Unhealthy consumption behaviors and their intergenerational persistence: The role of education

Accepted Manuscript Unhealthy consumption behaviors and their intergenerational persistence: The role of education

Yanjun Ren, Yanjie Zhang, Bente Castro Campos, Jens-Peter Loy PII: DOI: Reference:

S1043-951X(18)30104-4 doi:10.1016/j.chieco.2018.08.004 CHIECO 1208

To appear in:

China Economic Review

Received date: Revised date: Accepted date:

31 December 2017 27 July 2018 8 August 2018

Please cite this article as: Yanjun Ren, Yanjie Zhang, Bente Castro Campos, Jens-Peter Loy , Unhealthy consumption behaviors and their intergenerational persistence: The role of education. Chieco (2018), doi:10.1016/j.chieco.2018.08.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Unhealthy Consumption Behaviors and Their Intergenerational Persistence: the Role of Education

Yanjun REN 1,* [email protected]; Yanjie ZHANG2 [email protected] Leibniz; Bente CASTRO

Institute of Agricultural Economics, University of Kiel, Wilhelm-Seelig-Platz 6/7, 24118 Kiel,

RI

1

PT

CAMPOS1 [email protected]; Jens-Peter LOY1 [email protected]

Germany

Institute of Agricultural Development in Transition Economies (IAMO), Theodor-Lieser-Str. 2,

SC

2

EP T

ED

MA

Corresponding author.

AC C

*

NU

06120 Halle (Saale), Germany

ACCEPTED MANUSCRIPT 1. Introduction Cigarette smoking and alcohol drinking are two crucial behaviors that negatively affect health and longevity (Bazzano et al. 2007; Deng et al. 2006; Hao et al. 2004; Ruitenberg et al. 2002). The Chinese government is increasingly concerned about the public health expenditures caused by cigarette smoking and alcohol drinking. National legislators start to actively consider national

PT

bans on smoking in public and work places as well as on cigarette advertising. Nevertheless,

RI

tobacco control in China has remained particularly cumbersome due to the influence of the tobacco industry. There are over 300 million cigarette smokers in China consuming roughly one-

SC

third of the world’s cigarettes with substantial higher prevalence of smoking men (52.9%) than

NU

women (2.4%) compared with global smoking rates for men (36.9%) and women (7.3%) (WHO 2015). Almost 1.4 million people in China die annually from smoking-related diseases, and this

MA

number is expected to rise to over 3 million by 2050 if current smoking rates continue (Yang et al. 2015).

ED

Alcohol drinking is another health-related behavior. It is increasing faster in China than

EP T

elsewhere and shows a steady rise (Cochrane et al. 2003; Hao et al. 2005). The 2007 national survey of alcohol consumption has revealed that in China, 55.6% of men and 15.0% of women are frequent alcohol consumers (Li et al. 2011) 1 . Alcohol use disorders (AUDs) 2 are common

AC C

problems linked to disturbances in mental and physical health (Tang et al., 2013). Precisely, the AUD rates in China are 9.3% and 0.2% among men and women, respectively; these rates are comparatively larger than the AUD rates in other countries (WHO 2014). To control and better understand the determinants of cigarette smoking and alcohol consumption, a large number of studies have been conducted. These studies extensively focus on the impact of socio-demographic variables, such as income and price, on the occurrences of cigarette smoking

1 2

They consider drinking on 5-7 days per week as frequent drinking. AUDs encompass harmful patterns of drinking such as alcohol dependence and abuse.

ACCEPTED MANUSCRIPT and alcohol consumption. The impact of price on individuals’ alcohol and cigarette consumption has been extensively studied (Bishop et al. 2007; Tian and Liu 2011; Yu and Abler 2010). For instance, Bishop et al. (2007) find that price elasticity for cigarettes is approximately -0.5; however, income is less likely to affect cigarette smoking as the income elasticity is extremely low, ranging between 0.003 and 0.038. Yu and Abler (2010) find that neither price nor

PT

income has statistically significant impacts on cigarette smoking. Regarding alcohol

RI

consumption in China, Tian and Liu (2011) conclude that price has no significant impact since

SC

the price elasticity is virtually zero for beer and only −0.12 for liquor. Yen et al. (2009) also find that income does not affect alcohol consumption for men, while socio-demographic factors such

NU

as education, employment, and marital status play a significant role. Moreover, many studies suggest that parental consumption behavior may also influence filial

MA

consumption behavior (Charles et al. 2014; Melchior et al. 2010; Waldkirch et al. 2004; Wickrama et al. 1999). However, regarding cigarette smoking and alcohol drinking, the

ED

empirical results are mixed. Some studies find a positive correlation between parental and filial

EP T

alcohol consumption (Schmidt and Tauchmann 2011; Waldkirch et al. 2004), while others indicate no significant correlation (Yu 2003). Alternatively, filial consumption behavior might depend either on maternal or paternal consumption behavior. Francesconi et al. (2010) find that a

AC C

young adult who lives with an unmarried mother has a higher smoking propensity. Gundy (2002) finds that alcohol consumption is higher among children whose mothers typically have approximately three or more drinks daily or who drink on a weekly basis, while fathers’ alcohol consumption only positively influences sons’ drinking behavior. The studies suggest that parental consumption behavior influences filial health outcomes; nevertheless, it has largely been ignored in the literature on China. If there exists a high correlation of unhealthy consumption behaviors between generations, how could offspring be efficiently prevented from perpetuating their parents’ unhealthy consumption

ACCEPTED MANUSCRIPT behaviors? Education is expected to be one of the most reasonable prevention methods. However, a large body of literature has documented mixed results regarding the link between education and health behavior. Many studies reveal that education has a negative effect on the probability of smoking or binge drinking (Cowell 2006; Jensen and Lleras-Muney 2012; Kemptner et al. 2011; Nayga 1999), suggesting that more educated people are likely to have better health status

PT

and are less likely to engage in unhealthy behaviors, while other studies cannot provide strong

RI

evidence that education leads to healthier behavior (Reinhold and Jurges 2010; Xie and Mo 2014;

SC

Zhong 2015). Given the existing literature, however, the mechanism underlying the connection between education and unhealthy consumption behaviors is still not fully understood. To the best

NU

of our knowledge, only one study by Cowell (2006) has summarized three broad mechanisms that could be used to explain the connection between the education obtained and health

MA

behaviors: efficiency mechanisms, future opportunity costs, and unobserved heterogeneity. The efficiency mechanisms argue that people with higher education tend to allocate their resources

ED

more efficiently for a better health. For instance, education as a form of human capital investment can improve cognitive skills or health knowledge and enhance positive health

EP T

behavior accordingly (Berger and Leigh 1989; Kenkel 1991). While the mechanisms of future opportunity costs consider that education raises a person’s future utility that can be influenced by

AC C

current unhealthy behaviors, such as smoking and binge drinking, desire for a higher future utility can impel the higher educated to escape unhealthy consumption behaviors. According to the explanation from the mechanisms of unobserved heterogeneity, education and health behavior are correlated only because education is a proxy variable for unobserved variables, such as time preference or ability. Thus, it is particularly important to control for unobserved heterogeneity that is correlated with education and health or health behaviors (Cowell 2006). A number of recent studies has attempted to take unobserved heterogeneity or endogeneity of education into consideration, and most of them rely on instrumental variable methods (Dickson

ACCEPTED MANUSCRIPT 2013; Nayga 1999; Xie and Mo 2014); however, it is difficult to disentangle whether education impact is evidence of an efficiency mechanism or evidence of future opportunity costs or both of them. While the same issue exists in our study even through controlling for unobserved heterogeneity, our hypothesis is that education may discourage offspring from maintaining parents’ unhealthy

PT

habits, such as smoking and binge drinking, since more highly educated people might be more

RI

aware and knowledgeable of the harmful consequences of unhealthy consumption or are more

SC

likely to control their health-related behaviors. Nevertheless, focusing on investigating the causal relation between education and health render the precise mechanism irrelevant in our study.

NU

Meanwhile, to the best of our knowledge, there is limited research on measuring the impact of education on intergenerational persistence of unhealthy consumption behaviors, let alone on

MA

considering the unobserved heterogeneity of education in intergenerational persistence of these

ED

behaviors.

In addition to the problem of unobserved heterogeneity, the problem of reversed causality

EP T

between education and unhealthy behaviors also requires addressing the endogeneity problem. Currie and Hyson (1999) argue that a feedback relationship between education and health exists.

AC C

Positive health behavior could also positively affect the demand for education and school performance. For example, Zhao et al. (2012) find that smoking among teenagers can reduce their learning productivity. Based on mathematical test results, the authors find that smoking one cigarette per day can lower students’ scores by approximately 0.08 standard deviations. Balsa, Giuliano, and French (2011) also show that alcohol consumption negatively influences students’ academic performance. These findings suggest that there seems to be a strong reversed causal relationship between unhealthy consumption behaviors and education. Neglecting the endogeneity of education in estimating the impact of education on cigarette smoking and alcohol

ACCEPTED MANUSCRIPT consumption may lead to biased results. To obtain unbiased estimates requires carefully addressing the endogeneity of education. Changes in government regulations concerning education can serve as natural experiments to control for the endogeneity of education. For instance, law changes in education or schooling related aspects could affect individuals’ years of obtained education but are less likely to directly

PT

influence their health. These laws and regulations, therefore, generate exogenous variation in

RI

years of schooling both across regions and over time, such as changes in compulsory schooling

SC

laws (Kemptner et al. 2011; Xie and Mo 2014; Castro Campos et al. 2016), and the abolition of secondary school fees (Reinhold and Jurges 2010). Other schooling related aspects include, for

NU

example, college attendance during war time (de Walque 2007) and the number of academic track schools in a state (Amin et al. 2013). Xie and Mo (2014) investigate the causal effect of

MA

education on health using the Compulsory Schooling Law and the Provisions on the Prohibition of Using Child Labor as instruments for education; they cannot obtain conclusive results for

ED

smoking due to a violation of the exogeneity of their instruments. One possible reason might be

EP T

lack of controlling for cohort trends in their estimation, since two institutional changes operate based on the year of birth, failure to adequately control for cohort trends is likely to invalidate

AC C

the instrument (Brunello et al. 2013). Although many studies have estimated the causal link between health and education, studies have seldom investigated how education impacts intergenerational persistence of unhealthy consumption behaviors considering the endogeneity of education. In this paper, we address the question whether education has an impact on unhealthy consumption behaviors as well as on their intergenerational persistence. Our study contributes to the literature by (1) identifying the correlation of unhealthy consumption behaviors between offspring and parents, by (2) examining how education affects filial unhealthy consumption behaviors through two institutional changes

ACCEPTED MANUSCRIPT used as instrumental variables for education, and by (3) investigating the impact of education on intergenerational persistence of unhealthy consumption behaviors. The paper is organized as follows. In Section 2, we introduce the econometric models for unhealthy consumption participation and the estimation strategy of our study. In Section 3, we describe the CHNS data and present descriptive statistics of the main variables. In Section 4, we

PT

present our estimation results. In the last section, we discuss the results, draw conclusions, and

RI

provide policy recommendations.

SC

2. Empirical approach

NU

2.1 Benchmark model for intergenerational persistence of unhealthy consumption behavior

MA

Becker and Tomes (1979, 1986) initially presented an economic model of intergenerational persistence. Our benchmark model closely follows their model and is defined as: (1)

ED

𝑃𝑟𝑜𝑏(𝑌=1) = 𝛽0 + 𝜷𝟏 𝑷𝒂𝒓𝒆𝒏𝒕 + 𝜶𝑿 + 𝜀𝑖

EP T

where Y is a binary variable which is one if the child has a health-risk consumption behavior and zero if the child has no health-risk consumption behavior. Parent refers to parental consumption behaviors, and 𝜷𝟏 is used to examine the intergenerational effects of parental consumption

AC C

behaviors on the probability of children’s behavior. X is the vector of variables that control for individual and parents’ demographic variables, including gender, age, age squared, marriage status, residence, year of birth, logarithm of income, working status, and parental age and education. Moreover, year and regional dummies are included in X to control for year and regional fixed effects. 𝜀𝑖 is the disturbance and is assumed to be 𝜀𝑖 ~𝑁(0, 1). As Cowell (2006) points out, most empirical studies that focus on smoking and drinking are not only of concern for policy makers but also provide an interesting comparison to one another. In our study, we also estimate two main health-related behaviors: cigarette smoking and alcohol

ACCEPTED MANUSCRIPT drinking since both behaviors may have potential negative health consequences. Undoubtedly, smoking incorporates a high risk of premature death. In terms of alcohol consumption, a harmful effect of binge drinking on health has been widely observed (Bazzano et al. 2007), but some researchers also indicate that a certain light-to-moderate drinking might actually decrease allcause mortality (Hao et al. 2004; Murray et al. 2002). More recently, one significant study

PT

reveals the harmful effect of drinking by providing plausible evidence for the epidemiological

RI

association between alcohol consumption and enhanced cancer risk (Garaycoechea et al. 2018).

SC

Thus, we specify alcohol consumption into two indicators of “drinking” and “binge drinking” to consider the possible heterogeneity of education on alcohol consumption for different

NU

consumption levels.

MA

The main independent variable Parent is a set of dummy variables of maternal and paternal consumption behaviors. Six dummy variables are separately used to indicate whether the mother (father) is a current smoker, drinker, or binge drinker. First, we introduce maternal and paternal

ED

consumption behaviors in different model specifications to identify intergenerational persistence

EP T

of unhealthy consumption behaviors. As most of the studies have found that parental consumption behaviors have a positive impact on filial consumption behavior, 𝜷𝟏 is expected to

AC C

have a positive sign. Second, we gradually control income and working status to check the robustness of our results. Income and working status might be potentially endogenous but leaving them in the disturbance may bias estimation results. To control for correlation in the same region, we cluster standard errors at community level. 2.2 Impact of education on unhealthy consumption behaviors After identifying intergenerational persistence of unhealthy consumption behaviors, we further investigate the impact of education on unhealthy consumption behaviors (smoking, drinking, and binge drinking). Our empirical model is defined as:

ACCEPTED MANUSCRIPT 𝑃𝑟𝑜𝑏(𝑌=1) = 𝛽0′ + 𝜷′1 𝑷𝒂𝒓𝒆𝒏𝒕 + 𝛽2′ 𝑌𝑒𝑎𝑟𝑠𝑒𝑑𝑢𝑐 + 𝜶′ 𝑿 + 𝜇𝑖

(2)

Yearseduc is measured by the years of education an individual has acquired. X is the vector of control variables as given in Equation (1). 𝜇𝑖 is the disturbance with 𝜇𝑖 ~𝑁(0, 1). The coefficients 𝜷′𝟏 and 𝛽2′ are used to capture the intergenerational persistence of parents’

PT

unhealthy consumption behaviors on their children (as in Equation (1)) and the impact of

RI

education on these unhealthy consumption behaviors, respectively. We hypothesize that higher education leads to better health knowledge and self-control and could reduce the probability of

SC

unhealthy consumption behaviors such as smoking and binge drinking; thus, 𝛽2′ is expected to be

NU

negative. However, the impact of education on drinking is ambiguous since moderate drinking might not necessarily be an unhealthy consumption behavior. All control variables are the same

MA

as in model (1); standard errors are clustered on community level. To obtain a consistent estimate for education, the potential endogeneity of education has to be

ED

considered, since education might face reversed causality problem and unobserved heterogeneity

EP T

in health estimation. As mentioned earlier, law changes as quasi-experiments generate exogenous variation in years of schooling both across regions and over time. A change in the Compulsory Schooling Law in 1986 serves as a valid instrumental variable (Kemptner et al.,

AC C

2011; Xie and Mo, 2014; Huang, 2015; Castro Campos et al. 2016). Additionally, the enactment of the Provisions on the Prohibition of Using Child Labor in 1991, which aims to prohibit child labor in China, has increased educational attainment and serves as a valid instrument (Xie and Mo 2014; Castro Campos et al. 2016). The 9-year Compulsory Education Law (Law), which was officially enacted on July 1, 1986 is expected to boost middle school education and to eradicate illiteracy. The law regulates that children who have reached the age of six are to enroll in school without any tuition fee for nine compulsory years, covering six years of primary school and three years of junior middle school

ACCEPTED MANUSCRIPT (Fang et al. 2012). In general, young individuals complete the nine years of compulsory schooling at the age of 15; this indicates that individuals who were born after 1971 were affected by the law. Huang (2015) suggests that there are variations in the law’s implementation in different provinces. Thus, this study also considers geographic variation in the implementation of the compulsory schooling law in different parts of the country. Specifically, individuals born

PT

after 1971 are affected by the law in Beijing, Chongqing, Liaoning, and Heilongjiang; after 1972

RI

in Shandong, Jiangsu, Shanghai, Hubei, and Henan; after 1973 in Guizhou, and after 1976 in

SC

Hunan and Guangxi. We define individuals who were born after these threshold years with “1” because they are affected by the law and individuals who were born before these threshold years

NU

with “0” because they are unaffected by the law.

The other institutional change is the Provisions on the Prohibition of Using Child Labor

MA

(Provisions), which were enforced on April 15, 1991. As a complementary regulation to the Compulsory Education Law, these provisions indicate that children below 16 years of age are not

ED

allowed to be employed for any type of work, which enforced children of school age to continue

EP T

their education and start working only after completing their compulsory education at the age of 15. We define individuals who were born after 1975 with “1” because they are affected by the provisions, and individuals who were born before 1975 with “0” because they are unaffected by

AC C

the provisions. These two institutional changes serve as potentially valid instruments as they are supposed to highly influence individual’s education. However, individual’s unhealthy consumption is virtually impossible to be directly influenced by these two institutions. Hence, we introduce both institutional changes as instruments to address the endogeneity of education in all estimations 3 :

3

As argued by Xie and Mo (2014) that the five years gap between these two institutional changes in Ch ina might overlap with their effect on educational attain ment, the first stage to obtain the predicted years of education is based on equation (3) for the Law and the Provisions. Meanwhile, it is worth noting that the Prohibitions might influence health and health behaviors through other mechanisms. For instance, early entry to labor market may lead to heavy

ACCEPTED MANUSCRIPT ̂ 𝑌𝑒𝑎𝑟𝑠𝑒𝑑𝑢𝑐 = 𝛾0 + 𝛾1 𝐿𝑎𝑤 + 𝛾2 𝑃𝑟𝑜𝑣𝑖𝑠𝑖𝑜𝑛𝑠 + 𝛾3 𝑌𝑒𝑎𝑟𝑏𝑖𝑟𝑡ℎ + 𝝆𝑿 + 𝑢 𝑖

(3)

A valid instrument should have a high correlation with the instrumented variable and should be uncorrelated with the error component in the respective model. In other words, the law and the provision are assumed to affect individuals’ education but should not directly affect individuals’

PT

unhealthy consumption behaviors. Additionally, since years of compulsory education are increasing over cohorts, failure to adequately control for cohort trends is likely to invalidate the

RI

instrument. Because these legislative changes operate based on the year of birth, it is

SC

recommended that the estimation equations control for year of birth in the analysis (Brunello et al. 2013). Another potential problem for the estimation is the likelihood that standard errors are

NU

correlated within regions and cohorts. If that is the case, standard errors may be biased, leading

MA

to incorrect statistical inference. Therefore, all estimations are clustered by cohorts and communities to indicate that observations may be correlated within cohorts and communities but would be independent between cohorts and communities. In a similar vein, we also gradually

ED

control for income and working status to check the robustness of our results.

EP T

Moreover, a Wald test of exogeneity is applied to test whether the variable is indeed endogenous. If the test is rejected, the estimates from the regular Probit model (Equation 2) will be biased.

AC C

Under this circumstance, the instrumental Probit model should be preferred, but when instruments are weak, point estimators will be biased and the Wald test is unreliable. To remove the doubts of weak instruments, a joint F-test on the excluded instruments is used to test the relevance of the instruments for the potentially endogenous variable (Yearseduc). There are several alternative tests recommended by Finlay and Magnusson (2009) applicable when instruments are weak (e.g., the Anderson-Rubin (AR) test (Anderson and Rubin 1949), the Kleibergen-Moreira Lagrange Multiplier (LM) test (Kleibergen 2007; Moreira 2003), and the

work that could affect children’s health later on. Therefore, we also separately introd uce the two intuitional changes as instruments for education in all estimations, and the results are almost the same as when using both of them.

ACCEPTED MANUSCRIPT Conditional Likelihood-Ratio (CLR) test. In our study the CLR test is used to check the relevance of instruments because it is more powerful than the AR and LM tests (Andrews et al. 2006). Since two instruments are employed in each estimation, the J-test of overidentifying restrictions recommended by Finlay and Magnusson (2009) is applied in this study. Additionally, the K-J test as a combination of the K and J statistics is also applied to jointly test the structural

PT

parameter and exogeneity of instruments.

RI

2.3 Impact of education on intergenerational persistence of unhealthy consumption

SC

To detect how education impacts the intergenerational persistence of unhealthy consumption, we further introduce an interaction term between parental unhealthy consumption behaviors and

NU

offspring’s education. The estimated model is defined as follows:

𝜶′′ 𝑿 + 𝛿𝑖

MA

̂ ̂ 𝑷𝒓𝒐𝒃(𝒀=𝟏) = 𝛽0′′ + 𝜷′′𝟏 𝑷𝒂𝒓𝒆𝒏𝒕 + 𝛽2′ 𝑌𝑒𝑎𝑟𝑠𝑒𝑑𝑢𝑐 + 𝜷′𝟑 𝑷𝒂𝒓𝒆𝒏𝒕 ∗ 𝑌𝑒𝑎𝑟𝑠𝑒𝑑𝑢𝑐 + 𝛽4 𝑌𝑒𝑎𝑟𝑏𝑖𝑟𝑡ℎ + (4)

ED

̂ where 𝑌𝑒𝑎𝑟𝑠𝑒𝑑𝑢𝑐 is the estimated value of the years of education from Equation (3). The

EP T

definitions of dependent and control variables are the same as in Equation (1). Our main concern is the coefficient of the interaction term, 𝜷′𝟑. The null hypothesis is that additional education can

AC C

prevent individuals from perpetuating their parents’ unhealthy behaviors; therefore, 𝜷′𝟑 is assumed to be negative. Given the potential endogeneity problem of education, the interaction term between parental unhealthy consumption and the two instruments for education (the law and the provisions as discussed earlier) are used in Equation (4). 4 The controls are identical as in equation (2), with an additional term for year of birth (Yearbirth). The CLR, J, and K-J tests are similarly employed. 4

As we have controlled for the endogeneity of offspring’s education but not for the endogeneity of parental consumption behaviors, which could likewise be endogenous, caution should be exercised when talking about causal effects between education and intergenerational persistence. Nevertheless, it is hard to find an appropriate instrument that is correlated with parental consumption behaviors but uncorrelated with children’s consumption behaviors. In the framework of this paper, we thus assume that parental consumption behavior is exogenous leaving the endogeneity issue of parental consumption behavior for future research.

ACCEPTED MANUSCRIPT 3. Data and variables 3.1 Data

The data used in this study are from the China Health and Nutrition Survey (CHNS) over the period from 1991 to 2011. The CHNS uses a multistage, random cluster design in eight

PT

provinces (Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, and Guizhou) to select a stratified probability sample from the initial wave in 1989 with a partial sample. The selected

RI

provinces vary substantially in terms of geography, economic development, public resources,

SC

and health indicators. Following the sampling strategy, two cities (one large and one small,

NU

usually the provincial capital and a lower- income city) and four counties (stratified by income, one high-, one low-, and two middle- income) were selected in each province. Within cities, two

MA

urban and suburban communities were randomly selected. Within counties, one community in the capital city and three rural villages were randomly chosen. Afterwards twenty households per

ED

community were randomly selected for participation (Zhang et al. 2014). In doing so, the CHNS applies a multistage and random cluster process to draw a sample of roughly 4,400 households

EP T

with a total of approximately 26,000 individuals. By the survey in 2011, the CHNS included 9 provinces (Heilongjiang, Liaoning, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou)

AC C

and three autonomous cities (Beijing, Shanghai, Chongqing), consisting of 288 communities with 27,447 individuals from 5,884 households. The questions regarding our main dependent variables have been reported since 1991; thus, we drop the data for the first wave in 1989.

3.2 Variables Unhealthy behaviors in this study are identified by cigarette smoking and alcohol consumption. Regarding cigarette smoking, the respondents are asked “Do you still smoke cigarettes now?”; those who reported smoking (not smoking) are defined as smokers (non-smokers). Regarding alcohol consumption, the respondents are asked “Did you drink beer or any other alcoholic

ACCEPTED MANUSCRIPT beverage last year?”; those who reported drinking (not drinking) beer or any alcoholic beverage last year are defined as alcohol drinkers (non-alcohol drinkers). The respondents further reported the types of alcoholic beverages (beer, grape wine, rice wine, and liquor) and quantity (bottle for beer, Liang (Chinese ounce), or 50 grams for other alcoholic beverages) consumed in a typical week within the previous year. According to the quantity of each alcoholic beverage consumed,

PT

we calculated the percent of pure alcohol consumed based on Li et al. (2011). The percent of

RI

pure alcohol is 4% for beer, 52% for liquor, and 10% for wine; moreover, the bottle volume of beer is set at 640 ml, which is most common in China. According to the National Institute on

SC

Alcohol Abuse and Alcoholism (NIAAA) of the United States, one “standard” drink contains

NU

roughly 14 grams of pure alcohol. We define a male (female) who consumes 14 (7) drinks or

MA

more weekly as a binge drinker.5

We match filial alcohol and cigarette consumption with their parental consumption using

ED

parental-child identifiers to select a sample consisting of parental-offspring pairs. After merging all other variables regarding filial demographic, socio-economic and household background

EP T

information, we restrict our sample to offspring older than sixteen years of age given that in most countries moderate alcohol consumption and cigarette smoking is officially allowed after that

AC C

threshold year. Finally, our sample consists of 11,316 observations over the period from 1991 to 2011. There are approximately 30.4%, 35.3%, and 10.4% smokers, drinkers, and binge drinkers, respectively (Table 1). Regarding parental consumption behaviors, we observe that approximately 4.5% and 60.1% of mothers and fathers are smokers, respectively. 11.9% of

5

As defined by the National Institute on Alcohol Abuse and Alcoholism (NIAAA), for wo men, low-risk drinking is

no more than 3 drinks on any single day and no more than 7 drinks per week. For men, it is no more than 4 drinks on any single day and no more than 14 drin ks per week. The study by Li et al., (2011) has defined binge drinking as consumption of 50 grams or mo re pure alcohol for men/ 40 grams o r mo re for wo men on at least one day in the previous 12 months. Since we have no informat ion on daily drinking behavior, it shou ld be noted that the probability of binge drinking calculated from our measurement may be higher than that from the NIAAA.

ACCEPTED MANUSCRIPT mothers are drinkers and 4.5% are binge drinkers. Similar to smoking behavior, we also find that fathers have a higher probability of being drinkers and binge drinkers with proportions of 63.3% and 24.8%, respectively. In general, fathers are more likely to smoke and to drink as well as to be binge drinkers than are mothers. As stated by Tang et al. (2013) this is probably a cultural phenomenon in China that women are discouraged from drinking alcohol. Nevertheless, the

PT

authors also indicate that this difference between men and women might diminish in the future

RI

due to social change.

SC

Education is the key explanatory variable in this study; it is measured as formal years of education ranging from 0-21 years. In our sample, the average years of education obtained by

NU

offspring are approximately 9.637 years (Table 1), which indicates that most of them have

MA

obtained six years of primary education and three years of secondary education. The evidence from two institutional changes influencing individuals’ educational attainment shows that the

ED

Law and the Provisions affect approximately 59.6% and 47.3% of the total sample population, respectively. In addition, the proportions of smokers, drinkers, and binge-drinkers affected by

EP T

institutional change of the Law (the Provisions) are with 46.5% (35.6%), 51.6% (39.6%), and 54.6% (42.1%) comparatively smaller than that of non-smoker, non-drinker, and non-binge-

AC C

drinker which stand at 65.3% (52.4%), 63.9% (51.5%), and 60.2% (47.9%), respectively. There are significant differences in other key variables across various sub-samples. Regarding parental consumption behaviors, it is apparent that offspring who are smoker, binge drinker, and drinker are associated with higher frequencies of parents having the same behaviors. There is, moreover, a statistically significant difference in the years of education across sub-samples. On average non-smokers have higher education than do smokers, while drinkers have higher education than do non-drinkers. Although a slightly larger mean value of education is observable for binge-drinker than for non-binge-drinker, the T-test indicates that there is no statistical difference between them.

ACCEPTED MANUSCRIPT The other explanatory variables controlled for in this study are gender, age, age squared, working status (presently working), marital status, the logarithm of household income, residence (urban or rural), and parental age and education (Table 1). Regional dummies are included to account for disparities in education and healthcare investment across provinces, and for differences in provincial characteristics. Additionally, we also include dummies for the different survey waves

PT

to control for changes in individuals’ consumption behaviors over time. The results of crucial

RI

explanatory variables will be discussed in the next section.

SC

4. Results

NU

4.1 The intergenerational persistence of unhealthy consumption behaviors The estimates for the correlation between individuals’ unhealthy consumption behavior and their

MA

parents’ consumption behavior are presented in Table 2. We gradually introduce income and working status in the estimations to check the robustness of results. Estimates of key variables

ED

are almost the same no matter whether income and working status are controlled for or not. The

EP T

marginal effects are calculated based on the model with inclusion of income and working status. In line with previous studies (Green et al. 1991; Schmidt and Tauchmann 2011; Waldkirch et al. 2004), our results show that parents’ smoking and drinking behaviors have statistically

AC C

significantly positive correlations with children’s smoking and drinking behaviors. For the smoking estimation, the marginal effects show that individuals whose mothers or fathers are current smokers have a 9.5% or 6.4% higher probability of smoking than do those whose mothers or fathers are non-smokers, respectively. Similar findings are observed for drinking and binge drinking. Parents’ drinking and binge drinking behaviors have statistically significant positive correlations with children’s drinking behaviors in all model specifications. Specifically, the marginal effects for drinking show that individuals whose mothers or fathers are drinkers are 19.3% and 14.4%, respectively, more

ACCEPTED MANUSCRIPT likely to be also drinkers. Surprisingly, the results show that drinking and smoking mothers have higher contributions to the probability of offspring being smokers or drinkers than have smoking or drinking fathers. Regarding binge drinking, individuals whose mothers or fathers are binge drinkers have 12.2% or 11.9% higher probability to also be binge drinkers, respectively. The results indicate that (1) there exists a significant intergenerational persistence of cigarette and

PT

alcohol consumption, and (2) that offspring is more likely to persist with alcohol consumption

RI

rather than with cigarette smoking.

SC

Regarding demographic variables, men are more likely to be smoker, drinker, and binge drinker than are women. We find an inverted U-shape between age and unhealthy consumption

NU

behaviors, suggesting that with additional age, probabilities to consume cigarettes and alcohol first increase and then decrease. Interestingly, married people are more likely to smoke and drink,

MA

while marital status is not significant for binge drinking. People living in urban areas are more likely to be drinker and binge drinker than are people from rural areas. Importantly, parental

ED

education has significantly negative effects on children’s smoking behavior but is insignificant

EP T

for drinking behaviors implying that policies aimed at improving parental education might be an efficient way to reduce children’s cigarettes consumption. It is worth noting that when gradually introducing income and working status, the coefficients of the concerned variables remain

AC C

unchanged, suggesting that the estimates from our estimations are generally robust. Since the results for demographic variables are almost the same across different model specifications, interpretation is focused on the main variables in the following sections. 4.2 Impact of education on unhealthy consumption behaviors To

investigate

how education affects

unhealthy consumption behaviors and

their

intergenerational persistence, we introduce education in our benchmark model (Equation (2)). The estimation results are presented in Table 3. The coefficients of parental consumption

ACCEPTED MANUSCRIPT behaviors that indicate intergenerational persistence of smoking, drinking, and binge drinking are slightly different in magnitude compared with estimates from benchmark models, but significance levels are the same. We find that education has a statistically significant impact on smoking and drinking, but it shows no significant effect on binge drinking. The marginal effect of education is negative and

PT

statistically significant for smoking, suggesting that higher education decreases the probability of

RI

smoking. In particular, on average, an additional year of education leads to a 0.7% decrease in

SC

the probability of smoking, which is unexpectedly lower than in other studies (Cowell 2006; Huang 2015; Kemptner et al. 2011; Reinhold and Jurges 2010). With respect to drinking, the

NU

marginal effect of education is positive and statistically significant, indicating that an additional year of education might give rise to the probability of being a drinker by approximately 0.3%.

MA

However, as the coefficients of education from the estimations might be biased, the findings have to be treated with caution due to the potential endogeneity of education. To further address

ED

this issue, we use an instrumental Probit estimation to better understand the impact of education

EP T

on smoking, drinking, and binge drinking.

To further clarify the impact of education on these three consumption behaviors, two

AC C

institutional changes (Law and Provisions) are introduced as instruments for years of education, respectively. As discussed previously, a valid instrument is required to be highly correlated with the instrumented variable and to be uncorrelated with the disturbance. We employ OLS regressions to investigate the impact of the Law and the Provisions on years of education according to Equation (3). The results are presented in Table 4. The estimates from various model specifications suggest that these two instrumental variables have significantly positive impacts on educational attainment. Generally, individuals regulated by the Law or the Provisions receive approximately 0.3 or 1.0 more years of education, respectively (Table 4, column 9). The estimated coefficient of the Law is lower, but the coefficient of the provisions is higher than

ACCEPTED MANUSCRIPT those reported by Xie and Mo (2014). A possible reason could be related to different sample restrictions since Xie and Mo (2014) use the sample from 1997 to 2006, but we use data from 1991 to 2011 and only consider observations with parental information reported. Moreover, lack of controlling for year-of-birth in the analysis might be another reason given that these legislative changes operate based on the year-of-birth.

PT

After observing a high correlation between education and the two instrumental variables, we

RI

separately introduce the variables Law and Provisions into Equation (2) and use an IV-Probit

SC

model in our analysis. The estimation results are reported in Table 5. Education remains negative and statistically significantly correlated with smoking and binge drinking but with a larger

NU

magnitude compared to the regular Probit model. As expected, the coefficient of education for drinking changes from a statistically significantly positive sign in the regular Probit model to a

MA

statistically significantly negative sign in the instrumented Probit model suggesting that

ED

additional education, as expected, actually decreases the probability of drinking. From the Wald test of exogeneity for the instrumented variables, the null hypothesis of no

EP T

endogeneity is rejected as the P value is less than 0.05 (Table 5) for most model specifications; this indicates that the estimates from the regular Probit model are inappropriate and likely to be

AC C

biased. Therefore, an instrumented Probit model is employed. To check whether the instruments considered are only weakly correlated with included endogenous variables, F statistics and CLR results are reported at the bottom of Table 5. We find that there is no weak instruments problem in our estimation as the CLR tests are all rejected at the conventional level of significance (at the 5% level); similar evidence can be found from the F statistics on excluded instruments. To further examine the exogeneity of instrumental variables, the J-Test for over- identifying restriction is applied. The results show that the over- identification tests in the second stage cannot be rejected in all model specifications at the conventional level of significance (at the 5% level). Additionally, as a combination of weak instrument and over-identification test, the K-J

ACCEPTED MANUSCRIPT test is rejected, suggesting that the instrumental variables of the two institutional changes are jointly valid instruments (Table 5). The estimates from the instrumental model are consistent and convincing. We conclude that an additional year of education reduces the probability of smoking, drinking, and binge drinking by 4.1%, 3.4%, and 1.6%, respectively (Table 5). Our findings are in line

PT

with previous studies (Cowell 2006; Huang 2015; Kemptner et al. 2011; Reinhold and Jurges

RI

2010); while the estimated impact of education on smoking is higher than that reported by Huang

SC

(2015), who finds that an additional year of education will decrease the smoking probability by 1.5 percentage points. The reason is that we use a subsample from the whole population, and the

NU

average age (approximately 23 years) is lower than the 32.46 years reported by Huang (2015). This difference suggests that improving education to reduce smoking might have a specific

MA

influence on younger people.

ED

Regarding the coefficient of parents’ unhealthy consumption behaviors, we find that parents’ unhealthy consumption behaviors have statistically significant and positive correlations with

EP T

children’s unhealthy consumption behaviors; this is similar to the findings from the benchmark model (Table 2). Specifically, we find that the intergenerational persistence of smoking from

AC C

both the mother and the father is, with 5.8% and 4.5%, respectively, comparatively lower than in the benchmark model, where it stands at approximately 9.5% and 6.4% for the mother and the father, respectively. However, there are only slight changes in the intergenerational persistence of drinking (17.4% from the mother, 13.1% from the father) and binge drinking (12.1% from the mother, 12.3% from the father) compared with the regular Probit model, in which the intergenerational persistence is 19.3% (14.4%) from the mother (father) for drinking and 12.2% (12.2%) from the mother (father) for binge drinking. Our conclusion remains that the intergenerational persistence from the mother is higher than that from the father in smoking and drinking behaviors, which is in line with Brion's et al. (2010) findings.

ACCEPTED MANUSCRIPT 4.3 Impact of education on intergenerational persistence of unhealthy consumption behaviors The estimation results for Equation (4) considering the interaction term between years of education and parents’ unhealthy consumption behaviors are presented in Table 6. For the

PT

interaction term in Equation (4), we use an instrumental Probit model by including an interaction term between parental unhealthy consumption behaviors and education, respectively. As shown

RI

in columns (1)-(3), the interaction term between “Smoke Mother” and “Yearseduc” is

SC

insignificant, indicating that education has no specific impact on individuals whose mothers are current smokers compared with offspring whose mothers are non-smokers. However, the

NU

coefficient for the interaction term between “Smoke Father” and “Yearseduc” is statistically

MA

significantly negative; this suggests that an additional year of education can actually decrease the probability of smoking for individuals with smoking fathers. Regarding binge drinking, we find that an additional year of education cannot prevent intergenerational persistence of binge

ED

drinking from mothers, but we do find it can counteract the intergenerational persistence from

EP T

fathers, as the coefficient of the interaction term (Binge Father*Yearseduc) is statistically significantly negative as shown in columns (7)-(9). However, we also find that education has

AC C

neither a statistically significant impact on the intergenerational persistence of drinking from mothers nor from fathers.

The Wald test of exogeneity for the instrumented variables indicates that the estimates from the instrumented Probit model are consistent. The CLR tests are all rejected at the 5% level of significance, indicating that there is no weak instruments problem in our estimation. The overidentification test cannot be rejected in all model specifications for smoking, drinking, and binge drinking. The K-J test as a combination of weak instrument and overidentification test is not rejected, indicating that the instruments considered are valid. Hence, the estimates from the instrumental model are unbiased and convincing.

ACCEPTED MANUSCRIPT We conclude that an additional year of education can counteract intergenerational persistence of unhealthy consumption behaviors from fathers but not from mothers. Our findings imply that policies designed to decrease intergenerational persistence of unhealthy consumption behaviors through education may not be efficient when mothers have unhealthy consumption behaviors; it might require other measures that directly influence mothers’ consumption behaviors rather than

PT

filial education. One potential reason is that mothers are often the primer caregivers, especially

RI

during childhood. If mothers have unhealthy consumption behaviors, offspring would be easily

SC

affected. The other reason could be genetic factors. It has been well documented that maternal smoking during pregnancy is a major public health concern with clearly established negative

NU

consequences for both the mother and the baby. Maternal smoking during pregnancy can lead to increased psychological problems in offspring (Brion et al. 2010), which might increase the

MA

probability of smoking and the difficulty to stop smoking later in life.

ED

4.4 Robustness check for education impact: rural versus urban To further check the robustness of our findings, we classify our sample into a rural and an urban

EP T

sample, as they are significantly different in terms of economic development, education and public resources. Using both institutional changes as instruments for education 6 , we follow

AC C

Equation (3) to investigate the impact of education on unhealthy consumption behaviors for rural and urban samples. The estimation results are presented in Table A1. For the rural sample, the Wald test of exogeneity is rejected at the conventional level as presented in Table A1, suggesting that education is actually endogenous in the estimations. Therefore, IV Probit estimation should be preferred. Evidence from the weak instrumental test (CLR) indicates that there is no weak instrument problem for the rural sample in all estimations, indicating that the two instruments are highly correlated with years of education. The J test for 6

We also separately instrument education by the Nine-year Co mpulsory Education Law and the Provisions on the Prohibition of Using Child Labor; the results are identical with the results when using both of them.

ACCEPTED MANUSCRIPT over- identification restriction cannot be rejected for all estimations for the rural sample, implying no over- identification problem. The two instruments employed to address the endogeneity of education are valid for all unhealthy consumption estimations for the rural sample. Nevertheless, this is only true for smoking for the urban sample, for which all tests are qualified. The IV Probit estimation results for drinking and binge drinking might be biased due to weak instrument

PT

problems given that the null hypothesis from the weak instrument test (CLR) and K-J test cannot

RI

be rejected at the conventional level of significance (5% level of significance).

SC

Generally, education has a significantly negative effect on smoking; one additional year of education will reduce the probability of smoking by 4.2% for both rural and urban samples.

NU

Regarding drinking and binge drinking, the results show that an additional year of education can reduce the probability of drinking and binge drinking by 3.5% and 1.9%, respectively. However,

MA

it is difficult to obtain conclusive results for the urban sample due to the problem of weak

ED

instruments.

Regarding the estimation for the impact of education on intergenerational persistence of

EP T

unhealthy consumption behaviors as shown in Table A2, the results indicate that additional education can counteract intergenerational persistence of smoking and binge drinking from the

AC C

father but not from the mother for the rural sample.The weak instrument tests (CLR) and overidentification (J) test remove the doubts on existence of weak instruments problem and endogeneity of instruments for the rural sample (this is also proved by the K-J test). However, the K-J test is not rejected for the urban sample, suggesting that there is either a weak instrument problem or endogenous instruments, although the weak instrument and over- identification tests are qualified. Based upon these tests, we cannot obtain conclusive results for the urban sample because of weak instrument problem or endogenous instruments. To sum up, the results from the estimations for rural samples are generally in line with the findings for the full sample, but caution should be exercised when interpreting the results of the

ACCEPTED MANUSCRIPT impact of education on drinking and binge drinking for the urban sample due to weak instrument problems or endogenous instruments. 5. Conclusion This paper investigates the intergenerational persistence of unhealthy consumption behaviors and

PT

the impact of education on unhealthy consumption behaviors and its intergenerational persistence. After identifying a statistically significant correlation between parental and filial

RI

unhealthy consumption behaviors, we attempt to illustrate the impact of education on filial

SC

unhealthy consumption behaviors by first applying standard Probit models. However, due to the potential endogeneity of education, more rigorous estimations are required to obtain more

NU

reliable results. As quasi-experiments, the Nine-year Compulsory Education Law and the

MA

Provisions on the Prohibition of Using Child Labor have significant influence on educational attainment (Xie and Mo 2014; Castro Campos et al. 2016). We employ an instrumental variable estimation approach by using these two institutional changes as instruments for education in

ED

unhealthy consumption equations. By introducing an interaction term between parental

EP T

unhealthy consumption and filial education, we further clarify the impact of education on the intergenerational persistence of unhealthy consumption. The data used is from the CHNS over

AC C

the period from 1991 to 2011.

The empirical estimation results indicate that in China parental unhealthy consumption behaviors have statistically significant and positive correlations with children’s unhealthy consumption behaviors; therefore, our findings suggest that a significant intergenerational persistence of unhealthy consumption behaviors exists. Precisely, the probability of intergenerational persistence of smoking is approximately 5.8% and 4.5% from mothers and fathers, respectively, while that of binge drinking is approximately 12.1% and 12.3% from mothers and fathers, respectively. Drinking behavior has the largest probability of intergenerational persistence with approximately 17.4% and 13.1% from mothers and fathers, respectively.

ACCEPTED MANUSCRIPT We also find that an additional year of education has a statistically significant negative impact on smoking, drinking, and binge drinking. One additional year of education decreases the probability of smoking, drinking, and binge drinking by 4.1%, 3.4%, and 1.6%, respectively. Another interesting finding is that an additional year of education can counteract intergenerational persistence of smoking and binge drinking from the father, but it has no impact

PT

on intergenerational persistence from the mother. This is indeed a very interesting finding and

RI

needs to be further investigated such as with tools from behavioral economics to provide

SC

enhanced policy recommendations. Since drinking behavior might be ambiguous as an unhealthy consumption behavior, we find that there is no impact of education on the intergenerational

NU

persistence of drinking.

From a policy perspective, our findings of largely significant negative correlations between

MA

education and intergenerational persistence of unhealthy consumption behaviors suggest that social mobility is stagnant in China. Offspring from families with less education tend not only to

ED

be less educated as well but also to be unhealthier and poorer. Public policy might address

EP T

stagnant social mobility through enhancing equal access to opportunities. Regarding improving health outcomes, health education including anti-smoking and anti-alcohol

AC C

campaigns about addictive properties and other risks of alcohol and cigarette consumption might be crucial to raise people’s awareness and trigger positive change. It is necessary that parents are aware of teaching children about negative impacts of alcohol and cigarette consumption and be good “role-models”. Increased public education and health campaigns are crucial in this regard. Moreover, the welfare loss stemming from alcohol and cigarette consumption which trigger negative impacts on health and education needs to be counteracted through more rigorous alcohol and tobacco controls. In general, developed countries apply a policy- mix to control alcohol and cigarette consumption. Alcohol- related problems are largely addressed through taxation, consumption restriction for the underage, and limited availability of alcoholic

ACCEPTED MANUSCRIPT beverages. However, in China there is no official age- limit for purchasing or consuming alcohol. Alcohol use is prevalent among adolescents in China (Li et al. 1996); therefore, it is crucial that well-designed policies are addressed to limit alcohol consumption among adolescents taking into account the intergenerational impacts as this study has shown. Smoking related problems are largely addressed through the prohibitions of smoking in indoor

PT

public places. Smoking is banned in almost all public places in Europe, the US and Australia

RI

among other countries. Such restrictive measures on smoking stem from the fact that smoking

SC

not only negatively affects smokers but also non-smokers through exposure to passive smoking. Since 2011 smoking has been officially banned in public places in China but with little positive

NU

outcomes. The World Health Organization (WHO) Framework Convention on Tobacco Control

MA

(FCTC) might be useful for helping implement more rigorous tobacco controls in China. In particular, our results suggest that policies oriented to control smoking and binge drinking

ED

should take parents’ consumption behaviors into consideration; education is expected to be an efficient way to control unhealthy consumption behaviors in China; however, education may not

EP T

be an efficient way to prevent intergenerational persistence of unhealthy consumption from the mother which needs to be further explored with additional research.

AC C

A limitation of our study is that due to the absence of data on consumers’ daily alcohol consumption, the definition of binge drinking in our study is different from other studies (see Li et al., 2011), which may induce a higher probability of binge drinking in our sample. Further research needs to be conducted to shed light on the mechanisms or channels through which unhealthy consumption behaviors transmit to the next generation. More importantly, once mother’s unhealthy consumption behaviors persist in offspring’s consumption behaviors, they cannot be readily changed. From a public health perspective, policies and campaigns oriented towards maternal health might improve offspring’s health accordingly. Additionally, future research is suggested to distinguish between the mechanisms (efficiency and opportunity costs)

ACCEPTED MANUSCRIPT through which education impacts unhealthy consumption behaviors. By doing so, it would be valuable not only for individuals to benefit from better health from higher education but also for policy makers to improve public health nationwide. For instance, if efficiency mechanisms play a significant role in transferring education impact, any policy targeting at health production and resources use would promote individual’s health efficiently. Whereas, if future opportunity costs

PT

cannot be ignored in interpreting education impact, then any polic y that affects individual’s

SC

may also influence their unhealthy behavior (Cowell 2006).

RI

future opportunity costs (e.g., assistance in job mobility, job search, retirement, income taxation)

NU

This research uses data from the China Health and Nutrition Survey (CHNS). We thank the

MA

National Institute for Nutrition and Health, China Center for Disease Control and Prevention, Carolina Population Center (P2C HD050924, T32 HD007168), the University of North Carolina

ED

at Chapel Hill, the NIH (R01-HD30880, DK056350, R24 HD050924, and R01-HD38700) and the NIH Fogarty International Center (D43 TW009077, D43 TW007709) for financial support

EP T

for the CHNS data collection and analysis files from 1989 to 2015 and future surveys, and the China-Japan Friendship Hospital, Ministry of Health for support for CHNS 2009, Chinese

AC C

National Human Genome Center at Shanghai since 2009, and Beijing Municipal Center for Disease Prevention and Control since 2011. The authors also acknowledge the financial support from the China Scholarship Council for conducting this research.

ACCEPTED MANUSCRIPT References Amin, V., Behrman, J. R., & Spector, T. D. (2013). Does more schooling improve health outcomes and health related behaviors? Evidence from U.K. twins. Economics of Education Review, 35, 134–148. doi:10.1016/j.econedurev.2013.04.004 Anderson, T. W., & Rubin, H. (1949). Estimation of the Parameters of a Single Equation in a

PT

Complete System of Stochastic Equations. The Annals of Mathematical Statistics, 20(1),

RI

46–63. doi:10.1214/aoms/1177730090

SC

Andrews, D. W. K., Moreira, M. J., & Stock, J. H. (2006). Optimal Two-Sided Invariant Similar Tests for Instrumental Variables Regression. Econometrica, 74(3), 715–752.

achievement

in

high school.

NU

Balsa, A. I., Giuliano, L. M., & French, M. T. (2011). The effects of alcohol use on academic Economics of

review,

30(1),

1–15.

MA

doi:10.1016/j.econedurev.2010.06.015

education

Bazzano, L. A., Gu, D., Reynolds, K., Wu, X., Chen, C.-S., Duan, X., et al. (2007). Alcohol

ED

consumption and risk for stroke among Chinese men. Annals of Neurology, 62(6), 569– 578. doi:10.1002/ana.21194

EP T

Becker, G. S., & Tomes, N. (1986). Human capital and the rise and fall of families. Journal of Labor Economics, 4(3 Pt. 2), 1–47.

AC C

Becker, G., & Tomes, N. (1979). An Equilibrium Theory of the Distribution of Income and Intergenerational Mobility. Journal of Political Economy, 87(6), 1153–89. Berger, M. C., & Leigh, J. P. (1989). Schooling, Self-Selection, and Health. The Journal of Human Resources, 24(3), 433–455. doi:10.2307/145822 Bishop, J. A., Liu, H., & Meng, Q. (2007). Are Chinese smokers sensitive to price? China Economic Review, 18(2), 113–121. doi:10.1016/j.chieco.2006.12.001

ACCEPTED MANUSCRIPT Brion, M.-J., Victora, C., Matijasevich, A., Horta, B., Anselmi, L., Steer, C., et al. (2010). Maternal smoking and child psychological problems: Disentangling causal and noncausal effects. Pediatrics, 126(1), e57–e65. doi:10.1542/peds.2009-2754 Brunello, G., Fabbri, D., & Fort, M. (2013). The Causal Effect of Education on Body Mass: Evidence

from

Europe.

Journal

of

Labor

Economics,

195–223.

PT

doi:10.1086/667236

31(1),

RI

Castro Campos, B., Ren, Y., & Petrick, M. (2016). The impact of education on income

253–267. doi:10.1016/j.chieco.2016.10.007

SC

inequality between ethnic minorities and Han in China. China Economic Review, 41,

NU

Charles, K. K., Danziger, S., Li, G., & Schoeni, R. (2014). The Intergenerational Correlation of Consumption Expenditures. The American economic review, 104(5), 136.

MA

Cochrane, J., Chen, H., Conigrave, K. M., & Hao, W. (2003). Alcohol use in China. Alcohol and Alcoholism (Oxford, Oxfordshire), 38(6), 537–542.

ED

Cowell, A. J. (2006). The Relationship between Education and Health Behavior: Some Empirical Evidence. Health Economics, 15(2), 125–146.

EP T

Currie, J., & Hyson, R. (1999). Is the Impact of Health Shocks Cushioned by Socioeconomic Status? The Case of Low Birthweight. The American Economic Review, 89(2), 245–250.

AC C

de Walque, D. (2007). Does education affect smoking behaviors?: Evidence using the Vietnam draft as an instrument for college education. Journal of Health Economics, 26(5), 877– 895. doi:10.1016/j.jhealeco.2006.12.005 Deng, J., Zhou, D. H. D., Li, J., Wang, Y. J., Gao, C., & Chen, M. (2006). A 2-year follow-up study of alcohol consumption and risk o f dementia. Clinical Neurology and Neurosurgery, 108(4), 378–383. doi:10.1016/j.clineuro.2005.06.005 Dickson, M. (2013). The Causal Effect of Education on Wages Revisited. Oxford Bulletin of Economics and Statistics, 75(4), 477–498.

ACCEPTED MANUSCRIPT Fang, H., Eggleston, K. N., Rizzo, J. A., Rozelle, S., & Zeckhauser, R. J. (2012). The Returns to Education in China: Evidence from the 1986 Compulsory Education Law (Working Paper

No.

18189).

National

Bureau

of

Economic

Research.

http://www.nber.org/papers/w18189. Accessed 1 September 2015 Finlay, K., & Magnusson, L. M. (2009). Implementing Weak Instrument Robust Tests for a

PT

General Class of Instrumental Variables Models (Working Paper No. 0901). Tulane

RI

University, Department of Economics. https://ideas.repec.org/p/tul/wpaper/0901.html.

SC

Accessed 21 November 2016

Francesconi, M., Jenkins, S. P., & Siedler, T. (2010). The Effect of Lone Motherhood on the

NU

Smoking Behavior of Young Adults. Health Economics, 19(11), 1377–1384. Garaycoechea, J. I., Crossan, G. P., Langevin, F., Mulderrig, L., Louzada, S., Yang, F., et al.

MA

(2018). Alcohol and endogenous aldehydes damage chromosomes and mutate stem cells. Nature, 553(7687), 171–177. doi:10.1038/nature25154

ED

Green, G., Macintyre, S., West, P., & Ecob, R. (1991). Like parent like child? Associations between drinking and smoking behaviour of parents and their children. British Journal of

EP T

Addiction, 86(6), 745–758. doi:10.1111/j.1360-0443.1991.tb03100.x Gundy, K. V. (2002). Gender and Intergenerational Transmission of Alcohol Use Patterns: An

AC C

Analysis of Adult Children in Moscow. Substance Use & Misuse, 37(1), 65–87. doi:10.1081/JA-120001497 Hao, W., Chen, H., & Su, Z. (2005). China: alcohol today. Addiction, 100(6), 737–741. doi:10.1111/j.1360-0443.2005.01036.x Hao, W., Su, Z., Liu, B., Zhang, K., Yang, H., Chen, S., et al. (2004). Drinking and drinking patterns and health status in the general population of five areas of China. Alcohol and Alcoholism, 39(1), 43–52.

ACCEPTED MANUSCRIPT Huang, W. (2015). Understanding the Effects of Education on Health: Evidence from China (SSRN Scholarly Paper No. ID 2655246). Rochester, NY: Social Science Research Network. https://papers.ssrn.com/abstract=2655246. Accessed 4 November 2016 Jensen, R., & Lleras-Muney, A. (2012). Does Staying in School (and Not Working) Prevent Teen Smoking and Drinking? Journal of Health Economics, 31(4), 644–657.

PT

Kemptner, D., Jurges, H., & Reinhold, S. (2011). Changes in Compulsory Schooling and the

RI

Causal Effect of Education on Health: Evidence from Germany. Journal of Health

SC

Economics, 30(2), 340–354.

Kenkel, D. S. (1991). Health Behavior, Health Knowledge, and Schooling. Journal of Political

NU

Economy, 99(2), 287–305.

Kleibergen, F. (2007). Generalizing weak instrument robust IV statistics towards multiple

MA

parameters, unrestricted covariance matrices and identification statistics. Journal of Econometrics, 139(1), 181–216. doi:10.1016/j.jeconom.2006.06.010

ED

Li, X., Fang, X., Stanton, B., Feigelman, S., & Dong, Q. (1996). The rate and pattern of alcohol consumption among Chinese adolescents. Journal of adolescent health, 19(5), 353–361.

EP T

Li, Y., Jiang, Y., Zhang, M., Yin, P., Wu, F., & Zhao, W. (2011). Drinking behaviour among men and women in China: the 2007 China Chronic Disease and Risk Factor Surveillance.

AC C

Addiction, 106(11), 1946–1956. Melchior, M., Chastang, J.-F., Mackinnon, D., Galéra, C., & Fombonne, E. (2010). The intergenerational transmission of tobacco smoking—The role of parents’ long-term smoking

trajectories.

Drug

and

Alcohol

Dependence,

107(2–3),

257–260.

doi:10.1016/j.drugalcdep.2009.10.016 Moreira, M. J. (2003). A Conditional Likelihood Ratio Test for Structural Models. Econometrica, 71(4), 1027–1048. doi:10.1111/1468-0262.00438

ACCEPTED MANUSCRIPT Murray, R. P., Connett, J. E., Tyas, S. L., Bond, R., Ekuma, O., Silversides, C. K., & Barnes, G. E. (2002). Alcohol Volume, Drinking Pattern, and Cardiovascular Disease Morbidity and Mortality: Is There a U-shaped Function? American Journal of Epidemiology, 155(3), 242–248. doi:10.1093/aje/155.3.242 Nayga, R. M., Jr. (1999). A Note on Schooling and Smoking: The Issue Revisited. Education

PT

Economics, 7(3), 253–258.

RI

Reinhold, S., & Jurges, H. (2010). Secondary School Fees and the Causal Effect of Schooling on

SC

Health Behavior. Health Economics, 19(8), 994–1001.

Ruitenberg, A., van Swieten, J. C., Witteman, J., Mehta, K. M., van Duijn, C. M., Hofman, A., &

NU

Breteler, M. (2002). Alcohol consumption and risk of dementia: the Rotterdam Study. The Lancet, 359(9303), 281–286. Accessed 24 February 2014

MA

Schmidt, C. M., & Tauchmann, H. (2011). Heterogeneity in the intergenerational transmission of alcohol consumption: A quantile regression approach. Journal of Health Economics,

ED

30(1), 33–42. doi:10.1016/j.jhealeco.2010.09.005 Tang, Y., Xiang, X., Wang, X., Cubells, J. F., Babor, T. F., & Hao, W. (2013). Alcohol and

EP T

alcohol-related harm in China: policy changes needed. Bulletin of the World Health Organization, 91(4), 270–276.

AC C

Tian, G., & Liu, F. (2011). Is the Demand for Alcoholic Beverages in Developing Countries Sensitive to Price? Evidence from China. International Journal of Environmental Research and Public Health, 8(6), 2124–2131. doi:10.3390/ijerph8062124 Waldkirch, A., Ng, S., & Cox, D. (2004). Intergenerational linkages in consumption behavior. Journal of Human Resources, 39(2), 355–381. WHO.

(2014).

Global

status

report

on

alcohol

and

health

2014.

http://www.who.int/substance_abuse/publications/global_alcohol_report/en/. 17 November 2015

WHO. Accessed

ACCEPTED MANUSCRIPT WHO. (2015). Smoke- free Policies in China – Evidence of Effectiveness and Implications for Action. http://www.wpro.who.int/china/mediacentre/factsheets/smoke_free_20151019/en/. Accessed 30 November 2015 Wickrama, K. A. S., Conger, R. D., Wallace, L. E., & Elder, G. H. (1999). The Intergenerational

PT

Transmission of Health-Risk Behaviors: Adolescent Lifestyles and Gender Moderating

RI

Effects. Journal of Health and Social Behavior, 40(3), 258–272. doi:10.2307/2676351

29, 1–18. doi:10.1016/j.chieco.2013.12.003

SC

Xie, S., & Mo, T. (2014). The impact of education on health in China. China Economic Review,

NU

Yang, G., Wang, Y., Wu, Y., Yang, J., & Wan, X. (2015). The road to effective tobacco control in China. The Lancet, 385(9972), 1019–1028. doi:10.1016/S0140-6736(15)60174-X

censored

system

MA

Yen, S. T., Yuan, Y., & Liu, X. (2009). Alcohol co nsumption by men in China: A non-Gaussian approach.

China

Economic

Review,

20(2),

162–173.

ED

doi:10.1016/j.chieco.2009.03.001

Yu, J. (2003). The association between parental alcohol-related behaviors and children’s

EP T

drinking. Drug and Alcohol Dependence, 69(3), 253–262. doi:10.1016/S03768716(02)00324-1

AC C

Yu, X., & Abler, D. (2010). Interactions between cigarette and alcohol consumption in rural China.

The

European

Journal

of

Health

Economics,

11(2),

151–160.

doi:10.1007/s10198-009-0157-2 Zhang, B., Zhai, F., Du, S., & Popkin, B. M. (2014). The China Health and Nutrition Survey, 1989–2011. Obesity reviews : an official journal of the International Association for the Study of Obesity, 15(0 1). doi:10.1111/obr.12119 Zhao, M., Konishi, Y., & Glewwe, P. (2012). Does Smoking Affect Schooling? Evidence from Teenagers in Rural China. Journal of Health Economics, 31(4), 584–598.

ACCEPTED MANUSCRIPT Zhong, H. (2015). Does a college education cause better health and health behaviours? Applied

AC C

EP T

ED

MA

NU

SC

RI

PT

Economics, 47(7), 639–653. doi:10.1080/00036846.2014.978074

ACCEPTED MANUSCRIPT Table 1 Descriptive statistics of main variables. Full

Smoking

Drinking

Binge-drinking

sample All

Yes

No

Yes

No

Yes

No

0.653

0.264

(0.476)

(0.441)

1.000

0.278

(0.000)

(0.448)

Dependent

0.304

0.585

(0.460)

(0.493)

0.353

0.679

0.210

(0.478)

(0.467)

(0.408)

0.104

0.223

0.052

0.295

0.000

ED

drinking

MA

(0/1)

Binge

(0.358)

NU

Drinking

SC

(0/1)

EP T

(0/1)

(0.416)

(0.222)

(0.456)

(0.000)

0.045

0.063

0.037

0.051

0.042

0.060

0.043

(0.207)

(0.243)

(0.189)

(0.220)

(0.199)

(0.238)

(0.203)

0.601

0.637

0.585

0.599

0.601

0.607

0.600

(0.490)

(0.481)

(0.493)

(0.490)

(0.490)

(0.489)

(0.490)

0.119

0.130

0.113

0.183

0.084

0.209

0.108

Parents

consumption

behavior*

AC C

(0.305)

Smoker

0.151

RI

Smoking

PT

Variables

Mother (0/1)

Smoker Father (0/1)

Drinker

ACCEPTED MANUSCRIPT

Drinker

(0.323)

(0.337)

(0.317)

(0.386)

(0.277)

(0.407)

(0.310)

0.633

0.637

0.632

0.717

0.588

0.787

0.615

(0.482)

(0.481)

(0.482)

(0.451)

(0.492)

(0.410)

(0.487)

0.045

0.055

0.040

0.076

0.028

0.169

0.030

(0.206)

(0.227)

(0.196)

(0.264)

(0.164)

(0.375)

(0.171)

0.248

0.265

0.241

0.228

0.575

0.210

(0.432)

(0.441)

(0.428)

(0.451)

(0.420)

(0.495)

(0.407)

9.637

9.229

MA

Mother (0/1)

9.904

9.491

9.734

9.625

(4.226)

(3.782)

(4.394)

(4.167)

(4.251)

(4.126)

(4.237)

Binge

PT

Father (0/1)

Year

of

variables Law (0/1)

Provisions

0.596

0.465

0.653

0.516

0.639

0.546

0.602

(0.491)

(0.499)

(0.476)

(0.500)

(0.480)

(0.498)

(0.490)

0.473

0.356

0.524

0.396

0.515

0.421

0.479

(0.499)

(0.479)

(0.499)

(0.489)

(0.500)

(0.494)

(0.500)

0.654

0.989

0.508

0.909

0.515

0.910

0.624

AC C

Instrumental

EP T

education

9.815

ED

Education

0.285

NU

(0/1)

SC

Binge Father

RI

Mother (0/1)

(0/1)

Control variables Male (0/1)

ACCEPTED MANUSCRIPT

Age

Age squared

(0.476)

(0.105)

(0.500)

(0.288)

(0.500)

(0.286)

(0.484)

24.326

27.163

23.087

26.868

22.940

28.003

23.899

(7.159)

(7.266)

(6.748)

(7.463)

(6.587)

(7.640)

(6.977)

643.013

790.634

578.536

777.564

569.646

842.465

619.859

(424.989) (454.328) (394.583) (468.022) (380.042) (488.733) (410.758) 1976.204

1973.724

1976.075

1973.831

1975.409

(8.059)

(8.506)

(7.661)

(8.595)

(7.625)

(8.386)

(8.004)

0.310

0.498

0.228

0.471

0.222

0.511

0.287

(0.463)

(0.500)

(0.420)

(0.499)

(0.416)

(0.500)

(0.452)

0.300

0.292

0.304

0.325

0.286

0.302

0.300

(0.458)

(0.455)

(0.460)

(0.468)

(0.452)

(0.459)

(0.458)

9.404

9.501

9.362

9.583

9.307

9.760

9.363

(1.161)

(1.173)

(1.153)

(1.165)

(1.147)

(1.089)

(1.162)

0.849

0.693

0.829

0.692

0.820

0.731

(0.439)

(0.359)

(0.461)

(0.377)

(0.462)

(0.384)

(0.444)

51.468

54.362

50.203

54.176

49.991

54.906

51.069

(8.682)

(8.714)

(8.359)

(8.881)

(8.203)

(9.123)

(8.540)

53.552

56.548

52.243

56.234

52.089

56.749

53.180

(9.098)

(9.033)

(8.813)

(9.206)

(8.697)

(9.452)

(8.983)

4.471

3.868

4.734

4.318

4.488

4.379

4.520

EP T

of income

Presently

0.740

(0/1)

AC C

working

Age Mother

Age Father

Education

MA

Logarithm

ED

Urban (0/1)

NU

(0/1)

PT

1973.049

RI

Married

1975.245

SC

Year of birth

ACCEPTED MANUSCRIPT

(4.380)

(4.125)

(4.462)

(4.376)

(4.381)

(4.381)

(4.380)

6.724

6.156

6.972

6.633

6.734

6.670

6.753

(4.420)

(4.339)

(4.432)

(4.481)

(4.413)

(4.573)

(4.334)

Yr1991

0.148

0.138

0.153

0.127

0.160

RI

0.111

0.153

Yr1993

0.199

0.164

0.214

0.169

0.190

0.200

Yr1997

0.114

0.115

0.114

SC

0.215

0.112

0.115

0.172

0.107

Yr2000

0.096

0.109

0.090

0.104

0.091

0.139

0.091

Yr2004

0.085

0.095

0.081

0.096

0.079

0.110

0.082

Yr2006

0.092

0.110

MA

Mother

0.084

0.118

0.077

0.116

0.089

Yr2009

0.105

0.116

0.101

0.132

0.091

0.132

0.102

Yr2011

0.088

0.116

0.079

0.111

0.076

0.106

0.086

Education Father

PT

Time

controls

0.069

0.076

0.066

0.068

0.070

0.076

0.069

Heilongjiang

0.050

0.041

0.054

0.043

0.054

0.056

0.049

Jiangsu

0.124

0.117

0.127

0.130

0.120

0.120

0.124

Shandong

0.107

0.076

0.120

0.094

0.114

0.123

0.105

Henan

0.119

0.134

0.112

0.139

0.108

0.108

0.120

Hubei

0.100

0.084

0.108

0.103

0.099

0.136

0.096

Hunan

0.109

0.101

0.112

0.074

0.128

0.071

0.113

AC C

Liaoning

ED

EP T

Regional

NU

controls

ACCEPTED MANUSCRIPT 0.151

0.165

0.145

0.179

0.136

0.133

0.153

Guizhou

0.140

0.174

0.125

0.134

0.143

0.146

0.139

Beijing

0.008

0.009

0.007

0.010

0.007

0.015

0.007

Shanghai

0.017

0.015

0.018

0.017

0.017

0.006

0.019

Chongqing

0.006

0.008

0.005

0.009

0.005

0.010

0.006

N

11316

3440

7876

3993

7323

1177

10139

PT

Guangxi

RI

Notes: The standard deviation is given in parentheses. Law refers to whether or not the

SC

individual is affected by the Nine-Years Compulsory Education Law; Provisions refers to whether or not the individual is affected by the Provisions on the Prohibition of Using Child

NU

Labor. In total, we have 11,316 observations. *Parent’s consumption be havior can be interpreted as follows: for example, among children who are smoking (Smoking=Yes), 6.3% of their

MA

mothers are also smoking. Among children who are not smoking (Smoking=No), 3.7% of their

ED

mothers are smoking.

EP T

Source: Authors’ calculations based on CHNS samples. Table 2 Probit estimations for smoking, drinking, and binge drinking. Smoking (2)

(3)

AC C

(1)

Mar

(5)

(6)

Binge Drinking (7)

Mar

(9)

(10)

(11)

Mar

gina

gina

gina

l

l

l

Smo

0.39

0.39

0.39

0.09

ker

4**

8**

7**

5**

*

*

*

*

(0.0

(0.0

(0.0

(0.0

Moth

Drinking

er

ACCEPTED MANUSCRIPT 9)

9)

9)

2)

Smo

0.26

0.26

0.26

0.06

ker

4**

5**

7**

4**

*

*

*

*

(0.0

(0.0

(0.0

(0.0

4)

4)

4)

1)

Fathe

0.679

0.677

hol

***

***

er

0.512

ED

Alco

Bing e Moth

AC C

r

***

EP T

hol Fathe

(0.05)

MA

(0.05)

(0.04)

***

NU

Moth

0.686

0.19

SC

Alco

RI

PT

r

(0.05)

3** *

(0.0 1)

0.509

0.513

0.14

***

***

4** *

(0.04)

(0.04)

(0.0 1) 0.87 8**

0.874 0.874 ***

***

*

0.12 2** *

er (0.0

(0.07

(0.07

(0.0

7)

)

)

1)

ACCEPTED MANUSCRIPT Bing

0.87

e

5**

Fathe

0.855 0.854 ***

***

*

0.11 9** *

r (0.05

(0.0

5)

)

)

1)

2.24

2.24

0.54

1.236

1.238

1.243

0.34

4**

8**

9**

0**

***

***

***

RI

PT

(0.05

2.24

*

*

*

*

(0.0

(0.0

(0.0

(0.0

9)

9)

9)

2)

0.16

0.16

0.14

0.00

4**

2**

2**

2

*

*

*

(0.0

(0.0

(0.0

(0.0

5)

5)

5)

1)

Age

-

-

squar

0.00

0.00

ed

3**

3**

3**

*

*

*

(0.0

(0.0

(0.0

0)

0)

0)

-

-

-

-

-

-

-

0.05

0.05

0.05

0.01

0.106

0.106

4

4

1

2

***

***

Year of birth

0.033

MA

0.036

(0.04)

(0.04)

9**

SC

(0.05)

(0.05)

NU

(0.05)

ED

EP T

Age

0.013

(0.04)

0.80 0**

0.799 0.797 ***

***

0.11 1**

*

*

(0.0

(0.0

(0.06

(0.06

(0.0

1)

6)

)

)

1)

-

0.17

0.01

5**

0.00

8*

*

9

(0.0

(0.0

(0.05

(0.05

(0.0

1)

3)

)

)

1)

-

-

-

-

-

-

-

0.00

0.002

0.002

0.002

0.00

***

***

***

2**

AC C

Male

(0.0

*

0.015 0.011

-

0.002 0.001 ***

***

(0.0

(0.00

(0.00

0)

)

)

-

0.04

-

-

0.101

0.02

7

***

8**

* (0.00)

(0.00)

(0.00)

0.112 0.111 **

**

0.01 5**

ACCEPTED MANUSCRIPT * (0.0

(0.0

(0.0

(0.0

5)

5)

5)

1)

Marr

0.16

0.16

0.14

0.03

0.216

0.210

ied

8**

1**

7**

5**

***

***

*

*

*

*

(0.0

(0.0

(0.0

(0.0

4)

4)

4)

1)

0.04

0.02

0.06

0.01

0.117

0.107

2

9

4

5

***

**

(0.0

(0.0

(0.0

5)

5)

5)

1)

Age

-

-

-

-

Moth

0.00

0.00

0.00

0.00

8

8

8

2

(0.0

(0.0

1)

1)

Age

-

Fathe

EP T (0.0

(0.0

1)

0)

AC C

er

-

0.00

0.00

0.00

0.00

0

0

1

1

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Educ

-

-

-

-

ation

0.01

0.01

0.01

0.00

r

(0.0

(0.0

(0.04

(0.04

(0.0

1)

3)

)

)

1)

0.201

0.05

0.08

***

7**

4*

0.073 0.070

0.01 0

(0.04)

(0.04)

(0.0

(0.0

RI

(0.04)

PT

*

(0.05)

0.005

SC

1)

(0.04)

0.005

(0.01)

0.001

(0.01)

0.002

5)

0.138

0.03

0.09

***

9**

8*

(0.05)

0.004

ED

(0.0

(0.03)

NU

n

(0.03)

MA

Urba

(0.03)

(0.01)

0.002

(0.05

(0.05

(0.0

)

)

1)

0.086 0.092 *

3*

* (0.0

(0.0

(0.05

(0.05

(0.0

1)

5)

)

)

1)

0.00

0.00

1

6

(0.0

(0.0

(0.01

(0.01

(0.0

0)

1)

)

)

0)

0.00

-

-

-

-

1

0.00

0.006 0.006

0.004

(0.01)

0.003

(0.01)

0.005

0.00 1

0.006 0.006

6 (0.01)

0.01

0.00 1

(0.0

(0.0

(0.01

(0.01

(0.0

0)

1)

)

)

0)

0.00

-

-

-

-

1

0.00

0.002 0.002

0.00

ACCEPTED MANUSCRIPT Moth

1*

2**

1*

3*

2

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Educ

-

-

-

-

ation

0.01

0.01

0.01

0.00

Fathe

1*

2**

2**

3**

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Loga

0.05

0.03

0.00

0.045

rithm

2**

1

7

**

(0.0

(0.0

(0.0

2)

1)

0.29

0.07

0.266

0.07

3**

0**

***

5**

*

*

(0.0

(0.0

5)

1)

0

er (0.01)

(0.01

(0.01

(0.0

0)

1)

)

)

0)

-

-

-

-

-

-

0.001

0.000

0.00

0.00

0

2

2)

AC C

Prese

SC

(0.01)

(0.01)

NU

EP T

me

ED

Inco

(0.01)

0.026

MA

of

Wor

(0.0

-

r

ntly

(0.0

0.004 0.004

0.00 1

RI

0.001

(0.01)

PT

(0.01)

(0.02)

(0.02)

(0.0

(0.0

(0.01

(0.01

(0.0

0)

1)

)

)

0)

0.00

0.058 0.052

0.00

7

**

**

7**

(0.0

(0.03

(0.03

(0.0

1)

)

)

0)

0.066

0.00 9

*

king

Cons

103.

102.

96.0

tant

218

459

49

(0.04)

206.2

205.3

197.1

15*** 76*** 03***

(0.0

(0.05

(0.0

1)

)

1)

98.5

217.3 214.7 75**

96**

ACCEPTED MANUSCRIPT 67 (94.

(95.

(95.

(68.4

(68.6

(67.8

(63.

(89.2

(89.1

93)

43)

54)

0)

1)

9)

13)

4)

1)

113

113

113

113

113

113

1131

1131

113

16

16

16

16

16

16

6

6

16

129

130

136

1748.

1772.

1853.

6.64

5.96

0.54

622

245

332

6

1

3

Pseu

0.30

0.30

0.31

do R2

6

7

1

PT

115

RI

8.67

0.231

0.232

SC

Chi2

11316 11316 11316

0.235

1109. 1112. 851

456

0

0.23

0.236 0.236

1

NU

N

* p<0.10, ** p<0.05, *** p<0.010

MA

Notes: Robust standard errors (clustered on community id) are in parentheses. Regional controls include dummies for Chongqing (reference), Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu,

ED

Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004, 2006, 2009, 2011.

EP T

Source: Authors’ estimations based on CHNS samples.

behaviors.

AC C

Table 3 Probit estimations of the impact of education on individuals’ unhealthy consumption

Smoking

(1)

(2)

(3)

Drinking Mar

(5)

(6)

Binge Drinking (7)

Mar

(9)

(10)

(11)

Mar

gina

gina

gina

l

l

l

Year

-

-

-

-

0.010

0.009

0.009

0.00

-

-

-

-

sedu

0.02

0.02

0.02

0.00

**

*

**

3**

0.00

0.00

0.00

0.00

ACCEPTED MANUSCRIPT 8**

8**

7**

*

*

*

*

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Smo

0.38

0.38

0.38

0.09

ker

1**

5**

4**

2**

*

*

*

*

(0.0

(0.0

(0.0

(0.0

9)

9)

9)

2)

Smo

0.25

0.25

0.26

0.06

ker

8**

8**

0**

2**

Fath

*

*

*

*

(0.0

(0.0

(0.0

(0.0

4)

4)

4)

1)

Moth

(0.00) (0.00) (0.00)

er

EP T

AC C

Moth

6

1

(0.0

(0.01

(0.01

(0.01

(0.0

0)

)

)

)

0)

NU MA

ED

er

hol

6

SC

er

Alco

5

PT

6**

RI

c

0.679

0.678

0.686

0.19

***

***

***

3** *

(0.05) (0.05) (0.05)

(0.0 1)

Alco

0.512

0.510

0.514

0.14

hol

***

***

***

4**

Fath

*

ACCEPTED MANUSCRIPT er (0.04) (0.04) (0.04)

(0.0 1)

Bing

0.87

0.87

0.87

0.12

e

5***

2***

2***

2**

SC

RI

er

NU

Bing e

Age

4**

9**

*

*

(0.07

(0.07

(0.0

)

)

)

1)

0.85

0.85

0.85

0.12

7***

5***

5***

2**

(0.05

(0.05

(0.05

(0.0

)

)

)

1)

2.26

0.54

1.238

1.239

1.244

0.34

0.79

0.79

0.79

0.11

0**

0**

***

***

***

9**

7***

7***

5***

1**

*

*

EP T

2.25

AC C

2.25

ED

er

(0.07

*

MA

Fath

Male

*

PT

Moth

(0.0

(0.0

(0.0

(0.0

9)

9)

9)

2)

0.16

0.16

0.14

0.00

7**

4**

5**

2

*

*

*

(0.0

(0.0

(0.0

(0.0

5)

5)

5)

1)

* (0.05) (0.05) (0.05)

0.034

0.032

0.012

*

(0.0

(0.06

(0.06

(0.06

(0.0

1)

)

)

)

1)

-

0.01

0.01

0.01

0.01

0.01

7

5

2

2**

8* (0.04) (0.04) (0.04)

*

(0.0

(0.05

(0.05

(0.05

(0.0

1)

)

)

)

0)

ACCEPTED MANUSCRIPT Age

-

-

-

-

-

-

-

-

-

squar

0.00

0.00

0.00

0.002

0.002

0.002

0.00

0.00

0.00

ed

3**

3**

3**

***

***

***

2***

2***

1***

*

*

*

(0.0

(0.0

(0.0

(0.00

(0.00

(0.00

0)

0)

0)

)

)

)

-

-

-

-

-

-

-

-

-

-

0.00

0.05

0.05

0.05

0.01

0.105

0.105

0.101

0.11

0.11

0.11

6

6

6

2

3

***

***

3**

2**

1**

(0.0

(0.04

(0.04

(0.04

(0.0

1)

)

)

)

0)

birth

PT

RI

of

-

0.02

SC

Year

(0.00) (0.00) (0.00)

***

8**

NU

*

(0.0

(0.0

(0.0

5)

5)

5)

1)

Marr

0.16

0.15

0.14

0.03

0.218

0.213

0.205

0.05

0.08

0.07

0.06

0.01

ied

7**

8**

5**

5**

***

***

***

7**

2*

2

9

0

*

*

*

*

(0.0

(0.0

(0.0

(0.0

(0.0

(0.05

(0.05

(0.05

(0.0

4)

4)

4)

1)

1)

)

)

)

1)

0.07

0.06

0.09

0.02

0.105

0.096

0.127

0.03

0.10

0.09

0.09

0.01

3**

**

**

***

6**

6**

2*

8*

3*

(0.0

(0.05

(0.05

(0.05

(0.0

1)

)

)

)

1)

0.00

0.00

0.00

0.00

0.00

1

6

6

6

1

n

7*

2

ED

EP T

AC C

Urba

7**

*

(0.04) (0.04) (0.04)

*

(0.0

(0.0

(0.0

(0.0

5)

5)

5)

1)

Age

-

-

-

-

Moth

0.00

0.00

0.00

0.00

7

7

8

2

er

(0.03) (0.03) (0.03)

MA

(0.0

(0.05) (0.05) (0.05)

0.005

0.005

0.004

ACCEPTED MANUSCRIPT

1)

1)

1)

0)

Age

-

-

-

-

Fath

0.00

0.00

0.00

0.00

1

1

0

0

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Educ

-

-

-

-

ation

0.00

0.00

0.00

0.00

Moth

8

9

7

2

(0.0

(0.0

(0.0

(0.0

1)

1)

1)

0)

Educ

-

-

-

-

ation

0.00

0.00

0.00

0.00

Fath

5

7

6

2

(0.0

(0.0

(0.0

(0.0

er

(0.01) (0.01) (0.01)

0.002

Loga rithm of

0.003

0.002

0.004

(0.01

(0.01

(0.01

(0.0

0)

)

)

)

0)

0.00

-

-

-

-

1

0.00

0.00

0.00

0.00

6

6

6

1

(0.0

(0.01

(0.01

(0.01

(0.0

0)

)

)

)

0)

-

-

-

-

0.00

0.00

0.00

0.00

0

1

1

0

(0.0

(0.01

(0.01

(0.01

(0.0

0)

)

)

)

0)

0.00 1

MA

(0.01) (0.01) (0.01)

-

ED

EP T

AC C

1)

0.002

(0.01) (0.01) (0.01)

er

er

0.002

(0.0

PT

(0.0

RI

(0.0

SC

(0.0

NU

(0.0

0.001

-

-

-

-

-

-

-

0.002

0.002

0.00

0.00

0.00

0.00

0.00

1

2

3

3

0

(0.0

(0.01

(0.01

(0.01

(0.0

0)

)

)

)

0)

0.00

0.06

0.05

0.00

6

1**

6**

9**

(0.0

(0.03

(0.03

(0.0

(0.01) (0.01) (0.01)

1)

1)

0)

0.06

0.04

0.01

0.040

7**

6**

1**

**

(0.0

(0.0

0.021

*

Inco me (0.0

(0.02) (0.02)

ACCEPTED MANUSCRIPT 2)

2)

1)

1)

Prese

0.29

0.06

0.267

ntly

0**

9**

***

Wor

*

*

(0.0

(0.0

5)

1)

)

)

0)

0.07

0.06

0.00

5**

6

9

(0.0

(0.05

(0.0

1)

)

1)

*

105.

98.9

205.1

204.5

196.2

219.

217.

215.

tant

046

318

81

07**

11**

16**

759*

172*

*

*

*

*

*

*

(94.

(95.

(95.

(68.5

(68.7

(68.0

(89.0

(89.1

(89.0

61)

21)

35)

6)

4)

1)

3)

9)

5)

113

113

113

113

1131

1131

1131

113

1131

1131

1131

113

16

16

16

16

6

6

6

16

6

6

6

16

130

132

141

1781.

1798.

1869.

1127

1109

1111

8.73

5.43

5.72

890

032

748

.347

.147

.702

0

4

Pseu

0.30

0.31

0.232

0.232

0.236

0.23

0.23

0.23

do R2

9

0

5

6

6

Chi2

MA

5

0.31

AC C

N

NU

844*

EP T

SC

106.

ED

Cons

RI

(0.04)

PT

king

4

* p<0.10, ** p<0.05, *** p<0.010 Notes: Robust standard errors (clustered on community id) are in parentheses. Regional controls include dummies for Chongqing (reference), Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004, 2006, 2009, 2011. Source: Autho rs’ estimations based on CHNS samples.

ACCEPTED MANUSCRIPT Table 4 OLS estimations of the impact of institutional changes on individuals’ educational attainment.

0.660*

0.652*

0.648*

**

**

**

(0.15)

(0.17)

(0.17)

Provisio

(4)

Year of birth

Married

Urban

(9)

0.322*

0.308*

0.303*

(0.18)

(0.17)

(0.17)

1.042*

1.056*

1.060*

**

**

**

1.163*

**

**

(0.17)

(0.17)

(0.17)

(0.18)

(0.18)

(0.18)

1.165*

-0.014

-0.016

-0.030

-0.013

-0.015

**

-0.015

-0.032

(0.12)

(0.10)

(0.10)

(0.10)

(0.10)

(0.10)

(0.10)

(0.10)

(0.10)

0.134

0.101

0.125

0.156

0.124

0.149

0.164

0.131

0.156

(0.13)

(0.12)

(0.12)

(0.12)

(0.12)

(0.12)

(0.12)

(0.12)

(0.12)

-

-

-

-

-

-

-

-

-

0.004*

0.004*

0.004*

0.005*

0.004*

0.005*

0.005*

0.004*

0.005*

**

**

**

**

**

**

**

**

**

MA

NU

-0.013

ED

squared

(8)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

(0.00)

-0.142

EP T

Age

(7)

-0.031

-0.137

-0.143

-0.159

-0.155

-0.162

-0.170

-0.165

-0.172

(0.12)

(0.11)

(0.11)

(0.11)

(0.11)

(0.11)

(0.11)

(0.11)

(0.11)

-

-

-

-

-

-

-

-

-

0.295*

0.360*

0.353*

0.303*

0.368*

0.362*

0.302*

0.367*

0.361*

*

**

**

*

**

**

*

**

**

(0.13)

(0.12)

(0.12)

(0.13)

(0.12)

(0.12)

(0.12)

(0.12)

(0.12)

1.179*

1.050*

1.012*

1.157*

1.026*

0.987*

1.153*

1.023*

0.983*

AC C

Age

(6)

1.153*

ns

Male

(5)

PT

(3)

RI

(2)

SC

Law

(1)

Age

**

**

**

**

**

**

**

**

**

(0.20)

(0.13)

(0.13)

(0.13)

(0.13)

(0.13)

(0.13)

(0.13)

(0.13)

0.004

0.003

0.004

0.002

0.001

0.002

0.002

0.002

0.003

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

0.003

0.006

0.005

0.007

0.010

0.009

0.006

0.010

0.009

(0.02)

(0.01)

(0.01)

(0.01)

(0.01)

RI

ACCEPTED MANUSCRIPT

(0.01)

(0.01)

(0.01)

0.170*

0.158*

0.156*

0.169*

0.157*

0.155*

0.170*

0.158*

0.156*

**

**

**

**

**

**

**

**

**

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

(0.02)

0.212*

0.196*

0.195*

0.212*

0.196*

0.195*

0.211*

0.195*

0.194*

**

**

**

**

**

**

**

**

**

(0.02)

(0.01)

(0.01)

(0.01)

(0.01)

(0.01)

(0.01)

(0.01)

Age

PT

Mother

Father

hm

of

Income

0.565*

0.585*

0.569*

0.589*

0.568*

0.588*

**

**

**

**

**

**

(0.05)

(0.05)

(0.05)

(0.05)

(0.05)

AC C

Logarit

(0.05)

Presentl y Workin

(0.01)

EP T

Father

ED

on

MA

Mother

Educati

SC

on

NU

Educati

(0.01)

-

-

-

0.311*

0.321*

0.318*

**

**

**

(0.11)

(0.11)

(0.11)

g

ACCEPTED MANUSCRIPT 268.83

281.10

317.19

304.51

317.72

338.88

325.30

338.05

4

8

6

4

1

5

0

7

5

(237.3

(217.7

(217.4

(222.2

(217.5

(217.1

(223.0

(218.3

(217.9

0)

2)

5)

7)

1)

7)

4)

0)

7)

N

11316

11316

11316

11316

11316

11316

11316

11316

11316

F

90.312

110.41

106.75

110.78

113.09

109.37

107.57

110.00

106.63

2

5

8

4

5

9

2

7

0.336

0.337

0.329

0.340

0.329

0.341

0.342

t

0.324

0.341

SC

R2

PT

283.11

RI

Constan

* p<0.10, ** p<0.05, *** p<0.010 Notes: Robust standard errors (clustered on community id and

NU

year of birth) are in parentheses. Regional controls include dummies for Chongqing (reference), Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi,

MA

Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004, 2006, 2009, 2011.

EP T

ED

Source: Authors’ estimations based on CHNS samples.

AC C

Table 5 Instrumental Probit estimations of smoking, drinking and binge drinking. Smoking

(1)

(2)

Drinking

(3)

Mar

(5)

(6)

Binge drinking (7)

ginal

Mar

(9)

(10)

(11)

ginal

Mar ginal

Years

-

-

-

-

-

-

-

-

-

-

-

-

educ

0.17

0.17

0.17

0.04

0.11

0.11

0.11

0.03

0.10

0.10

0.10

0.01

4***

5***

6***

1***

6***

6***

9***

4***

2**

2**

3**

6*

(0.04

(0.04 (0.04 (0.01

(0.04 (0.04

(0.04 (0.01

(0.05 (0.05

(0.05 (0.01

ACCEPTED MANUSCRIPT )

)

)

)

Smok

0.24

0.25

0.24

0.05

e

2**

0**

6**

8**

)

)

)

)

)

)

)

)

Mothe r

)

)

)

Smok

0.18

0.19

0.19

0.04

e

7***

1***

2***

5***

(0.05

(0.05 (0.05 (0.01

RI

)

PT

(0.10 (0.10 (0.02

SC

(0.10

)

)

)

Alcoh

ol

AC C

Alcoh

EP T

Mothe r

0.61

0.61

0.61

0.17

7***

5***

8***

4***

ED

ol

MA

)

NU

Father

(0.06 (0.06

(0.06 (0.02

)

)

)

)

0.47

0.46

0.46

0.13

0***

5***

6***

1***

Father (0.04 (0.04 )

)

(0.04 (0.01 )

)

Binge

0.80

0.79

0.79

0.12

Mothe

2***

8***

7***

1***

ACCEPTED MANUSCRIPT r (0.09 (0.09

(0.09 (0.01

)

)

)

)

Binge

0.81

0.81

0.81

0.12

Father

4***

1***

0***

3***

PT

(0.06 (0.06

No

Yes

No

Yes

Yes

Yes

No

ithm

Incom

MA

e

No

Worki

No

Yes

Yes

Yes

No

Yes

Yes

EP T

ng

AC C

First

Law

Yes

ED

ntly

stage

)

NU

of

Prese

Yes

)

RI

Yes

)

SC

Logar

(0.06 (0.01

0.41

0.40

0.39

0.29

0.28

0.26

0.32

0.30

0.30

6**

5**

5**

0*

0*

9*

0*

8*

3*

(0.17

(0.16 (0.16

(0.17 (0.17

(0.17

(0.17 (0.17

(0.17

)

)

)

)

)

)

)

)

)

Provis

0.96

0.97

0.98

1.05

1.07

1.07

1.04

1.06

1.06

ions

2***

8***

5***

8***

2***

8***

3***

0***

3***

(0.19

(0.19 (0.19

(0.17 (0.17

(0.17

(0.18 (0.18

(0.17

)

ACCEPTED MANUSCRIPT )

)

)

)

)

)

instru

38.0

108.

58.6

38.2

110.

59.8

ments

5

37

9

8

78

Wald

0.00

0.00

0.00

0.00

0.01

test of

3

7

0

1

0.00

0.00

0

)

)

)

38.3

62.2

62.5

3

5

2

7

0.00

0.03

0.03

0.03

8

1

1

0

F statist ics on the

PT

exclud

neity

ment robust

0.00

0.00

0.00

0.00

0.02

0.02

0.02

0

0

8

8

6

9

9

8

0.18

0.18

0.18

0.61

0.67

0.63

0.63

0.70

0.71

5

5

0

4

7

9

5

9

4

AC C

J-Test

EP T

instru

(CLR)

SC

ED

Weak

test

NU

MA

exoge

0

RI

ed

for overidentif ying restric

ACCEPTED MANUSCRIPT tion 0.00

0.00

0.00

0.01

0.01

0.00

0.03

0.03

0.03

Test

0

0

0

0

0

8

5

5

4

1131

1131

1131

1131

1131

1131

1131

1131

1131

1131

1131

1131

6

6

6

6

6

6

6

6

6

6

6

6

1941

1973

2053

2383

2370

2434

1357

1340

1341

.037

.654

.069

.524

.023

.074

.371

.424

.974

N

RI

Chi2

PT

K-J

SC

* p<0.10, ** p<0.05, *** p<0.010. Notes: Robust standard errors (clustered on community and year of birth) are in parentheses. Regional controls include dummies for Chongqing (reference),

NU

Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004,

MA

2006, 2009, 2011. All other controls include male, age, age squared, married, urban, year of birth, logarithm of income, presently working, and mother’s and father’s age and education. Weak

Newey’s two-step estimator.

ED

instrument robust test (CLR), J-Test for over-identifying restriction, and K-J Test are based on

EP T

Source: Authors’ estimations based on the CHNS sample.

AC C

Table 6 Instrumental Probit estimations of the impact of education on intergenerational persistence of smoking, drinking, and binge drinking. Smoking

Yearse duc

Drinking

Binge drinking

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

-

-

-

-

-

-

-0.066

-0.066

-0.067

0.138*

0.135*

0.137*

0.083*

0.082*

0.088*

**

**

**

(0.04)

(0.04)

(0.04)

(0.05)

(0.05)

(0.05)

(0.05)

(0.05)

(0.05)

ACCEPTED MANUSCRIPT Smoke

0.993

1.107

1.104

(1.50)

(1.51)

(1.49)

Smoke

0.673*

0.706*

0.704*

Father

*

*

*

(0.31)

(0.31)

(0.31)

-0.094

-0.107

-0.107

(0.18)

(0.18)

(0.18)

Smoke

-

-

-

Father*

0.051*

0.054*

0.053*

(0.03)

(0.03)

RI

Smoke

PT

Mother

SC

Mother *

NU

Yearse

ED

duc

EP T

Yearse

(0.03) 0.022

0.006

-0.017

(0.45)

(0.45)

(0.45)

Drink

0.912*

0.918*

0.890*

Father

**

**

**

(0.32)

(0.31)

(0.31)

0.061

0.063

0.066

Mother

Drink Mother

AC C

Drink

MA

duc

ACCEPTED MANUSCRIPT * Yearse duc

Drink

(0.05)

(0.05)

(0.05)

-0.046

-0.047

-0.044

(0.03)

(0.03)

PT

Father*

RI

Yearse

(0.03) 0.411

0.392

0.389

(0.60)

(0.60)

(0.60)

1.624*

1.635*

1.634*

**

**

**

(0.39)

(0.39)

(0.39)

0.042

0.043

0.044

(0.06)

(0.06)

(0.06)

-

-

-

Father*

0.084*

0.086*

0.086*

Yearse

*

*

*

NU

Binge

SC

duc

MA

Mother

Binge

* Yearse duc

Binge

duc

EP T

Mother

AC C

Binge

ED

Father

ACCEPTED MANUSCRIPT

hm

(0.04)

(0.04)

Yes

No

Yes

Yes

No

Yes

Yes

No

Yes

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

0.060

0.066

0.062

0.005

0.005

0.002

0.002

0.002

MA

Logarit

(0.04)

of

Income Presentl

PT

y

RI

Workin

Wald of

exogen eity Weak

ent

0.001

0.003

0.003

0.003

0.001

0.001

0.001

0.860

0.770

0.159

0.147

0.150

AC C

test

for

0.001

EP T

robust

J-Test

0.002

ED

instrum

(CLR)

0.004

NU

test

SC

g

overidentify ing restricti on

0.165

0.175

0.178

0.807

ACCEPTED MANUSCRIPT K-J

0.002

0.002

0.002

0.004

0.004

0.003

0.001

0.001

0.001

N

11316

11316

11316

11316

11316

11316

11316

11316

11316

Chi2

6118.5

6330.5

6135.8

2349.9

2339.8

2395.7

1301.0

1290.2

1291.0

85

33

37

32

21

35

71

73

13

Test

PT

* p<0.10, ** p<0.05, *** p<0.010. Notes: Robust standard errors (clustered on community and

RI

year of birth) are in parentheses. Regional controls include dummies for Chongqing (reference),

SC

Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004,

NU

2006, 2009, 2011. All other controls include male, age, age squared, married, urban, year of birth, logarithm of income, presently working, and mother’s and father’s age and education. Weak

MA

instrument robust test (CLR), J-Test for over-identifying restriction, and K-J Test are based on Newey’s two-step estimator.

ED

Source: Authors’ estimations based on CHNS samples.

EP T

Appendix:

Table A1 Instrumental Probit estimations of smoking, drinking, and binge drinking by residence.

AC C

Smoking

Rural

Drinking

Urban

Rural

Binge drinking Urban

Rural

Urban

Coef

Mar

Coef

Mar

Coef

Mar

Coef

Mar

Coef

Mar

Coef

Mar

f.

ginal

f.

ginal

f.

ginal

f.

ginal

f.

ginal

f.

ginal

Years

-

-

-

-

-

-

-

-

-

-

-

-

educ

0.17

0.04

0.18

0.04

0.12

0.03

0.10

0.02

0.12

0.01

0.05

0.00

8***

2***

4***

2***

6**

5**

0*

9*

4**

9*

4

8

(0.05

(0.01 (0.05 (0.01

(0.05 (0.02

(0.05 (0.02

(0.06 (0.01

(0.08 (0.01

ACCEPTED MANUSCRIPT )

)

)

)

Smok

0.24

0.05

0.21

0.04

e

9**

9**

2

8

)

)

)

)

)

)

)

)

Mothe r

)

)

)

Smok

0.21

0.05

0.12

0.02

e

6***

1***

8

9

(0.06

(0.01 (0.08 (0.02

RI

)

PT

(0.03 (0.16 (0.04

SC

(0.12

)

)

)

Alcoh

ol

AC C

Alcoh

EP T

Mothe r

0.47

0.13

0.45

0.13

8***

2***

5***

2***

ED

ol

MA

)

NU

Father

(0.05 (0.01

(0.08 (0.02

)

)

)

)

0.63

0.17

0.60

0.17

9***

7***

1***

4***

Father (0.07 (0.02 )

)

(0.11 (0.03 )

)

Binge

0.80

0.12

0.83

0.11

Mothe

5***

5***

5***

8***

ACCEPTED MANUSCRIPT r (0.09 (0.01

(0.17 (0.02

)

)

)

)

Binge

0.76

0.11

0.95

0.13

Father

3***

9***

8***

5***

PT

(0.08 (0.01

SC

First stage 0.31

8**

3

5*

(0.18

(0.34

(0.18

)

)

Provis

0.77

1.24

ions

7***

9***

the exclud

0.34

0.22

1

5**

5

(0.35

(0.17

(0.36

)

)

)

)

0.85

1.46

0.83

1.42

0***

9***

1***

2***

AC C

statist ics on

)

EP T

F

0.15

NU

0.44

MA

0.40

ED

Law

)

RI

)

(0.09 (0.01

ed instru

38.1

19.1

38.7

19.8

38.6

19.5

ments

8

0

5

5

6

6

Wald

0.02

0.00

0.05

0.13

0.08

0.40

test of

1

3

2

9

2

1

)

ACCEPTED MANUSCRIPT exoge neity Weak

0.01

0.00

instru

9

1 0.04

0.11

0.07

0.40

robust

7

9

9

4

0.14

0.83

0.70

3

8

2

0.05

0.15

0.09

0.45

7

2

4

9

PT

ment

RI

test

for

0.36

0.24

0

7

over0.92

MA

identif

NU

J-Test

SC

(CLR)

4

ying

tion 0.04

Test

1

0.00

EP T

K-J

ED

restric

1

7920

7920

3396

Chi2

1380

879.

1562

895.

1110

397.

.144

128

.303

994

.603

108

AC C

N

3396

7920

7920

3396

3396

7920

7920

3396

3396

* p<0.10, ** p<0.05, *** p<0.010. Notes: Robust standard errors (clustered on community and year of birth) are in parentheses. Regional controls include dummies for Chongqing (reference), Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include year dummies for 1991 (reference), 1993, 1997, 2000, 2004, 2006, 2009, 2011. All other controls include male, age, age squared, married, urban, year of birth,

ACCEPTED MANUSCRIPT logarithm of income, presently working, and mother’s and father’s age and education. Weak instrument robust test (CLR), J-Test for over-identifying restriction, and K-J Test are based on

AC C

EP T

ED

MA

NU

SC

RI

PT

Newey’s two-step estimator. Source: Authors’ estimations based on CHNS samples.

ACCEPTED MANUSCRIPT Appendix: Table A2 Instrumental Probit estimations of the impact of education o n intergenerational persistence of smoking, drinking, and binge drinking by residence. Smoking

Smoke Father

Smoke

Mother*

Urban

Rural

Urban

Rural

Urban

-0.081

-0.168**

-0.083

-0.013

-0.048

-0.012

(0.06)

(0.07)

(0.06)

(0.07)

-0.025

-0.245

(0.78)

(1.15)

0.895***

1.193**

(0.32)

(0.55)

0.026

0.033

(0.10)

(0.12)

-0.070*

Yearseduc

AC

Drink Mother

CE

(0.04)

Drink Father

Drink

Mother*

IP

CR

ED

Father*

-0.089*

PT

Smoke

M

Yearseduc

T

Rural

US

Smoke Mother

Binge drinking

AN

Yearseduc

Drinking

(0.05) 0.186

-0.206

(0.40)

(0.65)

0.956*** 1.407*** (0.31)

(0.52)

0.066

0.079

(0.04)

(0.06)

Yearseduc

(0.07)

(0.08)

ACCEPTED MANUSCRIPT Drink

Father*

-0.032

-0.071

(0.03)

(0.05)

Yearseduc

CR

Mother*

AN

US

Yearseduc

Binge

Father*

ED

M

Yearseduc

First stage

(0.69)

(0.84)

(0.65)

0.041

0.011

(0.07)

(0.08)

-0.102**

-0.121**

(0.04)

(0.06)

0.729**

0.370**

0.729**

0.370**

0.405**

(0.18)

(0.32)

(0.18)

(0.32)

(0.18)

(0.13)

CE

Provisions

(0.39)

0.370**

PT

Law

0.896*** 1.023***

0.896*** 1.023***

0.896*** 0.756***

(0.19)

(0.38)

(0.19)

(0.38)

(0.19)

(0.14)

of

0.090

0.055

0.068

0.088

0.022

0.094

instrument

0.056

0.062

0.058

0.068

0.026

0.083

0.825

0.128

0.999

0.270

0.148

0.795

AC

Wald

0.745

1.816*** 2.390***

IP

Binge Father

Binge

0.592

T

Binge Mother

test

exogeneity Weak

robust test (CLR) J-Test

for

identifying

over-

ACCEPTED MANUSCRIPT restriction K-J Test

0.036

0.128

0.071

0.104

0.035

0.109

N

7920

3396

7920

3396

7920

3396

1386.480

622.027

1939.507

801.355

1169.218

486.010

Chi2

T

* p<0.10, ** p<0.05, *** p<0.010.

IP

Notes: Robust standard errors (clustered on community and Yearbirth) are in parentheses.

CR

Regional controls include dummies for Chongqing (reference), Beijing, Liaoning, Heilongjiang, Shanghai, Jiangsu, Shandong, Henan, Hubei, Hunan, Guangxi, Guizhou. Time controls include

US

year dummies for 1991 (reference), 1993, 1997, 2000, 2004, 2006, 2009, 2011. All other controls

AN

include male, age, age squared, married, urban, year of birth, logarithm of income, presently working, and mother’s and father’s age and education. Weak instrument robust test (CLR), J-Test

M

for over-identifying restriction, and K-J Test are based on Newey’s two-step estimator.

AC

CE

PT

ED

Source: Authors’ estimations based on CHNS samples.

ACCEPTED MANUSCRIPT Highlights 

Analyzes unhealthy consumption behaviors and their intergenerational persistence in China.



Utilizes two institutional changes as instruments to address the endogeneity problem of

Concludes that there exists significantly positive intergenerational persistence of smoking,

IP



T

education in health equations.

Finds a significantly negative impact of education on smoking, drinking, and binge

US



CR

drinking, and binge drinking.

drinking; an additional year of education decreases the probability of smoking, binge

M

Observes specifically negative impacts of education on intergenerational persistence of smoking and binge drinking from the father, but no impact on intergenerational

CE

PT

ED

persistence from the mother.

AC



AN

drinking, and drinking by 4.1%, 3.4%, and 1.6%, respectively.