The obesity loss effect of returns to education: Evidence from China

doi:10.21203/rs.3.rs-3400091/v1

Download PDF

Research Article

The obesity loss effect of returns to education: Evidence from China

https://doi.org/10.21203/rs.3.rs-3400091/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

As the world’s largest developing country, China’s overweight and obesity rate is higher than the global average and is growing. This paper investigates the effect of education on obesity and its mechanism to study the “weight loss” effect of education. The study finds that education has a nonremarkable effect on being overweight but a remarkable negative effect on obesity. Mechanism analysis showed that education can inhibit the occurrence of obesity by improving nutritional knowledge, enhancing health risk perception, improving diet regularity and increasing physical exercise. Heterogeneity analysis showed that the effect of education on obesity is more significant in rural residents and female samples compared with their counterparts. Further analysis revealed that education inhibits obesity mainly through improving nutritional knowledge for urban residents and female samples and by enhancing physical exercise for rural residents and male samples.

education

obesity

urban–rural differences

gender differences

Obesity is associated with many chronic noncommunicable diseases, such as diabetes, hypertension and cardiovascular diseases; affects the health and longevity of the population leading to a loss of national productivity and human capital and entails huge health care needs and expensive economic costs (Ayton and Ibrahim, 2019; James, 2018). Currently, more than 40% of the world’s adults are overweight or obese (WHO, 2020).

The reason for obesity is complex and includes congenital genetics, metabolic abnormalities, neuroendocrine disorders and other physiological factors as well as diet, physical exercise and other environmental factors. These factors especially environmental factors are more or less associated with various socioeconomic factors (Cockerham, 2022). At present, socioeconomic studies on the factors influencing obesity focused on the following three aspects. Firstly, at the micro level, individual characteristics, such as dietary habits, lifestyle, health awareness and gender, have different degrees of influence on obesity (Casey et al., 2008; Ding and Gebel, 2012; Durand et al., 2011; Kim, 2016; Menghetti et al., 2015). Secondly, at the community level, studies showed that obesity may be “contagious” in the community or amongst similar populations. For example, Christakis and Foweler (2007) found that the average BMI in the community has a significantly positive effect on individual BMI. This influence is transmitted through social network, association and environmental effects (de la Haye et al., 2011; Schaefer et al., 2011). Thirdly, at the macro level, the technological progress in the process of economic and social development, the level of urbanisation, environmental pollution, petrol prices and other socioeconomic factors may have different degrees of impact on obesity (Aiyar et al., 2021; Deschenes et al., 2020; Giuntella et al., 2020; Mary and Stoler, 2021; Philipson and Posner, 2003; Prakash et al., 2022).

Education and health have long been recognised as two important elements of human capital, and numerous studies showed that education and health are important contributors to economic growth and social development (Baldacci et al., 2008). However, an ongoing debate in the academic community about the determinants of the relationship between education and health within human capital remains (Liu et al., 2018; Dhrifi et al., 2021). Regarding obesity as a health concern, evidence on whether education can reduce the risk of obesity in people is limited and equally controversial. Some scholars believe that education can lower people’s body mass index (BMI), thus reducing the risk of obesity (Mazariegos et al., 2022; Zheng et al., 2022; Kong and Qi, 2017). Mazariegos et al. (2022) found a negative association between education and obesity in Latin American countries. Kong and Qi (2017) used data from the 2008 China Subhealth Survey and found that the inhibitory effect of education on obesity only works in the female population. Some scholars argued a correlation between education and obesity but not a clear causal relationship. Guo et al. (2019) used data from the China General Social Survey and found that education has no effect on health outcomes, such as obesity and being overweight.

The literature on the influencing factors and mechanisms of obesity has been relatively abundant. However, most of these studies were conducted in developed countries, and studies conducted in developing countries are relatively few. Given the differences between China and developed Western countries in terms of lifestyle, dietary habits and cultural concepts, using Chinese data to explore the relationship between education and obesity and its influencing mechanism is of international comparative value. Secondly, amongst the socioeconomic factors affecting obesity, the influence of education on obesity has always been a concern of the academic community. However, conclusions of the relevant literature on the influence of education on obesity are quite different. Reasons may be due to the sample selection bias arising from different data or the different measurement indicators of education and obesity. Thus, the heterogeneity of the relationship between education and obesity should be fully considered, and mechanism effects of variables, such as nutritional knowledge and diet regularity, between education and obesity should be further analysed to ensure the reliability of study conclusions.

This paper attempts to study the effect of education on obesity level based on a large survey data at national level in China. The results show that education has no remarkable effect on being overweight but has a remarkably negative effect on obesity. The analysis of the mechanism effects shows that education can have an effect on obesity by improving nutritional knowledge, enhancing health risk perception, improving dietary habits and increasing physical exercise. Further analysis reveals that the health effects of education and the mechanism effects of the above four factors are heterogeneous between urban and rural areas and between genders.

The possible marginal contributions of this paper are follows. First, there is an ongoing debate about the causal relationship between education and obesity and how education can lead to weight loss, the rich data in this study allow us to explore the impact of education on obesity and the underlying mechanisms. Secondly, we combined the heterogeneity analysis with the mechanism analysis to compare the magnitude of the various mechanisms of the effect of education on obesity at work in different samples. Thirdly, in the background of the prominent obesity problem, the findings of this paper provide insights for government to conduct health education-based obesity interventions.

The remainder of this paper is organised as follows. Section 2 presents the research hypotheses through theoretical analysis. Section 3 summarises the methods, including data sources, variable settings and econometric models. Section 4 presents the empirical analysis. Section 5 summarises the main findings of this paper and puts forward relevant policy recommendations.

Research on the relationship between education and health has originated from the health capital model theory, which views health as a long-term capital stock that depreciates as adults age but can be maintained or grown through various investments (Grossman, 1972). The model views health capital as endogenous and other types of human capital, such as education, as exogenous, thus viewing education as an exogenous investment to explain the positive effect of education on health capital. The health capital model theory assumes that holding other factors constant, an educated person efficiently produces health capital and has increased health capital. Return on the education theory suggests that educated people tend to have high economic incomes, giving them improved health resources and health entitlements, relatively minimally labour-intensive jobs, reduced health attrition and increased opportunities for health (Kemptner et al., 2011).

Bourdieu (1986) sees educational attainment as one of the important forms of cultural capital - institutionalised cultural capital. The Bourdieu’s theory of social practice can be understood as the practice of actors in different positions guided by their respective habitus and relying on the capital they possess (Bourdieu, 1986). According to the social practice theory, educated people have increased cultural capital, increased advantages under the guidance of the given field and habitus and increased practices developed in the direction of their health interests. In addition, cultural capital helps to improve or transform into economic and social capitals. The increase in economic capital helps to obtain improved healthcare services to cope with health shocks. The increase in social capital can increase the source of information channels, helping in obtaining health information, sharing and avoiding health risks and improving the health level (Petrou and Kupek, 2008).

Numerous studies showed that obesity is associated with a variety of chronic noncommunicable diseases and increased health risks, and obesity has been widely used as an important measure of health outcomes in real-world research (Casey et al., 2008). Based on the above theoretical analysis, this paper argues that education can have a negative effect on obesity. Through the above analysis, this paper proposes hypothesis H1:

H1: Education has a suppressive effect on obesity.

Regarding the mechanism of education on obesity, according to the health capital model theory, the inhibitory effect of education on obesity is achieved by increasing the productivity of health capital and reducing the production cost of health capital (Grossman, 1972). In this paper, education is suggested to increase the productivity of health capital from individual cognitive and behavioural factors and has a suppressive effect on obesity. Individual cognitive factors include nutritional knowledge and health risk perception, and individual behavioural factors include diet regularity and physical exercise.

Firstly, education can improve the individuals’ nutritional knowledge and health risk perceptions, which can help to improve their ability to eat healthily and be health conscious. This phenomenon reduces their risk of obesity. Ramirez-Rivera et al. (2021) found that nutrition education programs can reduce students’ BMI. Kaufman-Shriqui et al. (2016) found that a nutrition curriculum intervention can improve people’s nutrition knowledge and change their eating habits, thus increasing food variety and consumption of vegetables and fruits and reducing the consumption of sugary drinks. The above evidence suggests that increases in individual nutritional knowledge may have an impact on obesity through changes in dietary behaviour. In addition, the health belief model theory suggests that people’s risk (perceived susceptibility) and disease severity (perceived vulnerability) perceptions play an important role in predicting a range of health behaviours. People with high perceived health risk are likely to be aware of the severity of the obesity problem and the health risks associated with obesity, maintain a healthy weight and thus reducing the risk of obesity (Champion, 1984).

Secondly, education can also promote a change in people’s behaviour and help to develop good habits, and regular diet cycle and sufficient physical exercise can effectively reduce the risk of obesity (Kemptner et al., 2011). Extremely little sleep and irregular diet can easily lead to obesity (Reyna-Vargas et al., 2022). Adequate sleep and regular eating help in reducing the risk of obesity. In addition, long-term sedentary behaviour and screen time exposure are prone to obesity (Nieto and Suhrcke, 2021). Physical exercise plays an important role in maintaining a healthy weight as a major source of energy expenditure, effectively reducing sedentary behaviour and screen exposure time, and promotes calorie burning in the body, thereby reducing the risk of obesity (Lavie et al., 2019). In summary, this paper proposes hypothesis H2:

H2: Education can have a suppressive effect on obesity through nutritional knowledge, health risk perception, diet regularity and physical exercise.

Obesity amongst Chinese residents shows significant urban–rural differences, with obesity rates rising much faster in rural areas than in urban areas (Bureau of Disease Prevention and Control, National Health Commission of China, 2021). Previous studies examined the heterogeneity of returns to education from different perspectives and found evident urban–rural differences in terms of income level and occupational category (Mohapatra and Luckert, 2014; Zhang and Wu, 2017). Given a clear urban–rural dichotomy in most areas of China, the income and education levels of rural residents are lower than those of urban residents. This difference may have different degrees of influence on residents’ nutritional knowledge, health risk perception, diet regularity and physical exercise. Therefore, urban–rural differences may be present in the influence and mechanism of effect of education on obesity. Based on the above analysis, this paper further proposes hypothesis H3:

H3: The effect of education on obesity and the mechanism of action differs between urban and rural populations.

Obesity in China also shows significant gender differences, with obesity rates and obesity growth rates being higher amongst men than amongst women (Bureau of Disease Prevention and Control, National Health Commission of China, 2021). Although the traditional ideology of “preference for sons over daughters” prevails in China, the return to education is higher for women than for men (Hannum et al., 2013). Empirical studies showed that health awareness and frequency of healthy behaviours of women are usually higher than those of men, men tend to have a higher risk of obesity than women due to more daily work socialising, relatively higher work pressure and work intensity and relatively more unhealthy behaviours, such as smoking, drinking alcohol and staying up late (Anson and Sun, 2002). Therefore, the effect and mechanism of action of education on obesity may be equally specific to gender. Based on the above analysis, this paper further proposes hypothesis H4:

H4: The effect of education on obesity and the mechanism of action differs between men and women.

Combining the above analysis and assumptions, this paper proposes the theoretical logic diagram of education influencing obesity shown in Fig. 1.

3.1Data

This paper uses data from the Food Safety and Nutrition Health Survey (FNS), a nationwide survey conducted in June 2022 by the Institute of Food Safety Risk Management, Jiangnan University, in collaboration with several organisations. The survey is conducted online with a total sample size of over 100,000 respondents. The sample distribution is reported in Table A1. Compared with the offline face-to-face survey, the online survey can avoid the measurement error caused by social expectation bias as much as possible, and the accuracy of the online survey is better than the offline survey when respondents answer sensitive questions (Kreuter et al., 2008). The FNS questionnaire contains three main sections, including individual profile, food safety status and nutritional health. The variables involved in this paper are from the basic individual profile and nutritional health sections. Given that this paper focuses on obesity rather than malnutrition, the valid sample in this paper only includes those with BMI not less than 18.5. A total of 81,123 valid samples are obtained after data cleaning.

3.2 Variables

3.2.1 Dependent variable

The dependent variable in this paper is obesity. The BMI of most individuals is significantly and positively correlated with the percentage content of body fat, which can reflect the degree of obesity of the organism. Regarding the classification of obesity degree, the BMI threshold value set by the World Health Organization (WHO) is usually used internationally. Overweight is defined as BMI ≥ 25, and obesity is defined as BMI ≥ 30. The Working Group on Obesity in China (WGOC) of the China Office of the International Society of Life Sciences summarises and analyses the relationship between BMI and the prevalence of related diseases based on large-scale measurement data in China. The proposed cut-off values for judging the degree of overweight and obesity in Chinese adults are as follows: overweight, BMI ≥ 24 and obesity, BMI ≥ 28. In this paper, the classification of obesity level follows the criteria proposed by WGOC. Figure 2 shows the distribution of sample BMI, which is roughly normally distributed, and samples with BMI > 28, i.e. the obese group amongst the respondents, are on the right side of the reference line in the figure.

3.2.2 Independent variable

The independent variable is education, which is ordinal and measured through an individual’s educational attainment (1, junior high school or below; 2, high school; 3, junior college; 4, undergraduate and 5, graduate and above). Figure 3 demonstrates the obesity rate of the sample group under different education levels. A high education level results in low obesity rate.

3.2.3 Mechanism Variables

Based on the previous study hypotheses, the mechanism variables are nutrition knowledge, perceived health risks, diet regularity and physical exercise. Drawing on the study of Ren et al. (2019), nutrition knowledge is measured by the scores of relevant nutrition questions. Figure 4 shows the density of nutrition knowledge scores, which are roughly normally distributed with a relatively even distribution. The perceived health risk is measured by whether a person reads the nutrition label in daily shopping at supermarkets. The information on food ingredients conveyed by the nutrition label reflects the nutritional and health values of food to a certain extent. Thus, the frequency of reading the nutrition label in daily shopping can reflect the perceived risk intensity of food health attributes (Yang et al., 2021). The frequency of late-night snacking is chosen as a proxy variable for diet regularity because people who do late-night snacking usually stay up late and are often accompanied by unhealthy behaviours, such as waking up late in the morning and skipping breakfast. This variable can also reflect to some extent the degree of deviation from the normal state of three meal cycles per day. Physical exercise is measured by the specific time of exercise per day, and the definitions and descriptive statistics of all variables are shown in Table 1.

3.2.4 Control variables

Individual characteristics, such as gender, age, marriage and occupation, have different degrees of influence on obesity. Ren et al. (2019) found an inverted U-shaped relationship between age and obesity level, i.e. BMI increases with age and decreases with age when age passes a certain cut-off value. Kong and Qi (2017) found gender and urban–rural differences in the effects of socioeconomic factors on obesity in the population. In addition, married people and people with long working hours have a high chance of suffering from obesity (Bell and Jr. Thorpe, 2019; Virtanen et al., 2020). Therefore, in this paper, gender, age, marriage, family size, residence, occupation (e.g. civil servant, corporate employee and farmer) and income are used as control variables. Considering the close relationship between obesity and chronic diseases especially that obesity in certain populations is directly related to chronic diseases and their therapeutic drugs, we further control whether respondents suffer from chronic diseases. Table 1 reports all variable definitions and descriptive statistics.

Table 1

Variable name and definition
Variables	Definition	Mean	SD	Min	Max
Dependent variable
Obesity	BMI ≥ 28 = 1, Other = 0	0.070	0.250	0	1
Overweight	BMI ≥ 24 = 1, Other = 0	0.333	0.471	0	1
Independent variable
Education	Junior high school or below = 1 High school = 2 Junior college = 3 Undergraduate = 4 Graduate and above = 5	2.970	1.210	1	5
Control variables
Gender	Male = 1, Female = 0	0.500	0.500	0	1
Age	18–25 years old = 1 26–45 years old = 2 46–60 years old = 3 61 years old and above = 4	1.890	0.790	1	4
Marriage	Married = 1, Unmarried = 0	0.580	0.490	0	1
Rural or Urban	Urban = 1, Rural = 0	0.700	0.460	0	1
Career
Career1	Civil servant = 1, Other = 0	0.100	0.300	0	1
Career2	Corporate employee = 1, Other = 0	0.470	0.500	0	1
Career3	Farmer = 1, Other = 0	0.070	0.260	0	1
Personal Annual income	30,000 yuan and below = 1 30,000–50,000 yuan = 2 50,000–100,000 yuan = 3 100,000–150,000 yuan = 4 150,000–200,000 yuan = 5 200,000–500,000 yuan = 6 over 500,000 yuan = 7	2.610	1.640	1	7
Household size	Number of people living together	3.530	1.050	1	5
Chronic diseases	With chronic disease = 1, Not suffering = 0	0.360	0.480	0	1
Mechanism Variables
Nutrition knowledge	Nutrition knowledge score	4.575	1.485	0	7
Health risk perception	Whether the person doing the daily purchase of food in the supermarket in the mall reads the nutrition label: Does not read at all = 1 Basically does not read = 2 Occasionally reads = 3 Often reads = 4	2.799	0.961	1	4
Diet regularity	Whether the person has the habit of eating late-night snacks: Eats almost every day = 1 Often eats = 2 Rarely eats = 3 Never eats = 4	3.079	0.691	1	4
Physical exercise	Average daily active physical exercise (meaning exercise other than daily work and life): Never actively exercises = 1 Exercises for less than 30 minutes = 2 Exercises for 30–60 minutes = 3 Exercises for more than 60 minutes = 4	2.222	0.867	1	4

3.3 Econometric models

3.3.1 Baseline regression model

Given that the dependent variable obesity is a binary variable, we use Probit model to estimate the baseline regression with the following model settings:

$${obesity}_{i}=\alpha +\beta {edu}_{i}+\gamma {X}_{i}+{\epsilon }_{i}$$

The dependent variable ${obesity}_{i}$ in Eq. (1) indicates whether the sample is obese, with obesity assigned a value of 1, and others assigned a value of 0. The independent variable ${edu}_{i}$ denotes the education level of the sample, ${X}_{i}$ represents the other control variables of sample, and ${\epsilon }_{i}$ is the random error term.

3.3.2 Instrumental variable model

This paper further uses the IV-Probit model to correct the estimation of the baseline regression model and address the endogeneity problem caused by reverse causality or omit variables in the baseline regression model. Drawing on the study of Ren et al. (2019) and Ma et al. (2019), the instrumental variable is chosen as the average years of schooling of the population aged 15 years and above in the area where the sample is located. Descriptive statistics of instrumental variable is reported in Table A2. Theoretically, the average number of years of education in an individual’s region can reflect the importance of education in the local area and correlates with the education level of an individual. However, this variable is not related to whether an individual is obese, which satisfies the conditions for the use of instrumental variables. The instrumental variable model is set as follows:

${edu}_{i}={\alpha }_{1}+{\beta }_{1}{instrument}_{i}+{\gamma }_{1}{X}_{i}+{\epsilon }_{i}$ , and(2)

$${obesity}_{i}={\alpha }_{2}+{\beta }_{2}{\widehat{edu}}_{i}+{\gamma }_{2}{X}_{i}+{\epsilon }_{i}$$

Equations (2) and (3) are the 1st and 2nd-stage regression results, respectively. The $instrument$ in Eq. (2) is the instrumental variable, and $\widehat{edu}$ in Eq. (3) is the first-stage regression fitting result of Eq. (2).

3.3.3 Quantile regression model

The quantile regression model can measure the effect of independent variables on the dependent variables at different quartiles, thus providing a comprehensive description of the effect of education on obesity. Also, the quantile regression has relaxed assumptions, is not affected by non-normal distributions and heteroskedasticity and is minimally susceptible to extreme values when data have a distribution with spikes or thick tails, thus yielding robust results.

The overall q-quantile of the conditional distribution $y|x$, ${y}_{q}\left(x\right)$ is assumed to be a linear function of $x$, i.e. ${y}_{q}\left({x}_{i}\right)={x}_{i}^{{\prime }}{\beta }_{q}$, ${\beta }_{q}$ is then called the q-quantile regression coefficient. On the basis of the above analysis, the quantile regression model is designed in this paper as follows:

$${Q}_{q}={\alpha }_{1q}+{{\beta }_{1q}edu}_{i}+{\delta }_{1q}{X}_{i}+{\epsilon }_{i}$$

The independent variable ${edu}_{i}$ in Eq. (4) expresses the same meaning as the above. ${\beta }_{1q}$ denotes the q-quantile regression coefficient of the dependent variable ${edu}_{i}$, ${X}_{i}$ is the other control variables, and ${\epsilon }_{i}$ is the error term. The dependent variable of the quantile regression model here is the numerical continuous variable BMI, and the quantile regression model is applied at the robustness test of this paper to verify the reliability of the baseline regression results.

3.3.4. Mechanistic effect model

This paper uses the coefficient difference method and the KHB decomposition method for mechanism effects analysis, and the mechanism effects model for the coefficient difference method is designed as follows:

${channel}_{i}=\alpha +{\beta }_{3}{edu}_{i}+{\gamma }_{3}{X}_{i}+{\epsilon }_{i}$ , and(5)

$${obesity}_{i}=\alpha +{\beta }_{4}{edu}_{i}+\theta {channel}_{i}+{\gamma }_{4}{X}_{i}+{\epsilon }_{i}$$

The $channel$ in Equations (5) and (6) denotes the mechanism variable, and the other variables are named and defined similarly as before. The process of testing the mechanism effect by the coefficient difference method is to first test the positivity and significance of the coefficient ${\beta }_{3}$ in Eq. (5). If the coefficient is significant, the regression model test on Eq. (6) is continued, and if the coefficient $\theta$ is significant, the existence of a mechanism effect of the variable channel is indicated. The coefficient difference and significance between the coefficients ${\beta }_{4}$ and $\beta$ in Eq. (1) are further compared to determine whether the partial or full mechanism effect of the mechanism variable $channel$ is present. The coefficient difference method is applied to the case where the dependent variable is a continuous variable, and the dependent variable obesity is a binary variable in this paper. Mechanism effects are also determined by the KHB method, which is applicable when explanatory variables are binary and can extend the decomposability of the linear model to the nonlinear model, to avoid testing errors (Kohler et al., 2011).

5.1 Baseline regression model

Table 2 reports the results of the baseline regression and the instrumental variable models, with the dependent variable for the regression results in columns (1), (2) and (3) being overweight (BMI ≥ 24). Results show that the regression coefficient of educational attainment on being overweight is significantly negative in the regression model without the inclusion of instrumental variables and that the coefficient changes with the inclusion of instrumental variables is not statistically significant. The dependent variables for the regression results in columns (4), (5) and (6) are obesity (BMI ≥ 28), and the regression coefficients of educational attainment on obesity are all significantly negative at the 1% level. The above results indicate that educational attainment has a significant inhibitory effect on obesity, and baseline regression results tentatively verify hypothesis H1. The regression coefficients of instrumental variables in Table 2 are significantly positive at the 1% level, which proves the relevance of instrumental variables at the data level. The first-stage F value is greater than 10, and no weak instrumental variable problem is observed.

Notably, the effect of education on overweight is not significant after the inclusion of instrumental variables in the model. Possible explanations are as follows. Firstly, subjectively, people may have low perceptions of the health risks of being overweight and do not believe that being overweight brings high health risks and serious diseases. Secondly, objectively, the health risks associated with being overweight are lower compared with those with obesity, and people with high education levels may not adopt effective suppressive behaviours against being overweight.

Table 2

Baseline regression
Variables	Dependent variable: Overweight			Dependent variable: Obesity
	Probit	Probit	IV-Probit	Probit	Probit	IV-Probit
	(1)	(2)	(3)	(4)	(5)	(6)
Second stage
Education	−0.0758***	−0.0515***	−0.0779	−0.0817***	−0.0752***	−0.252***
	(0.0037)	(0.0043)	(0.0538)	(0.00563)	(0.0064)	(0.0799)
Constant	−0.208***	−1.051***	−0.973***	−1.254***	−1.517***	−1.002***
	(0.0120)	(0.0359)	(0.158)	(0.0173)	(0.0422)	(0.235)
Other control variables	N	Y	Y	N	Y	Y
First stage
Instrument variables			0.1408*** (0.006)			0.1408*** (0.006)
F value			1868.65			1868.65
Observations	81123	81123	81123	81123	81,123	81123
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

5.2 Robustness tests

5.2.1 Substitution of dependent variable

Firstly, this paper replaces the dependent variable with BMI, and the robustness of the baseline regression results is tested by OLS regression. The column (1) of Table 3 reports the OLS regression results, and the coefficient of educational attainment on BMI is significantly negative at the 1% level, which initially verifies the reliability of baseline regression results. Furthermore, given that BMI is a numerical continuous variable, this paper analyses the effect of education level on BMI at different quartiles by quantile regression. Columns (2) to (6) in Table 3 report the regression results of BMI at different quartiles. Results show that education has a significant inhibitory effect on BMI at different quartiles. Figure 5 shows the plots of the quantile regression coefficients at all quartiles and finds that the inhibitory effect of education on BMI increases with increasing BMI quantile.

Table 3

Robustness tests (1)
Variables	Dependent variable: BMI
	OLS	Q_10	Q_30	Q_50	Q_70	Q_90
	(1)	(2)	(3)	(4)	(5)	(6)
Education	−0.136***	−0.0107**	−0.0644***	−0.101***	−0.174***	−0.263***
	(0.0100)	(0.0048)	(0.0085)	(0.0100)	(0.0127)	(0.0212)
Constant	18.55***	18.48***	19.56***	20.82***	22.58***	26.05***
	(0.0469)	(0.0334)	(0.0584)	(0.0657)	(0.0854)	(0.139)
Other control variables	Y	Y	Y	Y	Y	Y
Observations	81123	81123	81123	81123	81123	81123
R-squared	0.111	--	--	--	--	--
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

5.2.2 Replacement regression model

The Probit regression model is chosen in the baseline regression model and is replaced here with a Logit model to test the previous regression results. Columns (1) and (2) in Table 4 report the regression results after replacing the regression model. Specifically, no control variable is added in column (1), and control variables are added in column (2). Results show that the regression results of education level on obesity are significantly negative at the 1% level.

5.2.3 Truncation processing

Given the large sample size, this paper further uses truncation processing to test the reliability of the baseline regression results and avoid the influence of outliers on regression results. The processing steps of the truncation processing are as follows. Firstly, the continuous variable BMI is truncated, and results after truncation are divided into dummy variables of obesity (BMI ≥ 28 = 1, others = 0). Finally, obesity is brought into the baseline model as the dependent variable for regression. Column (3) of Table 4 reports the regression results after truncating 5%, and all results show that the education level still has a significantly negative effect on the dependent variable after truncating the regression.

5.2.4 Correction of selection bias

This paper uses the PSM and IPWRA methods to test the robustness of the regression results and avoid sample selection bias. PSM can largely reduce the bias of the observed data, but its use usually requires two conditions to be satisfied. Firstly, PSM usually requires a relatively large sample size to obtain a high-quality match. Secondly, PSM requires a large common range of values for the propensity scores of the treatment and control groups. Otherwise, a large number of observations will be lost, resulting in a nonrepresentative sample for the rest. In this paper, the sample size is large, and Fig. 6 shows the common range of propensity scores. Most of the observations are within the common range. Only a small number of samples will be lost when PSM is performed, and both of these conditions can be satisfied. PSM uses one-to-one matching, and columns (4) and (5) in Table 4 report the estimation results based on PSM and IPWRA. The average treatment effect on the treated (ATT) are significantly negative at the 1% level, further validating the significantly negative effect of educational attainment on obesity. The results of the PSM balance test are reported in Table A3.

Table 4

Robustness tests (2)
Variables	Replacement regression model		Truncation processing	PSM (ATT)	IPWRA (ATT)
Variables	(1)	(2)	(3)	(4)	(5)
Education	−0.169***	−0.152***	−0.0759***	−0.0093***	−0.0151***
	(0.0117)	(0.0132)	(0.0104)	(0.0021)	(0.0019)
Constant	−2.129***	−2.734***	−2.052***	--	--
	(0.0352)	(0.0876)	(0.0669)	--	--
Other control variables	N	Y	Y	Y	Y
Observations	81123	81123	73049	81123	81123
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively. (3) The dependent variables for the regressions in columns 1 and 2 are BMI, and the dependent variables for the regressions in columns 3–5 are obesity.

5.3 Analysis of mechanism effects

Next, the mechanism effect analysis for different types of mechanism variables is conducted to investigate the pathway of education on obesity. The magnitude of mechanism effects generated by different types of mechanism variables is compared. In this paper, the coefficient difference method and KHB decomposition method are used to analyse the mechanism effects of nutritional knowledge, health risk perception, diet regularity and physical exercise. The regression results of the coefficient difference method are reported in Tables 5 and 6. Table 5 shows the regression results of Eq. (5), and Table 6 shows the regression results of Eq. (6). Results show that the regression coefficients of education on mechanism variables are significantly positive at the 1% level. The regression coefficients of education and mechanism variables on obesity are also significantly negative at the 1% level when the mechanism variables are used as control variables. The above results suggest that education can inhibit obesity by improving nutrition knowledge, enhancing health risk perception, improving diet regularity and enhancing physical exercise, thus verifying hypothesis H2.

Table 5

Coefficient difference method (1)
Variables	Dependent variable
	Nutrition knowledge	Health risk perception	Diet regularity	Physical exercise
	(1)	(2)	(3)	(4)
Education	0.1784***	0.0521***	0.0234***	0.0515***
	(0.0049)	(0.0035)	(0.0037)	(0.0035)
Other control variables	Y	Y	Y	Y
Observations	81123	81123	81123	81123
R-squared	0.0477	--	--	--
Wald chi2	--	1067.38	3968.06	1913.37
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

Table 6

Coefficient difference method (2)
Variables	Dependent variable: Obesity
Variables	(1)	(2)	(3)	(4)
Education	−0.0697***	−0.0734***	−0.0743***	−0.0714***
	(0.0065)	(0.0064)	(0.0064)	(0.0064)
Nutrition knowledge	−0.0333*** (0.0045)	--	--	--
		--	--	--
Health risk perception	--	−0.0403*** (0.0073)	--	--
	--		--	--
Diet regularity	--	--	−0.0617***	--
	--	--	(0.0103)	--
Physical exercise	--	--	--	−0.0908***
	--	--	--	(0.0082)
Other control variables	Y	Y	Y	Y
Observations	81123	81123	81123	81123
Wald chi-squared	1986.99	1926.80	1923.78	1987.43
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

Furthermore, this paper performs the mechanism effect test by the KHB decomposition method. Table 7 reports the decomposition results of the KHB decomposition method. Columns (1), (2), (3) and (4) report the mechanism effects of nutritional knowledge, perceived health risks, diet regularity and physical exercise, respectively, and the percentage of mechanism effects. Results show that the mechanism effects of all mechanism variables are significant at the 1% level, further validating hypothesis H2.

Further analysis reveals that the mechanism effects of nutritional knowledge, perceived health risks, diet regularity and physical exercise account for 12.9% of the total effect of education on obesity. Mechanism effects of nutritional knowledge, perceived health risks, diet regularity and physical exercise account for 6.57%, 0.94%, 1.03% and 4.36% of the total effects, 50.93%, 7.25%, 7.97% and 33.84% of the total mechanism effects. The results of the above analysis show that nutrition knowledge and physical exercise play the largest mechanism effect, reflecting the direct effect of diet and exercise on obesity.

Table 7

KHB decomposition method
Variables	Mechanism variables
	Nutrition knowledge	Health risk perception	Diet regularity	Physical exercise
	(1)	(2)	(3)	(4)
Total effect	−0.075***	−0.075***	−0.075***	−0.075***
	(0.006)	(0.006)	(0.006)	(0.006)
Direct effect	−0.069***	−0.073***	−0.074***	−0.071***
	(0.006)	(0.006)	(0.006)	(0.006)
Mechanism effects	−0.005*** (0.000)	−0.001*** (0.000)	−0.001*** (0.000)	−0.003*** (0.000)
Mechanism effect / Total effect	6.57%	0.94%	1.03%	4.36%
Mechanism effect / Total Mechanism effect	50.93%	7.25%	7.97%	33.84%
Other control variables	Y	Y	Y	Y
Observations	81123	81123	81123	81123
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

5.4 Heterogeneity analysis

5.4.1. Heterogeneity analysis: effect of education on obesity

A large amount of real-world evidence and previous studies showed significant urban–rural and gender differences in obesity in the Chinese population (Anson and Sun, 2002; Mohapatra and Luckert, 2014; Zhang and Wu, 2017). In this section, heterogeneity analysis based on urban–rural and gender perspectives is conducted to further investigate the urban–rural and gender differences in the effect of education on obesity and mechanism of action. The heterogeneity analysis is performed by group regression and between-group difference test. The results of the heterogeneity analysis are reported in Table 8. The results of the heterogeneity analysis based on urban–rural differences show that the coefficients of educational attainment are significantly negative at the 1% level for urban and rural residents. The empirical p-value of the between-group difference test is significant at the 10% level, indicating urban–rural differences in the effect of educational attainment on obesity. Comparing the regression coefficients can show that the effect of educational attainment on obesity is more significant for rural residents than for urban residents. The results of the heterogeneity analysis based on gender differences also show that the regression coefficient of educational attainment is significantly positive at the 1% level for women and men. The empirical p-value of the between-group difference test is significant at the 10% level. The comparison of the regression coefficients reveals that the effect of educational attainment on obesity is more significant for women than for men.

Table 8

Heterogeneity analysis (1): effect of education on obesity
Variables	Urban and rural groups		Gender groups
	Urban	Rural	Female	Male
	(1)	(2)	(3)	(4)
Education	−0.065***	−0.091***	−0.092***	−0.068***
	(0.007)	(0.011)	(0.010)	(0.008)
Constant	−1.672***	−1.496***	−1.404***	−1.354***
	(0.048)	(0.069)	(0.063)	(0.052)
Other control variables	Y	Y	Y	Y
LR chi2	1232.33***	772.86***	589.49***	1470.58***
Pseudo R²	0.046	0.056	0.035	0.063
Observations	56566	24557	40545	40578
T-test
p-value	0.078*		0.081*
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

5.4.2 Heterogeneity analysis: mechanisms of the effect of education on obesity

This paper further analyses whether the mechanism of the effect of education on obesity is equally specific to urban–rural and gender differences. The mechanism effect analysis uses the KHB decomposition method. Table 9 shows the results of the KHB decomposition based on urban–rural and gender differences.

The results of the mechanism effect test based on urban–rural differences show that the mechanism effects of four mechanism variables account for 16.53% of the total effect for urban resident samples. Specifically, nutritional knowledge accounts for the highest mechanism effect, explaining 7.06% of the mechanism effect followed by physical exercise, explaining 5.63% of the mechanism effect. For the sample of rural residents, the mechanism effect of mechanism variables only accounts for 3.43% of the total effect. Physical exercise accounts for the highest proportion of the mechanism effect at 2.65% followed by nutritional knowledge at 1.03%. Health risk perception and diet regularity do not pass the mechanism effect test, and no mechanism effect is observed. The above results reveal that nutrition knowledge and physical exercise play a certain proportion of the mechanism effect in urban samples. For rural samples, the mechanism effect of the four mechanism variables is relatively low. The effect mechanism of education on obesity in rural samples may be complex or easy to observe or may be caused by the low level of education of rural residents, and the role of education is not easy to distinguish.

The results of this paper also suggest that the effect of education on obesity may be complex or not observable in rural samples and that the mechanism effect of nutritional knowledge is largest in the urban sample. The mechanism effect of physical exercise is largest in the rural sample, indicating an urban–rural difference in the mechanism of the effect of education on obesity. Hypothesis H3 is verified by the above analysis.

The results of the mechanism effect test based on gender differences show that the mechanism effects of four mechanism variables account for 10.17% and 10.41% of the total effect in females and males, respectively, and no significant difference between the genders is observed. Further comparison of the mechanism effect sizes accounted by the four mechanism variables reveals that for the female group, nutritional knowledge exerts the largest mechanism effect (4.46%) followed by physical exercise (2.96%). For the male group, physical exercise plays the largest mechanism effect (5.63%) followed by nutritional knowledge (3.60%). The above results suggest a degree of heterogeneity in the mechanism of the effect of education on obesity. For females, education may have an effect on obesity through improving nutritional knowledge, whereas for males, education may have an inhibitory effect on obesity through enhancing physical exercise. Combined with the above analysis, hypothesis H4 is verified.

Table 9

Heterogeneity analysis (2): Mechanistic effects of education on obesity (KHB)
Variables	Nutrition knowledge	Health risk perception	Diet regularity	Physical exercise	Mechanism effect
Variables	(1)	(2)	(3)	(4)	Mechanism effect
Urban
Mechanism effect	−0.004*** (0.001)	−0.001*** (0.000)	−0.001*** (0.000)	−0.003*** (0.001)
Mechanism effect / Total effect	7.06%	2.09%	1.75%	5.63%	16.53%
Rural
Mechanism effect	−0.0001 (0.0014)	0.0004 (0.0006)	−0.0001 (0.0002)	−0.0023*** (0.0006)
Mechanism effect / Total effect	1.03%	-0.47%	0.22%	2.65%	3.43%
Female
Mechanism effect	−0.004*** (0.001)	−0.0007 (0.0007)	−0.0017*** (0.0004)	−0.0026*** (0.0005)
Mechanism effect / Total effect	4.46%	0.86%	1.90%	2.96%	10.17%
Male
Mechanism effect	−0.0024*** (0.0009)	−0.0005* (0.0003)	−0.0002 (0.0001)	−0.0038*** (0.0006)
Mechanism effect / Total effect	3.60%	0.77%	0.41%	5.63%	10.41%
Control variables	Y	Y	Y	Y
Note: (1) Robust standard errors are in parentheses. (2) , * and *** indicate passing the test at 10%, 5% and 1% significance levels, respectively.

This paper analyses the effect of education on obesity and the mechanism of action based on FNS data. Several main conclusions and implications can be drawn from this research. Firstly, education can have a significantly negative effect on obesity but not on being overweight. After the above results are tested for robustness by replacing the dependent variables and the regression model, truncation processing, PSM and IPWRA methods, regression results are still reliable. Secondly, education can exert inhibitory effects on obesity by improving nutrition knowledge, enhancing health risk perception, improving diet regularity and enhancing physical exercise. Mechanism effects due to nutrition knowledge and physical exercise are high. Third, the heterogeneity analysis has found that the effect of education on obesity is more significant in rural residents and female samples compared with their counterparts. The heterogeneity analysis of the mechanism of action found that for urban residents, education reduces obesity risk by improving nutrition knowledge, whereas for rural residents, education inhibits obesity by enhancing physical exercise. In terms of gender differences, the inhibitory effect of education on obesity in males is achieved by enhancing physical exercise, whereas that on obesity in females is achieved by improving nutritional knowledge.

Based on the above research findings, the following policy insights can be obtained. Firstly, with China’s economic development, the rising education level of the general public can be expected to have a positive impact on the prevention and control of obesity. However, we should continue to enhance school health education and incorporate health concepts, health literacy and health behaviours into the national education system based on quality education. We should build a school health education promotion mechanism and actively promote nutrition knowledge and the dangers of obesity to raise students’ awareness of obesity prevention. Secondly, obesity intervention should be tailored to the individual, and effective obesity intervention measures should be carried out for people with different characteristics to improve the effectiveness and reduce the cost of intervention. For urban residents or female population, the main focus should be on health knowledge and nutrition knowledge promotion. For rural residents or male population, the main focus of obesity intervention should be on promoting physical exercise and other outdoor sports. At the same time, the obesity intervention of key groups should be further strengthened, and the obesity intervention of rural residents should be strengthened under the realistic background of faster growth of obesity rate in rural areas. In addition, given that women’s health awareness is relatively high, the role of education is more evident to the female group than to the male group. Health interventions as a family unit and playing to the health advantages of the female group are possible to try to influence other people in the family by means of the family network effect. Third, the early intervention of obesity should be paid attention, and the residents’ awareness of overweight prevention should be raised to avoid the development of obesity in the future.

Ethical Approval and Consent to participate

Approval for this study was provided by the ethics review board of Anhui Agricultural University.

Consent for publication

Not applicable.

Availability of data and materials

The datasets generated or analysed during this study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported by National Social Science Foundation of China (21AGL022). We also thank the Institute of Food Safety Risk Management of Jiangnan University for providing the data.

Authors’ contributions

Shijiu Yin: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Writing – original draft, Writing – review & editing; ZongshuoYin: Conceptualization, Investigation, Methodology, Writing – original draft, Writing – review & editing; Mo Chen: Data curation, Formal analysis, Software; Nianyu Du: Writing – original draft, Writing – review & editing; Kai Li: Writing – original draft, Writing – review & editing.

Acknowledgements

We thank the Institute of Food Safety Risk Management of Jiangnan University for providing the data.

Aiyar, A., Rahman, A., Pingali, P., 2021. India's rural transformation and rising obesity burden. World Dev. 138. https://doi.org/10.1016/j.worlddev.2020.105258.
Anson, O., Sun, S., 2002. Gender and health in rural China: evidence from Hebei province. Soc. Sci. Med. 55 (6), 1039-1054. https://doi.org/10.1016/S0277-9536(01)00227-1.
Ayton, A., Ibrahim, A., 2019. Obesity is a public health emergency. BMJ Br. Med. J. 366. https://doi.org/10.1136/bmj.l5463.
Baldacci, E., Clements, B., Gupta, S., Cui, Q., 2008. Social spending, human capital, and growth in developing countries. World Dev. 36 (8), 1317-1341. https://doi.org/10.1016/j.worlddev.2007.08.003.
Bell, C.N., Jr. Thorpe, R.J., 2019. Income and marital status interact on obesity among black and white men. Am. J. Mens Health 13 (1). https://doi.org/10.1177/1557988319829952.
Bourdieu P., 1986. The Forms of Capital, in Richardson, J. (ed.): Handbook of Theory and Research for the Sociology of Education. Greenwood Press. https://www.jstor.org/stable/41162948.
Casey, A.A., Elliott, M., Glanz, K., Haire-Joshu, D., Lovegreen, S.L., Saelens, B.E., et al., 2008. Impact of the food environment and physical activity environment on behaviors and weight status in rural us communities. Prev. Med. 47 (6), 600-604. https://doi.org/10.1016/j.ypmed.2008.10.001.
Champion, V.L., 1984. Instrument development for health belief model constructs. ANS. Advances in nursing science 6 (3), 73-85. https://doi.org/10.1097/00012272-198404000-00011.
Christakis, N.A., Fowler, J.H., 2007. The spread of obesity in a large social network over 32 years. N. Engl. J. Med. 357 (4), 370-379. https://doi.org/10.1056/NEJMsa066082.
Cockerham, W.C., 2022. Theoretical approaches to research on the social determinants of obesity. Am. J. Prev. Med. 63 (1), S8-S17. https://doi.org/10.1016/j.amepre.2022.01.030.
de la Haye, K., Robins, G., Mohr, P., Wilson, C., 2011. Homophily and contagion as explanations for weight similarities among adolescent friends. J. Adolesc. Health 49 (4), 421-427. https://doi.org/10.1016/j.jadohealth.2011.02.008.
Deschenes, O., Wang, H., Wang, S., Zhang, P., 2020. The effect of air pollution on body weight and obesity: evidence from China. J. Dev. Econ. 145. https://doi.org/10.1016/j.jdeveco.2020.102461.
Dhrifi, A., Alnahdi, S., Jaziri, R., 2021. The causal links among economic growth, education and health: evidence from developed and developing countries. J. Knowl. Econ. 12 (3), 1477-1493. https://doi.org/10.1007/s13132-020-00678-6.
Ding, D., Gebel, K., 2012. Built environment, physical activity, and obesity: what have we learned from reviewing the literature? Health Place 18 (1), 100-105. https://doi.org/10.1016/j.healthplace.2011.08.021.
Durand, C.P., Andalib, M., Dunton, G.F., Wolch, J., Pentz, M.A., 2011. A systematic review of built environment factors related to physical activity and obesity risk: implications for smart growth urban planning. Obes. Rev. 12 (501), e173-e182. https://doi.org/10.1111/j.1467-789X.2010.00826.x.
Giuntella, O., Rieger, M., Rotunno, L., 2020. Weight gains from trade in foods: evidence from Mexico. J. Int. Econ. 122. https://doi.org/10.1016/j.jinteco.2019.103277.
Grossman, M., 1972. On the concept of health capital and the demand for health. Journal of Political Economy, 80(2), 223–255. https://doi.org/10.7312/gros17812-004.
Guo, S., Zhang, M., Cao, J., 2021. Can education really affect health? Evidence from the implementation of China's 1986 Compulsory Education Law. Eco. J. 6(03), 148-187.http://doi.org/10.16513/j.cnki.cje.20190730.001.
Hannum, E., Zhang, Y., Wang, M., 2013. Why are returns to education higher for women than for men in urban China? China Q. (215), 616-640. https://doi.org/10.1017/S0305741013000696.
James, W., 2018. Obesity: a global public health challenge. Clin. Chem. 64 (1), 24-29. https://doi.org/10.1373/clinchem.2017.273052.
Kaufman-Shriqui, V., Fraser, D., Friger, M., Geva, D., Bilenko, N., Vardi, H., et al., 2016. Effect of a school-based intervention on nutritional knowledge and habits of low-socioeconomic school children in Israel: a cluster-randomized controlled trial. Nutrients 8 (4). https://doi.org/10.3390/nu8040234.
Kemptner, D., Juerges, H., Reinhold, S., 2011. Changes in compulsory schooling and the causal effect of education on health: evidence from Germany. J. Health Econ. 30 (2), 340-354. https://doi.org/10.1016/j.jhealeco.2011.01.004.
Kim, J., 2016. Personality traits and body weight: evidence using sibling comparisons. Soc. Sci. Med. 163, 54-62. https://doi.org/10.1016/j.socscimed.2016.06.054.
Kong, G., Qi, Y., 2017. Socioeconomic factors influencing obesity among residents: gender and urban-rural differences. Sociological Rev. 5 (05), 79-96. http://doi.org/10.3969/j.issn.2095-5154.2017.05.006.
Kohler, U., Karlson, K.B., Holm, A., 2011. Comparing coefficients of nested nonlinear probability models. Stata J. 11 (3), 420-438. https://doi.org/10.1177/1536867X1101100306.
Kreuter, F., Presser, S., Tourangeau, R., 2008. Social desirability bias in cati, ivr, and web surveys the effects of mode and question sensitivity. Public Opin. Q. 72 (5), 847-865. https://doi.org/10.1093/poq/nfn063.
Lavie, C.J., Ozemek, C., Carbone, S., Katzmarzyk, P.T., Blair, S.N., 2019. Sedentary behavior, exercise, and cardiovascular health. Circ. Res. 124 (5), 799-815. https://doi.org/10.1161/CIRCRESAHA.118.312669.
Liu, S., Wu, P., Huang, T., 2018. Nonlinear causality between education and health: the role of human development index. Appl. Res. Qual. Life 13 (3), 761-777. https://doi.org/10.1007/s11482-017-9557-0.
Mary, S., Stoler, A., 2021. Does agricultural trade liberalization increase obesity in developing countries? Rev. Dev. Econ. 25 (3), 1326-1350. https://doi.org/10.1111/rode.12757.
Ma, M., 2019. Does children’s education matter for parents’ health and cognition? Evidence from China. J. Health Econ. 66, 222-240. https://doi.org/10.1016/j.jhealeco.2019.06.004.
Mazariegos, M., Auchincloss, A.H., Braverman-Bronstein, A., Kroker-Lobos, M.F., Ramirez-Zea, M., Hessel, P., et al., 2022. Educational inequalities in obesity: a multilevel analysis of survey data from cities in lain America. Public Health Nutr. 25 (7), 1790-1798. https://doi.org/10.1017/S1368980021002457.
Menghetti, E., Strisciuglio, P., Spagnolo, A., Carletti, M., Paciotti, G., Muzzi, G., et al., 2015. Hypertension and obesity in Italian school children: the role of diet, lifestyle and family history. Nutr. Metab. Carbiovasc. Dis. 25 (6), 602-607. https://doi.org/10.1016/j.ijedudev.2014.01.002.
Mohapatra, S., Luckert, M.K., 2014. Educational returns beyond the mean: differences along wage distributions of men and women in India’s formal labor market. Int. J. Educ. Dev. 36, 22-32. https://doi.org/10.1016/j.ijedudev.2014.01.002.
National Health Commission Bureau of Disease Control and Prevention, 2021. Report on Nutrition and Chronic Diseases of Chinese Residents (2020). People's Medical Publishing House. http://www.gov.cn/xinwen/2020-12/24/content_5572983.htm.
Nieto, A., Suhrcke, M., 2021. The effect of tv viewing on children's obesity risk and mental well-being: evidence from the UK digital switchover. J. Health Econ. 80. https://doi.org/10.1016/j.jhealeco.2021.102543.
Petrou, S., Kupek, E., 2008. Social capital and its relationship with measures of health status: evidence from the health survey for England 2003. Health Econ. 17 (1), 127-143. https://doi.org/10.1002/hec.1242.
Philipson, T.J., Posner, R.A., 2003. The long-run growth in obesity as a function of technological change. Perspect. Biol. Med. 46 (3 Suppl), S87-S107. https://doi.org/10.3386/w7423
Prakash, K., Awaworyi Churchill, S., Smyth, R., 2022. Petrol prices and obesity. Health Econ. 31 (7), 1381-1401. https://doi.org/10.1002/hec.4513.
Ramirez-Rivera, D.L., Martinez-Contreras, T., Villegas-Valle, R.C., Henry-Mejia, G., Quizan-Plata, T., Haby, M.M., et al., 2021. Preliminary results of the planet nutrition program on obesity parameters in mexican schoolchildren: pilot single-school randomized controlled trial. Int. J. Environmental Res. Public Health 18 (2). https://doi.org/10.3390/ijerph18020790.
Ren, Y., Li, H., Wang, X., 2019. Family income and nutrition-related health: evidence from food consumption in China. Soc. Sci. Med. 232, 58-76. https://doi.org/10.1016/j.socscimed.2019.04.016.
Reyna-Vargas, M.E., Parmar, A., Lefebvre, D.L., Azad, M.B., Becker, A.B., Turvey, S.E., et al., 2022. Longitudinal associations between sleep habits, screen time and overweight, obesity in preschool children. Nat. Sci. Sleep 14, 1237-1247. https://doi.org/10.2147/NSS.S363211.
Schaefer, D.R., Simpkins, S.D., Vest, A.E., Price, C.D., 2011. The contribution of extracurricular activities to adolescent friendships: new insights through social network analysis. Dev. Psychol. 47 (4), 1141-1152. https://doi.org/10.1037/a0024091.
Virtanen, M., Jokela, M., Lallukka, T., Hanson, L.M., Pentti, J., Nyberg, S.T., et al., 2020. Long working hours and change in body weight: analysis of individual-participant data from 19 cohort studies. Int. J. Obes. 44 (6), 1368-1375. https://doi.org/10.1038/s41366-019-0480-3.
World Health Organization (WHO). 2020. Obesity and overweight facts. World Health Organization.
Yang, Q., Liu, W., Zhang, D., 2021. Small labels, big effects: the food labeling effect in marketing. Adv. Psychol. Sci. 29 (09), 1669-1683. https://doi.org/10.3724/SP.J.1042.2021.01669.
Zhang, Z., Wu, X., 2017. Occupational segregation and earnings inequality: rural migrants and local workers in urban China. Soc. Sci. Res. 61, 57-74. https://doi.org/10.1016/j.ssresearch.2016.06.020.
Zheng, K., Gao, W., Cao, W., Lv, J., Yu, C., Wang, S., et al., 2022. Education, income, and obesity: a nationwide chinese twin study. Obes. 30 (4), 931-942. https://doi.org/10.1002/oby.23390.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

The obesity loss effect of returns to education: Evidence from China

Status:

Version 1

Abstract

Figures

1. lntroduction

2. Theoretical analysis and research hypothesis

3. Methods

3.1Data

3.2 Variables

3.2.1 Dependent variable

3.2.2 Independent variable

3.2.3 Mechanism Variables

3.2.4 Control variables

3.3 Econometric models

3.3.1 Baseline regression model

3.3.2 Instrumental variable model

3.3.3 Quantile regression model

3.3.4. Mechanistic effect model

4. Results

5.1 Baseline regression model

5.2 Robustness tests

5.2.1 Substitution of dependent variable

5.2.2 Replacement regression model

5.2.3 Truncation processing

5.2.4 Correction of selection bias

5.3 Analysis of mechanism effects

5.4 Heterogeneity analysis

5.4.1. Heterogeneity analysis: effect of education on obesity

5.4.2 Heterogeneity analysis: mechanisms of the effect of education on obesity

5. Conclusion and Implications

Declarations

References

Additional Declarations

Status:

Version 1