, therefore, the extra education will not give them any competitive advantage within their own cohort. They will, however, to some extent also compete with older, less educated cohorts. Thus, any health effect of an educational reform of this kind might partly be absolute. Some of the twentieth century educational reforms intended to facilitate secondary level education after the end of compulsory education. These reforms also affected individuals with low education. Yet, since secondary level education was voluntary, not
Downloaded from http://eurpub.oxfordjournals.org/ by guest on November 14, 2015
Background: Health is substantially worse in less educated people, and extended education might potentially improve their health. A prerequisite for a beneficial health effect of education is that the effect is absolute. An absolute effect of education means that the health effect comes about independently of any effect on other persons. A relative effect, on the other hand, only contributes to individual competitiveness in relation to others. Studies of natural experiments of extended compulsory education, and other educational-policy changes, provide an option for the analysis of absolute effects of education. Published studies, however, present conflicting results. Methods: A meta-analysis was performed of European studies where the health effects of extended compulsory or secondary level education on low-educated individuals were investigated. Results: Twenty-two relevant publications were identified. The meta-analysis indicated statistically significant favourable effects of educational reforms on rates of mortality, self-reported poor health and obesity. The effects were, however, small, 1-4%. Conclusions: An educational reform that typically added one educational year in the least educated group was associated with a mean 2.1% reduction in mortality in men before age 40. This effect might be compared with the total educational gradients of mortality rates in Swedish men at ages 30-64. One extra year of education after compulsory education corresponds to a 41% reduction in mortality, which is 20 times more than the absolute effect of education found in this meta-analysis. Thus, it unlikely that extended compulsory education will substantially improve the health of the least educated individuals.
588
European Journal of Public Health
Aim The aim of the study is to address the question: can the health of the least educated be improved by extended education?
Methods In order to identify relevant studies, the following databases were searched: EMBASE (all fields + text), Eric (anywhere), Global Health (all fields), IDEAS (whole record, word forms: exact), PubMed (all fields), Scopus (all fields) and Web of Science (topic), with the search string: (‘compulsory school’ OR ‘compulsory schooling’) AND (alcohol OR ‘cardiovascular disease’ OR exercise OR health OR hospitalization OR hypertension OR memory OR mortality OR obesity OR ‘self-rated health’ OR smoking OR ‘physical activity’). Google Scholar was also used, with the search string: ‘natural experiment’ AND ‘compulsory school’ AND (alcohol OR ‘cardiovascular disease’ OR exercise OR health OR hospitalization OR hypertension OR mortality OR obesity OR ‘physical activity’ OR ‘self-rated health’ OR smoking). The searches were performed in October/November 2013. In PubMed, MESH terms were also tried, but no additional studies were retrieved.
First, the titles were screened, and then the abstracts of promising titles. This was followed by the reading of full articles. The screening and reading was done in parallel by the two authors. Screening of the reference lists of the relevant articles completed the search. The inclusion criteria were as follows: (1) The study reported effects on health or health behaviours of extended primary or secondary level education by means of changes in educational laws and policies in year 1900 or later. (2) The study populations were living in Europe and were affected by educational reforms at ages 10-20. (3) The study reported results from analyses in which an IV, an RD, or a similar approach was adopted. Studies that only reported on linear regression models, like ordinary least squares (OLS), were excluded. (4) The study reported results for the two genders separately, because females and males experience disparate educational outcomes and fulfil gender-specific roles, both during education and in larger society.18 (5) The study presented original data and was published in English in a peer-reviewed journal or in a published working paper. If a study was published two or more times, only the most recent publication was included. Meta-analyses were performed for outcomes that had been analysed in at least five studies. The educational-reform effects were usually presented as regression coefficients. Studies of the effects on mortality were presented as odds ratios (ORs) or hazard ratios (HRs), which were regarded as equal. In studies of mortality, regression coefficients were transformed into OR = EXP(). The educational-reform effect and the standard error (SE) of a given reform were extracted from each published paper. When only 95% confidence intervals (CIs) were presented, the intervals were approximated to represent 2 SE. A random-effect model was used to get a weighted average of the separate study estimates.19 The random-effect model was chosen because the material is heterogeneous, with different states of health in the study populations and varying proportions affected by the educational reforms. CMA software was used (www.Meta-Analysis.com). Results are presented as means and 95% CI. In epidemiological studies, selective attrition is an important source of bias. Yet, this was not regarded as an important problem in this analysis, because the methods used for collecting information were identical in the unexposed and the exposed cohorts.
Results The numbers of hits obtained in the searched databases are given within brackets: EMBASE (199), Eric (26), Global Health (57), Google Scholar (558), IDEAS (35), PubMed (281), Scopus (165) and Web of Science (105). The titles, and easily available abstracts (as in PubMed), were screened by both authors. Thirty-three potential articles were identified for full text reading. The two authors agreed to include 22 publications and exclude 11 (see Supplementary tables S1 and S2). The statistical models that were judged to meet the inclusion criteria, as well as being based on unmodified IV and RD approaches, were: the second stage results of two-stage leastsquares computations (2SLS); the results of two-stage conditional maximum likelihood estimation including a random variable (2SCMLR)20; a IVprobit model21; a Fuzzy RD design22 and a reduced form difference in difference/IV model.23,24 Three studies exploited variation between school districts within one country in a reduced form of linear probability model considered to mirror the RD approach.25 Educational reforms had been implemented in 18 different countries between 1903 and 1990, and health outcomes were assessed in 1990-2011. The following outcomes were presented in at least five studies: body mass index (BMI)/obesity, cognitive
Downloaded from http://eurpub.oxfordjournals.org/ by guest on November 14, 2015
all will have participated. After finishing their education, adolescents with a longer education, due to the reform, competed with those with only compulsory education. Thus, the longer-educated individuals obtained a relative advantage, and any health effect of reform might partly be regarded as a relative effect. Several analyses of the health effects of educational reforms have been published. Most deal with the extension of compulsory education during the 1900s. In the analyses, adult health in cohorts affected by reform is compared with adult health in previous, unaffected cohorts. Any analysis of the health effects of an educational reform has to take into account the continuous improvement in health that occurred in most countries during the 1900s. A commonly used statistical method to achieve that end is regression discontinuity (RD).10 Researchers construct a time trend for the development of health in birth cohorts affected by a reform and also a separate trend for earlier unaffected cohorts. Any difference in time trends at the time of an educational reform is ascribed to the educational reform. Another statistical method, which has primarily been used in economics but has recently appeared in epidemiological studies,11 is based on the instrumental variables (IVs) approach, which adjusts for both observed and unobserved confounders. Yet, these methods do not preclude the drawing of fallacious conclusions. An educational reform might have been caused by rapid economic and social development at a particular point in time. Any health effect at the time of the reform might be attributable to the economic and social development rather than to the educational reform.12 Economic and social conditions, however, develop continuously, whereas the methods used analyse a break in a time trend. Accordingly, this explanation is not a likely one, but the health effect of the educational reform might still be inflated. Recent reviews report conflicting results from natural experiments of educational reforms, with some studies indicating favourable health effects, but others not.13–16 The conflicting results have not been explained by varying contexts, different initial levels of education, the fractions of individuals affected by the reforms, the outcomes that were studied or the statistical methods that were employed. One meta-analysis of educational reforms in Europe has been presented.17 It suggests that educational reforms result in a 2.9% reduction in mortality in men below 40 years of age but have no significant effect on women. The authors, however, only included the outcomes of analyses that they performed themselves, and only reported effects on mortality. Thus, an additional, more comprehensive, meta-analysis is called for.
Might extended education decrease inequalities in health?
589
Table 1 Effects of school reforms on mean BMI, BMI = 25 to < 30 and BMI 30 age 20-60 References
School reform
Braga and Bratti22 Brunello et al.
21
Brunello, 2013 26
Compulsory Compulsory
Compulsory
Statistical model
Fuzzy RD design IV probit (window 3 years) 2SLS
Secondary level
2SLS (age left school) 2SLS
Ju¨rges et al.27 Kemptner, 2011
Secondary level Compulsory
Mazzonna, 2014
Compulsory
2SLS
Reinhold and Ju¨rges28
Secondary level
2SLS
2SLS
Mean effect (95% CI) Heterogeneity I2
Overweight (BMI = 25 to < 30)
Obesity (BMI 30)
Female effect (SE)
Male effect (SE)
Female effect (SE)
Male effect (SE)
Female effect (SE)
Male effect (SE)
3.254 (1.053) 0.479 (0.236)
0.263 (1.172)
0.285 (0.195)
0.041 (0.253)
0.243 (0.063) 0.093 (0.096)
0.144 (0.147)
0.414 (0.149) 1.028 (0.493)
0.73 (0.178) 0.674 (0.379)
0.106 (0.040)
0.009 (0.068)
0.065 (0.006) 0.0644 (0.0379)
0.011 (0.081) 0.0616 (0.0386)
0.088 (0.063) 0.031 (0.015)
0.144 (0.167) 0.03 (0.015)
0.018 (0.038) 0.004 (0.010) 0.074 (0.046) 0.120 (0.070) 0.107 (0.215 to 0.000) 84.9
0.229 (0.166) 0.03 (0.014) 0.043 (0.064) 0.03 (0.059) 0.030 (0.054 to 0.006) 0
0.194 (0.133)
0.532 (0.924 to 0.141) 64.7
0.301 (0.121)
0.497 (0.776 to 0.218) 31.3
0.188 (0.109) 0.014 (0.100 to 0.071) 68.8
0.057 (0.078) 0.025 (0.053 to 0.003) 0
Effects are given as regression coefficients.
ability, mortality, self-rated poor health and smoking. All studies of mortality considered the effects of prolonging compulsory education. For the remaining outcomes, three studies26–28 that reported effects of secondary education reforms were also included. BMI were presented as a value, as prevalence of overweight (BMI 25 to
587
......................................................................................................... European Journal of Public Health, Vol. 25, No. 4, 587–592 ß The Author 2015. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. doi:10.1093/eurpub/cku243 Advance Access published on 24 January 2015
.........................................................................................................
Might extended education decrease inequalities in health?—a meta-analysis Sofia Ljungdahl1, Sven G. Bremberg1,2 1 The Public Health Agency of Sweden, Stockholm, Sweden 2 Department of Public Health Sciences, Karolinska Institute, Stockholm, Sweden Correspondence: Sven Bremberg, Fa¨rgarga˚rdstorget 52, 11643 Stockholm, Sweden. Tel: +46 70 68 99753, Fax: +46 8328330, e-mail: [email protected]
.........................................................................................................
Introduction 1
ess educated people have poorer health. Thus, it has been
Lsuggested that health in these people might be improved by 1–4
extending their education. There are a large number of potential causal chains that might explain educational effects on health.5 Many are unknown, but it is possible to discern two groups of causal chains. In a first group, education improves the ‘absolute’ competences of the individual, independently of other individuals.6 Education might improve individual ability to interpret information and to translate acquired knowledge into practice, including any health-promoting practice.2 Thus, education might increase the individual’s ability to make use of a resource. This effect does not depend on any effect of education on other individuals. Two sociologists, Mirowsky and Ross, describe this effect of education as boosting individual ‘healthefficiency’.6 In the second ‘relative’ causal chain, education provides individuals with abilities that they can use in competition with other individuals. Thus, a long education increases opportunities for privileged employment, with a high salary and high social prestige.7 Such resources contribute to health but are constrained; that is, if one individual has access to more, others get less. Education, understood in this way, contributes to the distribution of resources among individuals but does not increase collective assets. Thus, in order to improve individual health in the least educated group, without deteriorating health in other groups, absolute effects of education are required. It is, however, not clear whether absolute or relative educational effects on health are predominant. Accordingly, the issue has to be clarified.
Methods for differentiating between absolute and relative effects of education The effects of education on health outcomes can be analysed in cohort studies. Such a study design, however, faces two important problems. The first is confounding.8 Statistical methods will control for confounders, but many potential confounders are not known. When a potentially important variable is not included, the effects of education on health might be overestimated. The second refers to the large number of long potential causal chains, which are only partly understood. These two problems make cohort studies less able to determine whether absolute or relative effects of education predominate. Studies of natural experiments may be more helpful in elucidating the significance of the absolute effects of education.9 During the 1900s, the number of years in education increased in most countries, partly due to reforms that extended compulsory education. The reforms affected only students who otherwise would have left education at the end of compulsory education, and did not affect the significant number of young people who, both before and after the reform continued into post-compulsory education. Accordingly, such reforms are useful for analysing the effects of extended education on individuals with the least education. These students still have the least education and, therefore, the extra education will not give them any competitive advantage within their own cohort. They will, however, to some extent also compete with older, less educated cohorts. Thus, any health effect of an educational reform of this kind might partly be absolute. Some of the twentieth century educational reforms intended to facilitate secondary level education after the end of compulsory education. These reforms also affected individuals with low education. Yet, since secondary level education was voluntary, not
Downloaded from http://eurpub.oxfordjournals.org/ by guest on November 14, 2015
Background: Health is substantially worse in less educated people, and extended education might potentially improve their health. A prerequisite for a beneficial health effect of education is that the effect is absolute. An absolute effect of education means that the health effect comes about independently of any effect on other persons. A relative effect, on the other hand, only contributes to individual competitiveness in relation to others. Studies of natural experiments of extended compulsory education, and other educational-policy changes, provide an option for the analysis of absolute effects of education. Published studies, however, present conflicting results. Methods: A meta-analysis was performed of European studies where the health effects of extended compulsory or secondary level education on low-educated individuals were investigated. Results: Twenty-two relevant publications were identified. The meta-analysis indicated statistically significant favourable effects of educational reforms on rates of mortality, self-reported poor health and obesity. The effects were, however, small, 1-4%. Conclusions: An educational reform that typically added one educational year in the least educated group was associated with a mean 2.1% reduction in mortality in men before age 40. This effect might be compared with the total educational gradients of mortality rates in Swedish men at ages 30-64. One extra year of education after compulsory education corresponds to a 41% reduction in mortality, which is 20 times more than the absolute effect of education found in this meta-analysis. Thus, it unlikely that extended compulsory education will substantially improve the health of the least educated individuals.
588
European Journal of Public Health
Aim The aim of the study is to address the question: can the health of the least educated be improved by extended education?
Methods In order to identify relevant studies, the following databases were searched: EMBASE (all fields + text), Eric (anywhere), Global Health (all fields), IDEAS (whole record, word forms: exact), PubMed (all fields), Scopus (all fields) and Web of Science (topic), with the search string: (‘compulsory school’ OR ‘compulsory schooling’) AND (alcohol OR ‘cardiovascular disease’ OR exercise OR health OR hospitalization OR hypertension OR memory OR mortality OR obesity OR ‘self-rated health’ OR smoking OR ‘physical activity’). Google Scholar was also used, with the search string: ‘natural experiment’ AND ‘compulsory school’ AND (alcohol OR ‘cardiovascular disease’ OR exercise OR health OR hospitalization OR hypertension OR mortality OR obesity OR ‘physical activity’ OR ‘self-rated health’ OR smoking). The searches were performed in October/November 2013. In PubMed, MESH terms were also tried, but no additional studies were retrieved.
First, the titles were screened, and then the abstracts of promising titles. This was followed by the reading of full articles. The screening and reading was done in parallel by the two authors. Screening of the reference lists of the relevant articles completed the search. The inclusion criteria were as follows: (1) The study reported effects on health or health behaviours of extended primary or secondary level education by means of changes in educational laws and policies in year 1900 or later. (2) The study populations were living in Europe and were affected by educational reforms at ages 10-20. (3) The study reported results from analyses in which an IV, an RD, or a similar approach was adopted. Studies that only reported on linear regression models, like ordinary least squares (OLS), were excluded. (4) The study reported results for the two genders separately, because females and males experience disparate educational outcomes and fulfil gender-specific roles, both during education and in larger society.18 (5) The study presented original data and was published in English in a peer-reviewed journal or in a published working paper. If a study was published two or more times, only the most recent publication was included. Meta-analyses were performed for outcomes that had been analysed in at least five studies. The educational-reform effects were usually presented as regression coefficients. Studies of the effects on mortality were presented as odds ratios (ORs) or hazard ratios (HRs), which were regarded as equal. In studies of mortality, regression coefficients were transformed into OR = EXP(). The educational-reform effect and the standard error (SE) of a given reform were extracted from each published paper. When only 95% confidence intervals (CIs) were presented, the intervals were approximated to represent 2 SE. A random-effect model was used to get a weighted average of the separate study estimates.19 The random-effect model was chosen because the material is heterogeneous, with different states of health in the study populations and varying proportions affected by the educational reforms. CMA software was used (www.Meta-Analysis.com). Results are presented as means and 95% CI. In epidemiological studies, selective attrition is an important source of bias. Yet, this was not regarded as an important problem in this analysis, because the methods used for collecting information were identical in the unexposed and the exposed cohorts.
Results The numbers of hits obtained in the searched databases are given within brackets: EMBASE (199), Eric (26), Global Health (57), Google Scholar (558), IDEAS (35), PubMed (281), Scopus (165) and Web of Science (105). The titles, and easily available abstracts (as in PubMed), were screened by both authors. Thirty-three potential articles were identified for full text reading. The two authors agreed to include 22 publications and exclude 11 (see Supplementary tables S1 and S2). The statistical models that were judged to meet the inclusion criteria, as well as being based on unmodified IV and RD approaches, were: the second stage results of two-stage leastsquares computations (2SLS); the results of two-stage conditional maximum likelihood estimation including a random variable (2SCMLR)20; a IVprobit model21; a Fuzzy RD design22 and a reduced form difference in difference/IV model.23,24 Three studies exploited variation between school districts within one country in a reduced form of linear probability model considered to mirror the RD approach.25 Educational reforms had been implemented in 18 different countries between 1903 and 1990, and health outcomes were assessed in 1990-2011. The following outcomes were presented in at least five studies: body mass index (BMI)/obesity, cognitive
Downloaded from http://eurpub.oxfordjournals.org/ by guest on November 14, 2015
all will have participated. After finishing their education, adolescents with a longer education, due to the reform, competed with those with only compulsory education. Thus, the longer-educated individuals obtained a relative advantage, and any health effect of reform might partly be regarded as a relative effect. Several analyses of the health effects of educational reforms have been published. Most deal with the extension of compulsory education during the 1900s. In the analyses, adult health in cohorts affected by reform is compared with adult health in previous, unaffected cohorts. Any analysis of the health effects of an educational reform has to take into account the continuous improvement in health that occurred in most countries during the 1900s. A commonly used statistical method to achieve that end is regression discontinuity (RD).10 Researchers construct a time trend for the development of health in birth cohorts affected by a reform and also a separate trend for earlier unaffected cohorts. Any difference in time trends at the time of an educational reform is ascribed to the educational reform. Another statistical method, which has primarily been used in economics but has recently appeared in epidemiological studies,11 is based on the instrumental variables (IVs) approach, which adjusts for both observed and unobserved confounders. Yet, these methods do not preclude the drawing of fallacious conclusions. An educational reform might have been caused by rapid economic and social development at a particular point in time. Any health effect at the time of the reform might be attributable to the economic and social development rather than to the educational reform.12 Economic and social conditions, however, develop continuously, whereas the methods used analyse a break in a time trend. Accordingly, this explanation is not a likely one, but the health effect of the educational reform might still be inflated. Recent reviews report conflicting results from natural experiments of educational reforms, with some studies indicating favourable health effects, but others not.13–16 The conflicting results have not been explained by varying contexts, different initial levels of education, the fractions of individuals affected by the reforms, the outcomes that were studied or the statistical methods that were employed. One meta-analysis of educational reforms in Europe has been presented.17 It suggests that educational reforms result in a 2.9% reduction in mortality in men below 40 years of age but have no significant effect on women. The authors, however, only included the outcomes of analyses that they performed themselves, and only reported effects on mortality. Thus, an additional, more comprehensive, meta-analysis is called for.
Might extended education decrease inequalities in health?
589
Table 1 Effects of school reforms on mean BMI, BMI = 25 to < 30 and BMI 30 age 20-60 References
School reform
Braga and Bratti22 Brunello et al.
21
Brunello, 2013 26
Compulsory Compulsory
Compulsory
Statistical model
Fuzzy RD design IV probit (window 3 years) 2SLS
Secondary level
2SLS (age left school) 2SLS
Ju¨rges et al.27 Kemptner, 2011
Secondary level Compulsory
Mazzonna, 2014
Compulsory
2SLS
Reinhold and Ju¨rges28
Secondary level
2SLS
2SLS
Mean effect (95% CI) Heterogeneity I2
Overweight (BMI = 25 to < 30)
Obesity (BMI 30)
Female effect (SE)
Male effect (SE)
Female effect (SE)
Male effect (SE)
Female effect (SE)
Male effect (SE)
3.254 (1.053) 0.479 (0.236)
0.263 (1.172)
0.285 (0.195)
0.041 (0.253)
0.243 (0.063) 0.093 (0.096)
0.144 (0.147)
0.414 (0.149) 1.028 (0.493)
0.73 (0.178) 0.674 (0.379)
0.106 (0.040)
0.009 (0.068)
0.065 (0.006) 0.0644 (0.0379)
0.011 (0.081) 0.0616 (0.0386)
0.088 (0.063) 0.031 (0.015)
0.144 (0.167) 0.03 (0.015)
0.018 (0.038) 0.004 (0.010) 0.074 (0.046) 0.120 (0.070) 0.107 (0.215 to 0.000) 84.9
0.229 (0.166) 0.03 (0.014) 0.043 (0.064) 0.03 (0.059) 0.030 (0.054 to 0.006) 0
0.194 (0.133)
0.532 (0.924 to 0.141) 64.7
0.301 (0.121)
0.497 (0.776 to 0.218) 31.3
0.188 (0.109) 0.014 (0.100 to 0.071) 68.8
0.057 (0.078) 0.025 (0.053 to 0.003) 0
Effects are given as regression coefficients.
ability, mortality, self-rated poor health and smoking. All studies of mortality considered the effects of prolonging compulsory education. For the remaining outcomes, three studies26–28 that reported effects of secondary education reforms were also included. BMI were presented as a value, as prevalence of overweight (BMI 25 to
