Gender Imbalance and Parental Substance Use in Rural China

Gender Imbalance and Parental Substance Use in Rural China
Xi Chen 1
Yale University
June 2014
Abstract
China and some other countries have experienced unbalanced sex ratios in the marriage
market, which triggers intense competition and pressure to get married. This paper utilizes two
household longitudinal datasets from rural China – a secondary national survey and a primary
regional survey – to examine parental substance use in response to skewed sex ratios. Sex
ratios are calculated using a 1‰ sample of the 2000 China Population Census. Strikingly,
paternal smoking and alcohol use are more intense for families with son living in communities
with higher sex ratios. In contrast, those with daughter do not demonstrate this pattern. Both
direct and indirect evidence suggests that coping with the marriage market pressure is a more
plausible pathway linking the observed skewed sex ratios and intense substance use, while we
do not find evidence supporting income effect and status seeking motives. Considering the
highly competitive marriage market in the coming decade and prevalent substance use that
generates lasting health impacts and large negative externalities to society, policies that correct
the skewed sex ratios could lead to substantial welfare gains.
Keywords: Skewed Sex Ratios, Marriage market, Substance Use, Smoking, Alcohol Use
JEL: J13, J22
1 Assistant Professor of Public Health (Health Policy) and Faculty of Arts and Sciences, Yale University,
[email protected]. Financial support from the National Science Fund of China (NSFC) (Approval numbers 70525003
and 70828002) and data collection support provided by IFPRI, China Academy of Agricultural Sciences, and
Guizhou University are acknowledged. I am grateful to Jeremy Berhman, Simon Chang, Jason Fletcher, Amanda
Kowalski, Jody Sindelar and Xiaobo Zhang for helpful comments and discussions. Seminar audiences at Stanford
University, IZA/CEPR 2013 workshop on labor economics, Yale University (Yale Social Network Working Group
Meeting 2013 and Yale Health Economics Workshop 2013) and Central University of Finance and Economics are
acknowledged. The views expressed herein and any remaining errors are the author's and do not represent any
official agency.
1
Electronic copy available at: http://ssrn.com/abstract=2270032
1. Introduction
Substance use, such as smoking and alcohol use, spreads widely among human beings.
Much substance use behavior is addictive, impose long term negative health impacts and even
a large number of premature deaths (Chatterji et al.,2004; Hu at al., 2006), impair labor market
performance (Balsa and French, 2010), and raise negative externalities to society through
accidents (Lovenheim and Steefel, 2011), increased use of public health care and addiction
treatment services (Balsa et al.,2009), child abuse (Markowitz and Grossman, 2000) and partner
abuse (Markowitz, 2000), violence and crime (Carpenter, 2007; Corman and Mocan, 2013).
Understanding and suppressing the determinants of substance use could lead to substantial
welfare gains.
This paper aims to examine how demographic factors, in particular more men than women
in the marriage market, affect substance use. The widely available ultrasound technology in
recent decades, the ingrained culture of son preference, together with one of the most radical
birth control policies in history lead to highly skewed sex ratios favoring women in China.
According to the China Population Census, sex ratio at birth (SRB) in China has been increasing
from 106.32 in 1975 to 118.06 in 2010. Compared to the developed eastern region, sex ratios
are even more unbalanced in the impoverished western and central China (Figure 1).
Meanwhile, the 2000 population census and a 1% 2005 population census indicate that rural
areas possess more skewed sex ratios compared to their urban counterparts (Ebenstein and
Sharygin, 2009). The scale of involuntarily single men is frightening. The number of excess
Chinese men under age 20 exceeded 32 million in 2005, which is greater than the entire male
population of Italy or Canada (Zhu, Lu, and Hesketh, 2009).
Skewed sex ratios widen the dispersion of marriage market rewards, while those of low
socioeconomic status and unable to get married have to bear grave consequences. Faced with
the pressure to get married, men tend to invest more in education, spend more on positional
goods, throw extravagant wedding parties, pay high brideprice and build fancy houses for
marriage, which occupy a great proportion of lifetime income (Foreign Policy, 2012). However,
almost none of these expenses is incurred by brides’ families (Appendix I). To make ends meet,
men have to work harder and take more risky jobs (Robson, 1996; Hopkins, 2011; Wei and
2
Electronic copy available at: http://ssrn.com/abstract=2270032
Zhang, 2011a). Meanwhile, the savings rate for grooms’ families peaks, in the year before the
wedding, while it is almost always lower for brides’ families (Wei and Zhang, 2011b).
Consequently, substance use can be promoted as a stress coping strategy among the
grooms’ families. In contrast, brighter prospects for marriage among females reduce the
occurrence of a number of substance use behaviors (Umberson, 1987). For example, nicotine is
a psychoactive (mood altering) drug, tobacco use might make the subjective effects of stress
(such as feelings of frustration, anger, and anxiety) less severe (Peski, 2013). Psychological
studies link increasing psychosocial strain with more excessive use of alcohol to self-medicate
anxiety disorders (Kushner et al., 1990; Shaw et al., 2011).
We study substance use in response to skewed sex ratios in rural China. To the best of our
knowledge, the contributions of this paper involve three aspects: this is among the first studies
to examine the link between sex ratios and substance use as well as its differential gender
responses; meanwhile, it is the first time that parental substance use responses to skewed sex
ratios are investigated; our primary data composed of a unique social network dataset and a
rich survey distinguishing tobacco products of different visibility enables us to explore the
potential mechanisms that promote smoking and alcohol drinking, involving income effect,
marriage market stress, and status signaling.
We focus on comparing families with first child being a son versus being a daughter. Much
evidence suggests that there are very few gender selections at the first birth parity in rural
China: no strict fertility control policy has been implemented for ethnic minorities in China
(Scharping, 2003); sex selections at the first birth are low in rural areas, where at least two
children are allowed; sex ratio for the 1st birth parity has been almost constant over time
(Ebenstein, 2009); mothers who are faced with different fertility policies at the 1st birth
demonstrate similar sex ratios at birth (Ebenstein, 2010); sex ratio at birth by parity shows
availability of ultrasound does not affect the 1st birth but higher parities (Chen, Li and Meng,
2010); Chinese parents generally prefer one daughter one son to two sons. Moreover, evidence
also suggests that endogenous fertility decisions on the first-born child may not be a concern in
our rural sample: regressing the number of children (or whether stopping at the second child)
on household minority status finds no significant results, suggesting that both ethnic minorities
3
Electronic copy available at: http://ssrn.com/abstract=2270032
and the major Han group are not subject to binding fertility control policy; summary statistics
indicate that sex ratios at the 1st birth parity are similar to the natural rate (Ebenstein, 2009).
China has more smokers and drink more alcohol than any other countries in the world. The
empirical investigations in this paper focus on paternal smoking and alcohol drinking behavior
for at least three reasons. First, smoking and alcohol drinking by men is deeply ingrained in
Chinese culture, while there have been strong social norms against women smoke or drink
alcohol. In China, men smoke at a much higher rate than women (53% vs. 2% in 2010) (The
Economist, 2012) and drink 13.4 times more than women (Cochrane et al., 2003). Second, men
are the major income earners in most Chinese families who naturally bear much of the financial
burden in preparation for children’s marriage. Third, substance use among sons tends to be
both motivated by the pressure to get married and suppressed due to their weak bargaining
power to smoke in the marriage market that favors women. Fortunately, the latter is irrelevant
when we investigate paternal substance use, which gives us a cleaner identification of the
marriage market pressure.
Our results suggest that fathers having son in the marriage market with more men than
women smoke more and drink more alcohol, especially for the poor. In contrast, those with
daughter do not demonstrate this pattern. Compared to income effect and status seeking
motives, coping with the marriage market pressure, from both own stress and stress spilled
over from peers in social networks, is a more plausible pathway linking the observed
unbalanced sex ratios and the intensified substance use. Some indirect evidence on reduced
paternal life satisfaction and happiness in communities with skewed sex ratios also suggests
that stress may matter. The placebo test using sex ratios of less relevant age cohorts and the
growing marginal effect as a son approaches the marriage age all suggest that our identified
effect is likely to be causal.
The rest of the paper is organized as follows. Section 2 reviews the investigated
consequences of skewed sex ratios in the literature. Section 3 introduces data collection for this
paper and documents basic trends from the data. Section 4 presents main results, placebo tests,
and robustness, and investigates the potential mechanisms. Finally, section 5 concludes.
4
2. The Consequences of Skewed Sex Ratios: A Brief Review
The most direct effect of skewed sex ratios is on the marriage market. Empirical results are
largely consistent with the theories that high sex ratios increase female bargaining power in the
marriage market. Angrist (2002) shows that high sex ratios have a large positive effect on the
likelihood of female marriage. Using the 2005 inter-census China national survey, Zhu, Lu and
Hesketh (2009) find large size of the surplus Chinese men in the marriage market. Using the
national census data, Ebenstein and Sharygin (2009) document that there were 22 million more
men than women in cohorts born between 1980 and 2000. Based on their simulations, about
10.4 percent of these additional men will fail to marry. However, Edlund (1996) documents that
dowry payments have been deteriorating despite a worsening shortage of brides in India.
Edlund argues that female scarcity may work against female marriage market status via
increasing the return on male human capital investment.
Skewed sex ratios may bear macroeconomic impacts through more intense marriage
market competition. Wei and Zhang (2011a) provide evidence that the gender imbalance in
mainland China may stimulate economic growth by inducing more entrepreneurship and hard
work. Utilizing the defeat of the Kuomintang Party in China in the late 1940s with more than
one million soldiers and civilians (mainly young males) retreated to Taiwan as a natural
experiment, Chang and Zhang (2012) find that young men were more likely to become
entrepreneurs, work longer hours, save more, and amass more assets. Wei and Zhang (2011b)
argue that Chinese parents with a son save competitively to improve their son’s relative
attractiveness for marriage as the sex ratios rise, which may account for half the recent
increase in the household savings rate. Wei, Zhang and Liu (2012) find that rising sex ratios
accounts for around two fifths of the rise in real urban housing prices in China due to the status
good feature of housing in the marriage market.
Skewed sex ratios may affect social security. Ebenstein and Sharygin (2009) discuss the
concern over China's ability to care for its elderly, with a particular focus on elderly males who
fail to marry. The current social security system provides little support for the childless elderly.
Evidence suggests that skewed sex ratios can account for many public security issues.
Analyzing data from 70 countries, Barber (2000) argues that low sex ratios (more females than
5
males) are likely to correspond to increased family conflict and aggression, and these societies
are predicted to have higher rates of violent crimes, such as homicide, rape and assaults. Barber
concludes that skewed sex ratios explain a substantial amount of the cross-national variance in
violent crimes. However, crime rates nearly double with the markedly rising sex ratios in China
in the last two decades. These seemingly contradictory trends can be reconciled with the fact
that many males are not able to get married with the highly skewed sex ratios. In other words,
marriage fails to act more effectively as a socializing force. Edlund et al. (2007) find that a 0.01
increase in the sex ratios raise violent and property crime rates by 3 percent, and the rise in sex
ratios may account for up to one-seventh of the overall rise in crime. Sex ratios raise security
concerns across country borders. Den Boer and Hudson (2004) argue that possibilities of
meaningful democracy and peaceful foreign policy might be diminished as a result of high sex
ratio induced internal instability and therefore altered security calculus for the state. Den Boer
and Hudson further predict that the high sex ratios in China and India, in particular, have
implications for the long-term security of these nations and the Asian region more broadly.
Skewed sex ratios also affect labor market. Ebenstein and Sharygin (2009) document the
unprecedented internal migration from the west to the east region in China partly motivated by
marriage purpose to seek better lives. This migration trend makes the unbalanced sex ratios
even more unbalanced in rural western China, as more females are married up to the more
developed eastern region. The migration trend may further determine labor market dynamics,
industry agglomeration and economic growth momentum. Angrist (2002) finds a large negative
effect of high sex ratios on female labor force participation. Higher sex ratios raise male
earnings and the incomes of parents with young children.
Unbalanced sex ratios have been found to affect public health. Ebenstein and Sharygin
(2009) argue that as males greatly outnumber females in the marriage market, more men are
likely to pay for sex. Consequently, prostitution and sexually transmitted infections, such as
HIV/AIDS, may become more prevalent. Hu and Goldman (1990) analyze marital-status-specific
death rates for a large number of developed countries. The results indicate that the excess
mortality of unmarried persons relative to the married has been generally increasing over the
past few decades. Moreover, as high sex ratios increase female bargaining power in the
6
marriage market and evidence suggests that divorced males have the highest death rates
among the unmarried groups, one may anticipate that the economic and physical well-being of
men who divorce or fail to marry will be of special concern. Further, some studies document
the relationship between marital status and psychological distress among the never married
and formerly married people, while some other studies examine the depressive consequences
of economic hardship, social isolation and parental responsibilities that unmarried people are
especially vulnerable and exposed to (Pearlin and Johnson, 1977).
Though the impact of gender imbalance on stress and stress-related illness is studied, very
few studies investigate its impact on stress coping behavior, especially its impact on behavior
and well-being of the parental generation. Little attention is paid towards gender differential
responses to unbalanced sex ratios, the focus of this paper. Compared to the existing literature,
different mechanisms, such as income effect, stress and status signaling, are further
distinguished. Meanwhile, we utilize social network data to innovatively gauge the marriage
market pressure imposed by other people in the networks.
3. Datasets
We utilize two household longitudinal datasets from rural China – a primary census-type
survey in Guizhou province during 2004-2009 and a secondary national survey in nine provinces
during 1991-2006 - to examine parental substance use responses to skewed sex ratios, such as
tobacco use and alcohol drinking. The national sample of China Health and Nutrition Survey
(CHNS) covers a wide range of nationally representative counties. We follow the literature to
merge the survey with sex ratios at the county level based on a 1‰ sample of the 2000 China
Population Census. Though the Guizhou survey only covers villages in one county, its census
feature allows us to accurately gauge sex ratios at the village level, which can be an appropriate
measure capturing localized marriage market competition in impoverished rural China. 2
2
First, the marriage market competition can be quite localized. More than half of the marriages in the Guizhou
survey are within villages (Appendix II). It has been well documented that grooms’ families build fancy houses and
spend lavishly in social events to signal to local matchmakers and improve the relative standing in the local
marriage market (Brown et al., 2011). More generally, Mangyo and Park (2011) find that village reference groups
are the most salient for residents living in close proximity in rural China. Second, county-level sex ratios calculated
using the Chinese population census may oversample residents living in county seat where sex ratios tend to be
7
For the purpose of this study, the rural sample of CHNS is employed.5 The CHNS covers
nine provinces in China that vary substantially in geography and economic development
(Appendix III). Each province is drawn following a multistage, random cluster process. Stratified
by income, a weighted sampling scheme was used to randomly select four counties in each
province. Villages and townships within the counties were selected randomly. There are about
18,000 individuals in some 4,200 rural households surveyed. We utilize the information on
cigarette consumption per day and liquor drinks per week in six waves of the survey during
1991-2006. Table 1A suggests that around half of the fathers smoke, and their smoking rate
increases slightly from 51 percent to 53 percent in 2000-2006. The average number of
cigarettes consumed per day increases from 5.66 to 6.32 between 2000 and 2006, while alcohol
consumption reduces from 286.5 gram to 233.5 gram in the same period. Annual cigarettes
consumption in packs is higher than the national average, which are 72.1 packs (in 2004) and
71.8 packs (in 2006). 6
The county level sex ratios merged from a 1‰ sample of the 2000 China Population
Census keep increasing from 109 males (per 100 females) to 117 males (per 100 females)
during 1991-2006 (Table 1B). Together with the worsening skewed sex ratios, their standard
deviations increase as well, suggesting that the gender gap among communities may widen.
The population census data suggests that the national average county level sex ratios at the 1st,
2nd and 3rd birth parities are 108.4, 143.2 and 152.9, respectively.
Figure 2 plots the positive relationship between sex ratios and parental tobacco and
alcohol consumption and distinguishes the relationship by child gender composition. No matter
for tobacco consumption or alcohol consumption, families with one son and no daughter show
positive association between sex ratios and tobacco and alcohol consumption, especially when
sex ratios are more biased favoring females. However, no clear association is found for families
with one daughter and no son.
less biased and therefore underestimate the actual sex ratios in rural areas. Third, the mountainous landform in
our surveyed area in Guizhou leads each village to function like an isolated community.
5
We thank the National Institute of Nutrition and Food Safety, China Center for Disease Control and Prevention;
the Carolina Population Center, University of North Carolina at Chapel Hill; the National Institutes of Health (NIH;
R01-HD30880, DK056350, and R01-HD38700); and the Fogarty International Center, NIH, for financial support for
the CHNS data collection and analysis files since 1989.
6
China Tobacco Yearbook (2007).
8
Figure 3 compares tobacco and alcohol consumption patterns among households along
the income distribution. We distinguish families with son and those with daughter, and families
living in high sex ratio counties and those living in low sex ratio counties. The two types of
counties are categorized by their sex ratios relative to the median level in our sample. The
overall increase in smoking and alcohol drinking with income may capture positive income
effect. The two left figures suggest that poor families with son living in high sex ratio counties
smoke more and drink more alcohol, while the intensity is much lower for their poor
counterparts living in counties with low sex ratios. This may suggest that these families are
especially stressful and are less capable to cope with the marriage market pressure. However,
in the right figures no similar pattern is found for families with daughter.
Moreover, families living in high sex ratio counties with son demonstrate significantly
higher smoking and alcohol drinking intensity (left figures of Figure 3) than those with daughter
(right figures of Figure 3), while a comparison between the two types of families in low sex ratio
counties generates no distinct pattern. The findings in Figure 3 consistently suggest that the
marriage market pressure can be explicitly distinguished from other factors, such as income
and demographic effect.
We also utilize a three-wave Guizhou survey administered in 2004-2009. Though the
survey has limited geographic scope (Appendix III), it covers all households and their members
in 26 randomly selected villages. The rich information on both manufactured tobacco products
and homemade tobacco products (Table 1A) helps us distinguish other potential motives from
status seeking in driving tobacco use and alcohol drinking. Meanwhile, the unique spontaneous
gift records collected during the Guizhou survey enable us to separate the direct marriage
market pressure from its indirect effect imposed by peers in the networks. The details of this
social network data are illustrated in Chen (2014). Compared to the national average tobacco
consumption of 281.08 yuan (in 2004) and 324.73 yuan (in 2006)7, our sampled individuals
spend more on tobacco. The reported more tobacco consumption in the Guizhou survey may
be originated from its inclusion of homemade tobacco products, so is its much higher smoking
rate than the reported value in the CHNS. Table 1B indicates that the village level sex ratios at
7
China Tobacco Yearbook (2007).
9
the 1st, 2nd and 3rd birth parities in our sample are 106.0, 119.7 and 138.9, respectively. Figure 2
and Figure 3 using the Guizhou survey demonstrate very similar patterns, so they are not
presented in the paper but are available upon request.
4. Empirical Results
4.1 Main Results
The first set of main results using the CHNS data is reported in Table 2. The first column for
each outcome variable uses sex ratios of 5-19 age cohorts. The second column through the
fourth column for each outcome variable adopt sex ratios of 5-9, 10-14, and 15-19 age cohorts,
respectively. Sex ratios of the 15-19 age cohort capture larger and more significant effect.
Because most sampled households have one or two children, most families in the 5-9, 10-14,
and 15-19 age cohorts do not overlap, the marginal effect of the first column (age cohort 5-19)
is close to the sum of the effects for the other three age cohorts.
To distinguish families with son from those with daughter and families with different
number of children, the first column for each outcome variable in Table 3 estimates,
=
yijt α1sexratio jt + β1son1stij + γ 1sexratio jt * son1stij + X ijt Γ + µi + ν t + eijt
(1)
where i denotes family; j represents county (for the CHNS data) or village (for the Guizhou
survey); t is year. sexratio jt is sex ratio at the county level (for the CHNS data) or the village
level (for the Guizhou survey), and son1stij is a dummy variable equals one when the first child
of the family is a son. yijt denotes four outcome variables, including expenditures on tobacco
use and alcohol drinking in the Guizhou survey and number of cigarettes one smoke per day
and grams of liquor drinks per week in the CHNS. Only outcomes that capture the intensive
margin of tobacco use and alcohol drinking are investigated since people tend to form the habit
of smoking or alcohol drinking in much younger age and are less likely to change this decision.
X ijt are covariates, including household income per capita, paternal year of education,
household head gender and age, marital status, share of the elderly, share of youth, household
size, whether suffer from major diseases, and ethnicity. ν t denotes year fixed effects, and µi
represents household fixed effects. Since our main interested variables, i.e., tobacco use and
10
alcohol drink, may reflect individual habit and local norms, our estimations utilizing withinhousehold variation in consumption should mitigate these concerns.
Restricting the sample to households with no more than two children, having a son first
does not affect substance use. However, the combination of having a son and living in a
community with more skewed sex ratios is associated with more smoking and alcohol drinking
among parents. Further, releasing the assumption that parental inclination to engage in such
behavior is identical regardless of the sex composition of children, we estimate the equation (2)
for nuclear families with a son (the second columns) and those with a daughter (the third
columns) in Table 3. Results suggest more substance use among nuclear families with a son,
while no such pattern is found for those with a daughter.
=
yijt α1sexratio jt + X ijt Γ + µi + ν t + eijt
(2)
The marriage market pressure is expected to exert a bigger impact on households with son
approaching marriage age. Using the CHNS, we estimate the equation (1) with various age
cohorts from 1-5 to 26-30 to gauge the heterogeneous effects and draw their marginal effects
in Figure 4. The growing marginal effects of the interaction term (sex ratio*first child being son)
suggests that marriage market pressure becomes more intensified as a son grows up. Though
the small sample size in the Guizhou survey prohibits us from testing the heterogeneous effects,
a separation into 1-11 and 12-19 age cohorts 8 suggests that the 12-19 cohorts demonstrate
much higher marginal effect than the 1-11 cohorts.
4.2 Robustness of the Main Findings
First, the number of children in a household might be an endogenous choice of parents.
Therefore, in Appendix IV we check Heckman two-step estimations that model parental fertility
choice to stop at one child. Following Wei and Zhang (2011), we use minority status of the
household, age of the first child and whether the first child disables or suffers from big diseases
as the exclusive variables in the selection equation of whether parents stop at the first child.
8
In rural Guizhou province, families with boy reaching 12 hold a coming-of-age ceremony, signaling to the
community that the boy grows up into the marriage age. Our age cohort classification is consistent with this norm.
11
The main equations for nuclear families with first child being son still maintain statistical
significance, while no such result is obtained for nuclear families with first child being daughter.
Second, Table 4 shows the result of a falsification test that replaces the sex ratios (age
cohort 5-19) best capturing the mating competition by sex ratios of less relevant age cohort
(30-40). The effects disappear and even change signs, indicating that the combination of having
unmarried son and living in an area with skewed sex ratio at marriage age, rather than
unobserved potential trend, promote parents’ substance use.
Third, all the estimations cluster standard errors at the village level. However, the results
are robust to cluster at the county level when the CHNS national sample is utilized. For our
Guizhou survey collected from 26 villages in one county, the Cameron-Gelbach-Miller (2011)
bootstrapping method is adopted to address the issue of small number of clusters, and
adjusted p-values are presented in the brackets in Table 3.
4.3 Potential Mechanisms
Do sex ratios motivate more substance use through income effect? Evidence suggests that
unbalanced sex ratios stimulate grooms’ families to work harder to earn more money (Wei and
Zhang, 2011). To test whether income effect dominates the pathway, in Table 5A we regress
per capita income on local sex ratios and compare the effects for families with different
demographic structures. When the sample includes all families, we do not find positive impact
of sex ratios on income. When interact sex ratio with first child being son and restrict the
sample to families with one or two children, we find positive but insignificant effect on income.
We further restrict the sample to families with one child. Results even suggest a negative,
though insignificant, effect of sex ratio on income for families with son (column 3) and a
positive but insignificant effect for families with daughter (column 4).
Meanwhile, if skewed sex ratios affect substance use through a positive income effect, we
should at least expect more substance use among richer families living in communities with
high sex ratios. Re-estimating the main results in Table 3 but interacting local sex ratios with
four income quartiles, Table 5B suggest that smoking and alcohol drinking are biased towards
the lower income quartiles, which echoes the pattern in Figure 3 that poorer households with
12
son living in high sex ratio communities tend to consume more tobacco and alcohol. These
pieces of evidence clearly do not support the story of income effect.
Since both cigarette smoking and alcohol drinking are positional (Heffetz, 2011), do poor
households consume as a means to signal and improve social status? Though the positional
feature of tobacco and alcohol consumption makes it difficult to fully distinguish marriage
market pressure from status seeking motive, the rich Guizhou survey data on different types of
tobacco consumption enables us to partially distinguish the two mechanisms. Relative to
packed cigarettes with publicly recognizable brands, tobacco pipe smoking generally has little
to do with status signaling. If status signaling is the dominant motive, we should find little
evidence on tobacco pipe smoking. However, results from Table 5C show families with a son
living in high sex ratio villages experience more tobacco pipe smoking. In other words, factors
other than status signaling should be in effect.
Coping with stress can be a key motive for intensified smoking and alcohol drinking (Peski,
2013). It is plausible that those who are less able to manage the marriage market pressure tend
to consume more such goods. Though stress is not directly measured in the CHNS data or the
Guizhou survey, we are able to indirectly test whether having a son and living in a community
with skewed sex ratios favoring females actually reduce parental life satisfaction and happiness.
Results are presented in Table 5D. For families with one or two children, having a son and
experiencing skewed sex ratios in the marriage market indeed predict lower life satisfaction and
happiness. This pattern is salient for families with son but not for families with daughter when
the sample is restricted to one child families.
Where is the marriage market pressure from? Besides one’s own pressure as a result of having a
son approaching the marriage age (the direct effect), friends and neighbors also having an unmarried
son may further intensify the marriage market pressure (the indirect effect). We estimate spatial
econometric models (Appendix V) to distinguish the direct effect from the indirect effect.
Results in Appendix V confirm the significance of the direct effect, while they also indicate that
the indirect effect spills over to this family and further promotes parental smoking and alcohol
drinking. Depending on the specific substance use behavior under investigation, the identified
13
indirect effect in the spatial estimations ranges from 24.7 percent to 33.8 percent of the
identified direct effect.
5. Conclusions
China and some other East Asian countries have experienced increasingly skewed sex
ratios. Utilizing two longitudinal datasets, the CHNS national sample and the Guizhou survey,
we find that the marriage market competition promotes substance use. Parents with son
consume more tobacco and alcohol, while those with daughter do not demonstrate this pattern.
Our results may underestimate the marriage market effect as rural China is subject to less
stringent family planning policy.
Due to the strong social norms in China that discourage women from substance use, we
limit our analysis to the male subsample. We find that paternal smoking and alcohol use are
more intense for families with son living in communities with higher sex ratios. In contrast,
those with daughter do not demonstrate this pattern. Both direct and indirect evidence
suggests that coping with the marriage market pressure, from both own stress and stress
spilled over from peers in social networks, is the most plausible pathway connecting the
observed skewed sex ratios and the intensified substance use, while there is no solid evidence
supporting income effect and status seeking motives.
Since sex ratio imbalance in China will probably become worse in the next decade,
investigating the marriage market pressure and behavior consequences could help design
effective policies that improve parental well-being through rebalancing skewed sex ratios.
Moreover, widespread substance use in the world imposes large negative health impacts,
affects functioning of the labor market, and brings various negative externalities to society.
Therefore, understanding and suppressing the determinants of substance use, such as skewed
sex ratios in this paper, could lead to substantial welfare gains.
14
References
Angrist, J. 2002. “How Do Sex Ratios Affect Marriage and Labor Markets? Evidence from
America’s Second Generation.” Quarterly Journal of Economics, 117 (3): 997-1038.
Balsa, A. I. and French, M. T. 2010. Alcohol Use and the Labor Market in Uruguay. Health
Economics, 19, 833-854.
Balsa, Ana I., French, Michael T., Maclean, Catherine and Norton, Edward C. 2009. From Pubs to
Scrubs: Alcohol Misuse and Health Care Use. Health Services Research 44, 1480-1503.
Barber, N. 2000. “The sex ratio as a predictor of cross-national variation in violent crime.” CrossCultural Research, 34, 264-282.
Cameron, A.C., J.B. Gelbach. and D. L. Miller. 2011. “Robust Inference with Multi-Way
Clustering.” Journal of Business and Economic Statistics 29(2): 238-249.
Carpenter, Christopher 2007. Heavy alcohol use and crime: Evidence from underage drunk
driving laws. Journal of Law & Economics, 50, 539-557.
Chang, S. and X. Zhang. 2012. "The Economic Consequences of Excess Men: Evidence from a
Natural Experiment in Taiwan." IFPRI discussion papers 1203.
Chatterji, Pinka, Dave, Dhaval, Kaestner, Robert and Markowitz, Sara. 2004. “Alcohol abuse and
suicide attempts among youth.” Econ Hum Biol 2, 159-80.
Chen, X. 2014. “Gift-giving and Network Structure in Rural China: Utilizing Long-term
Spontaneous Gift Records.” Forthcoming, PLoS ONE.
Cochrane, Johanne. H. Chen, K. Conigrave, and W. Hao. 2003. “Alcohol Use in China.” Alcohol
and Alcoholism 38 (6): 537-542.
Corman, Hope and Mocan, Naci H. 2013. Alcohol Consumption, Deterrence and Crime in New
York City. NBER Working Paper Series.
Den Boer, A. and V. Hudson. 2004. “The Security Threat of Asia’s Sex Ratios.” SAIA Review 24(2):
27-43.
Ebenstein, AY., and EJ. Sharygin. 2009. “The consequences of the ‘missing girls’ of China.” The
World Bank Economic Review 23 (3), 399-425.
Ebenstein, Avraham. 2010. “The Missing Girls of China and the Unintended Consequences of
the One Child Policy.” Journal of Human Resources. 87-115.
Edlund, L. 1996. “Dear Son – Expensive Daughter: Do Scarce Women Pay to Marry?” Essay Two
in The Marriage Market: How Do You Compare. Dissertation, Stockholm School of Economics.
Edlund, L., H. Li., J. Yi., J. Zhang. 2007. "Sex Ratios and Crime: Evidence from China’s One-Child
Policy." IZA Discussion Papers 3214.
Elhorst, J. 2012. "Dynamic spatial panels: models, methods, and inferences." Journal of
Geographical Systems 14(1): 5-28.
Foreign Policy. 2012. “Bachelor Padding.” September 28.
Heffetz, Ori. 2011. "A Test of Conspicuous Consumption: Visibility and Income Elasticities."
Review of Economics and Statistics 93.4: 1101-1117.
Hopkins, Ed. 2011. "Inequality and Risk-Taking Behaviour." ESE Discussion Papers 204,
Edinburgh School of Economics, University of Edinburgh.
T-W Hu, Z Mao, M Ong, E Tong, M Tao, H Jiang, K Hammond, K R Smith, J de Beyer, A Yurekli.
2006. “China at the crossroads: the economics of tobacco and health.” Tobacco Control
15(Suppl I):i37–i41.
15
Hu, Y and N. Goldman. 1990. “Mortality Differentials by Marital Status: An International
Comparison.” Demography. 27(2): 233-250.
Kushner, M. G., Sher, K. J. and Beitman, B. D. 1990. The relation between alcohol problems and
the anxiety disorders. Am J Psychiatry 147: 685-95.
Lovenheim, Micael F. and Steefel, Daniel P. 2011. “Do Blue Laws Save Lives? The Effect of
Sunday Alcohol Sales Bans on Fatal Vehicle Accidents.” Journal of Policy Analysis and
Management, 30, 798-820.
Mangyo, E. and A. Park. 2011. Relative deprivation and health: Which reference groups matter?
Journal of Human Resources 46(3): 459-481.
Markowitz, Sara 2000. “The price of alcohol, wife abuse, and husband abuse.” Southern
Economic Journal 67, 279-303.
Markowitz, Sara and Grossman, Michael 2000. “The effects of beer taxes on physical child
abuse.” Journal of Health Economics 19, 271-282.
Pearlin, L. and J. Johnson. 1977. "Marital Status, Life Strains, and Depression." American
Sociological Review 42:704-15.
Peski, M.F. 2013. “Stress and Smoking: Associations with Terrorism and Causal Impact.”
Contemporary Economic Policy. doi: 10.1111/coep.12021.
Robson, Arthur. 1996. “The Evolution of Attitudes to Risk: Lottery Tickets and Relative Wealth.”
Games and Economic Behavior 14, 190-207.
Scharping, Thomas. 2003. “Birth Control in China 1949–2000.” New York, NY: Routledge Curzon.
Umberson, Debra. 1987. "Family status and health behaviors: Social control as a dimension of
social integration." Journal of Health and Social Behavior, 28(3): 306-319.
Shaw, B. A., Agahi, N. and Krause, N. 2011. “Are changes in financial strain associated with
changes in alcohol use and smoking among older adults?” J Stud Alcohol Drugs 72: 917-25.
The Economist. 2012. “The Big Smoke.” March 21.
Wei, S. and X. Zhang. 2011a. "Sex Ratios, Entrepreneurship, and Economic Growth in the
People’s Republic of China." NBER Working Papers 16800.
Wei, S. and X. Zhang. 2011b. "The Competitive Saving Motive: Evidence from Rising Sex Ratios
and Savings Rates in China." Journal of Political Economy, 119(3): 511 - 564.
Wei, S., X. Zhang. and Y. Liu. 2012."Status Competition and Housing Prices." NBER Working
Papers 18000.
Zhu, W., L. Lu. and T. Hesketh. 2009. “China’s excess males, sex selective abortion, and one
child policy: analysis of data from 2005 national intercensus survey.” British Medical Journal,
2009 (338): b1211.
16
Figure 1 Sex ratios in the marriage market
Source: National Bureau of Statistics of China (2000)
Notes: Sex ratios are defined as number of males per female.
17
Figure 2 Sex Ratios and Tobacco & Alcohol Consumption, by Child Gender Composition
Alcohol
2
2
Alcohol consumption (# 50g per week)
3
4
5
Tobacco consumption (# cigarettes per day)
3
4
5
6
6
Tobacco
.5
1
1.5
Sex ratio
2
.5
2.5
1
1.5
Sex ratio
2
One son, no daughter
One son, no daughter
No son, one daughter
No son, one daughter
2.5
Source: CHNS household survey data (1991-2006).
Notes: Vertical axis denotes parents’ tobacco & alcohol consumption. This Figure only plots families with one child.
Sex ratios are defined as number of males per female.
18
With daughter
0
0
2
2
4
4
6
6
8
8
With son
2
4
6
Per capita income (log)
8
2
10
4
6
Per capita income (log)
8
10
8
10
With daughter
0
2
2
4
4
6
6
8
8
With son
0
Alcohol consumption (# 50g per week)
Tobacco consumption (# cigarettes per day)
Figure 3 Parental Tobacco and Alcohol Consumption by Income
(Counties with high sex ratio versus counties with low sex ratio, families with son versus families with daughter)
2
4
6
Per capita income (log)
High sex ratio
8
10
2
4
6
Per capita income (log)
High sex ratio
Low sex ratio
Source: CHNS household survey data (1991-2006).
19
Low sex ratio
Marginal effects of marriage market pressure
1
by age cohorts, first child being son
.8
26~30
26~30
16~20
17~21
.6
.4
1~5 2~6
1~5
2~6
10~14 12~16
13~17 15~19
9~13
14~18
8~12
7~11
16~20
15~19
11~15
12~16
17~21
5~96~10
13~17
5~9
18~22
3~7 4~8
25~29
19~23
18~22 20~24
20~24
19~23
11~15
.2
Percent increase in consumption per SD higher sex ratios
Figure 4 Marginal Effects of Marriage Market Pressure
23~27
21~25
24~28
23~27
22~26
21~25
14~18
10~14
6~107~11
8~129~13
24~28
22~26
3~7 4~8
0
10
20
Age cohorts (starting age ~ end age)
Ln(# cigarettes smokes per day)
Ln(# 50 gms liquor drinks per week)
Source: CHNS household survey data (1991-2006).
20
30
Table 1A Summary Statistics for Key Left Hand Side Variables
Mean
Guizhou sample
Dummy for parental smoking in 2004
Dummy for parental smoking in 2006
Dummy for parental smoking in 2009
Parental tobacco consumption expenses in 2004
Parental tobacco consumption expenses in 2006
Parental tobacco consumption expenses in 2009
Parental homemade tobacco consumption in 2004 (in 50g)
Parental homemade tobacco consumption in 2006 (in 50g)
Parental homemade tobacco consumption in 2009 (in 50g)
Parental alcohol consumption expenses in 2004
Parental alcohol consumption expenses in 2006
Parental alcohol consumption expenses in 2009
China Health and Nutrition Survey (CHNS) National Sample
Dummy for parental smoking in 1991-2006
Dummy for parental smoking in 2000
Dummy for parental smoking in 2004
Dummy for parental smoking in 2006
Parental tobacco consumption in 1991-2006 (# cigarettes per day)
Parental tobacco consumption in 2000 (# cigarettes per day)
Parental tobacco consumption in 2004 (# cigarettes per day)
Parental tobacco consumption in 2006 (# cigarettes per day)
Parental alcohol consumption in 1991-2006 (# 50g per week)
Parental alcohol consumption in 2000 (# 50g per week)
Parental alcohol consumption in 2004 (# 50g per week)
Parental alcohol consumption in 2006 (# 50g per week)
Life satisfaction (1=least satisfied, 5= most satisfied)
Happiness (1=least happy, 5=most happy)
Standard Deviation
0.66
0.65
0.71
265.39
387.86
536.44
30.96
27.15
26.82
281.89
423.77
587.85
0.35
0.36
0.39
293.81
540.21
709.43
34.28
37.82
35.47
327.48
613.60
764.55
0.52
0.51
0.49
0.53
5.32
5.66
5.38
6.32
4.54
5.73
5.84
4.67
3.97
3.17
0.50
0.50
0.50
0.50
7.44
7.53
8.02
8.25
10.40
11.54
12.65
10.01
0.85
0.71
Source: Author’s Guizhou household survey data (2004-2009); CHNS household survey data (1991-2006).
21
Table 1B Summary statistics for Key Right Hand Side Variables
Mean
Standard Deviation
Sex ratios inferred from a census-type survey in Guizhou
Sex ratio for the age cohort 5-19 average over 2004-2009
Sex ratio at first birth (# males per female)
Sex ratio at second birth (# males per female)
Sex ratio at third birth (# males per female)
1.16
1.06
1.20
1.39
0.08
0.14
0.15
0.15
Sex ratios inferred from a 1‰ sample of the 2000 China Population Census
Sex ratio for the age cohort 5-19 in 1991 (# males per female)
Sex ratio for the age cohort 5-19 in 1993 (# males per female)
Sex ratio for the age cohort 5-19 in 1997 (# males per female)
Sex ratio for the age cohort 5-19 in 2000 (# males per female)
Sex ratio for the age cohort 5-19 in 2004 (# males per female)
Sex ratio for the age cohort 5-19 in 2006 (# males per female)
Sex ratio at first birth (# males per female)
Sex ratio at second birth (# males per female)
Sex ratio at third birth (# males per female)
1.09
1.09
1.10
1.12
1.13
1.17
1.08
1.43
1.53
0.16
0.16
0.16
0.16
0.17
0.20
0.12
0.12
0.12
Source: Author’s Guizhou household survey data (2004-2009); CHNS Data (1991-2006); A 1‰ sample of the 2000 China Population Census.
Notes: The sex ratios for the age cohorts 5-19 in 1991, 1993, 1997, 2004, and 2006 are respectively inferred from the age cohorts 14-28, 12-26, 8-22, 1-15, and
0-13 in the 2000 population census. Sex ratios are defined as number of males per female.
22
Table 2 Baseline Results: Sex Ratios and Substance Use – Smoking and Alcohol Use
Sex ratio
Adjusted R2
AIC
N
5-19
0.541***
(0.128)
0.443
8310.47
21321
Ln(# Cigarettes Smokes Per Day)
5-9
10-14
0.090
0.115*
(0.060)
(0.061)
0.422
0.372
8282.45
8300.67
21321
21321
15-19
0.104**
(0.047)
0.328
8340.37
21321
Ln(# 50 Gms Liquor Drinks Per Week)
5-19
5-9
10-14
15-19
0.602***
0.100
0.131*
0.113**
(0.149)
(0.072)
(0.072)
(0.050)
0.287
0.298
0.273
0.269
9006.12
8970.62
8991.91
9034.38
20088
20088
20088
20088
Source: CHNS household survey data (1991-2006).
Notes: Sex ratios are measured at the county level using a 1‰ sample of the 2000 China Population Census data. Robust standard errors are in the brackets.
*, **, *** denote statistical significance at the 10%, 5% and 1% level, respectively. All regressions have a constant term that is not reported.
23
Table 3 Main Results: Sex Ratios, Family Composition and Substance Use – Smoking and Alcohol Use
Guizhou Sample
China Health and Nutrition Survey (CHNS) National Sample
Sex ratio
for age cohort 5-19
Sex ratio*first child
being a son
Year FEs,
Village/county FEs
Year *
Village/county FEs
Adjusted R2
AIC
N
Ln(Tobacco Consumption
Expenditure)
One or
two
One
children
One son daughter
-0.128
0.763***
0.046
(0.520)
(0.000)
(0.842)
0.946*
(0.079)
Ln(Alcohol Consumption
Expenditure)
One or
two
One
children One son daughter
-0.215
1.069**
0.219
(0.441)
(0.041)
(0.558)
1.303*
(0.076)
Ln(# Cigarettes Smokes Per Day)
One or
two
children
-0.211
(0.133)
2.335***
(0.330)
One son
1.289***
(0.278)
One
daughter
-0.286
(0.226)
Ln(# 50 Gms Liquor Drinks Per
Week)
One or
two
One
children
One son daughter
0.003
1.082***
0.161
(0.162)
(0.535)
(0.198)
2.356***
(0.371)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.123
3614.98
1261
0.141
693.94
310
0.094
500.97
250
0.102
3943.59
1261
0.124
778.78
310
0.087
568.97
250
0.465
2119.04
15380
0.425
6057.94
3407
0.483
2882.86
4136
0.307
2860.18
14506
0.331
8320.58
3821
0.283
2641.97
4113
Source: Author’s Guizhou household survey data (2004-2009), CHNS household survey data (1991-2006), and 1‰ sample of the 2000 China Population Census.
Notes:
[1] Control variables included but not reported here. Standard errors are clustered at the village level.
[2] Left panel (Guizhou sample): Sex ratios are measured at the village level using the 26-village census-type survey. The Cameron-Gelbach-Miller (2011)
bootstrapping method is adopted to address the issue of small number of clusters, and adjusted p-values are presented in the brackets.
[3] Right panel (CHNS national sample): Sex ratios are measured at the county level using a 1‰ sample of the 2000 China Population Census data. Robust
standard errors are presented in the brackets.
24
Table 4 Falsification Test: Using Sex Ratios for Age Cohort 30-40
Sex ratio for age cohort 30-40
Sex ratio*first child being a son
Adjusted R2
N
Ln(# Cigarettes Smokes Per Day)
One or
One
Two
One son daughter
Children
-0.330
-0.381
-0.433
(0.396)
(0.657)
(0.356)
0.053
(0.443)
0.178
0.143
0.066
3407
4136
15380
Ln(# 50 Gms Liquor Drinks Per Week)
One or
One
Two
One son
daughter
Children
-0.126
-0.498
-0.635
(0.611)
(0.876)
(0.475)
0.013
(0.619)
0.155
0.099
0.040
3821
4113
14506
Source: CHNS household survey data (1991-2006).
Notes: sex ratios are calculated at the county level for the age cohort 30-40. Robust errors are in the brackets. Standard errors clustered at the village level.
Control variables included but not reported here.
25
Table 5A Testing Potential Mechanisms – Income Effect
All families
Sex ratio for age cohort 5-19
-0.122
(0.095)
Sex ratio*first child being a son
Adjusted R2
N
0.207
23500
ln(per capita income)
Families with one or
One son
two children
-0.557*
-0.028
(0.421)
(0.227)
0.490
(0.443)
0.213
0.230
15380
3407
One daughter
0.727
(1.238)
0.192
4136
Source: CHNS household survey data (1991-2006).
Notes: Sex ratios are calculated at the county level for the age cohort 5-19. Robust errors are in the brackets. Standard errors clustered at the village level.
Control variables included but not reported here.
Table 5B Testing Potential Mechanisms – Effects by Income Quartiles (for one child family)
sex ratio*have a son
sex ratio*poorest income quartile
sex ratio*second income quartile
sex ratio*third income quartile
sex ratio*richest income quartile
Adjusted R2
N
Ln(# Cigarettes Smokes Per Day)
One
One
One son daughter
Child
1.269*
(0.656)
0.886***
0.031
0.402**
(0.188)
(0.279)
(0.198)
0.716***
0.368
0.311
(0.175)
(0.967)
(0.333)
0.701*** -3.367*
0.190
(0.212)
(1.811)
(0.247)
0.438
-1.061
0.571
(0.578)
(0.836)
(0.641)
0.294
0.155
0.157
3407
4136
7543
Ln(# 50 Gms Liquor Drinks Per Week)
One
One son
daughter
One Child
1.610**
(0.704)
1.176***
-0.280
0.283
(0.378)
(0.431)
(0.349)
1.057***
-0.774
0.108
(0.333)
(1.637)
(0.582)
0.712***
-1.510
-0.166
(0.211)
(1.801)
(0.372)
0.520
-1.270
0.588
(0.706)
(1.232)
(0.820)
0.253
0.155
0.136
3821
4113
7934
Source: CHNS household survey data (1991-2006).
Notes: sex ratios are calculated at the county level for the age cohort 5-19. Robust errors are in the brackets. Standard errors clustered at the village level.
Control variables included but not reported here.
26
Table 5C Testing Potential Mechanisms – Signaling Socioeconomic Status
All families
Village sex ratio for age cohort 5-19
0.569***
(0.008)
Sex ratio*first child being a son
Adjusted R2
N
0.046
2591
ln(tobacco pipe consumption (in 50g))
Families with one or
One son
One daughter
two children
-0.122
0.560***
-0.069
(0.763)
(0.010)
(0.569)
0.729**
(0.041)
0.060
0.154
0.124
1261
310
250
Source: Author’s Guizhou household survey data (2004-2009)
Notes: Sex ratios are calculated at the village level for the age cohort 5-19. Robust errors are in the brackets. Standard errors clustered at the village level.
Control variables included but not reported here.
Table 5D - Testing Potential Mechanisms – Stress Coping
Sex ratio for age cohort 5-19
Sex ratio*first child being a son
Year FEs, Village/county FEs
Year * Village/county FEs
Adjusted R2
N
One or two
children
-0.466***
(0.163)
-0.873*
(0.444)
Yes
Yes
0.249
16736
Life satisfaction
One son
-2.281***
(0.275)
One
daughter
0.181
(0.259)
Yes
Yes
0.215
4220
Yes
Yes
0.399
3991
One or two
children
-0.308***
(0.090)
-0.789***
(0.282)
Yes
Yes
0.102
16082
Happiness
One son
-1.685***
(0.230)
One
daughter
0.156
(0.119)
Yes
Yes
0.050
3855
Yes
Yes
0.351
3873
Source: CHNS household survey data (1991-2006).
Notes: sex ratios are calculated at the county level for the age cohort 5-19. Robust errors are in the brackets. Standard errors clustered at the village level.
Control variables included but not reported here. Life satisfaction ranges from 1 to 5 with 5 the most satisfied. Happiness ranges from 1 to 5 with 5 the
happiest.
27
Appendix I
Table 1 Median expenditures (CNY) in organizing major marriage related ceremonies (1996–
2009)
Year
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Coming-of-age
per event
3,208 (1.95)
3,387 (2.62)
4,284 (2.75)
8,046 (5.50)
8,154 (5.51)
Wedding (groom’s family)
per event
4,500 (3.00)
3,852 (2.84)
5,211 (3.85)
3,634 (2.64)
6,250 (4.85)
7,371 (5.81)
7,347 (5.20)
7,891 (6.22)
10,423 (8.24)
9,486 (5.76)
11,805 (9.14)
8,569 (5.50)
13,983 (9.56)
15,066 (10.18)
Wedding (bride’s family)
per event
3,157 (2.10)
3,100 (2.29)
3,025 (2.23)
3,829 (2.79)
2,929 (2.27)
5,644 (4.45)
4,536 (3.21)
5,143 (4.05)
4,243 (3.35)
7,633 (4.63)
7,502 (5.81)
4,927 (3.16)
5,833 (3.99)
7,766 (5.25)
House Building Expenses
per unit
30,064 (18.54)
49,000 (24.05)
41,323 (20.29)
51,887 (14.62)
66,675 (18.79)
Source: Author’s Guizhou household survey data.
Notes:
1. All expenditures have been adjusted for inflation based on CPI figures in the appropriate years’ China Statistical
Yearbook published by the National Bureau of Statistics (NBS) of China, various issues. All values are in CNY.
2. Recall data for coming-of-age ceremonies were collected only since 2005.
3. Numbers in parentheses denote expenditure as proportion of yearly average per capita income in the villages.
4. The average housing value per unit (1,000RMB) in 2004 is 53,000, and housing value to household monthly
income ratio is 73.
Appendix II
Table 2 Marriage Migration Pattern (The Guizhou Survey)
Stay in the Same township
Same county
village
different village different township
2004
53.5%
12.4%
24.2%
2006
54.0%
13.4%
24.5%
2009
54.6%
10.5%
24.9%
Source: Author’s Guizhou household survey data (2004-2009)
28
Same province
different county
8.9%
7.5%
9.4%
Different
province
1.0%
0.6%
0.6%
Appendix III
Geographic Location of the Two Datasets
Figure 1 China Health and Nutrition Survey (1991-2006)
Figure 2 Guizhou Survey (2004-2009)
29
Appendix IV Table 3 Heckman Two-step Estimations on the Likelihood of Family Bearing Only One Child
DVs in the main equation = Parents’ tobacco / alcohol consumption; DV in the selection equation = likelihood that parents stop at the 1st child
Nuclear family with
Nuclear family with
1st child a son
1st child a daughter
1st child a son
1st child a daughter
Ln(# Cigarettes
Stop at
Ln(# Cigarettes
Stop at
Ln(# 50 Gms Liquor Stop at Ln(# 50 Gms Liquor Stop at
Smokes Per Day) 1st child Smokes Per Day) 1st child
Drinks Per Week) 1st child Drinks Per Week) 1st child
0.94***
0.04
-0.44
0.03
1.03***
0.04
-0.45
0.03
Sex ratio for age cohort
5-19
(0.19)
(0.09)
(0.55)
(0.13)
(0.21)
(0.09)
(0.63)
(0.13)
Ln(per capita income)
0.21
0.06
0.35
0.07
0.24
0.06
0.4
0.07
Year of education
Household head gender
Marriage status
Household head age
Share of the elderly
Share of youth
Household size
Whether parents suffer
From major diseases
Minority status of the
household
Age of the first child
Whether the 1st child disables
or suffers from big diseases
Log Likelihood
N
(0.14)
-0.04
(0.04)
1.34***
(0.44)
-0.24
(0.21)
0.00
(0.01)
-0.07
(0.49)
-0.26
(0.45)
0.17
(0.11)
-0.42**
(0.17)
(0.05)
-0.01
(0.02)
-0.38*
(0.22)
-0.08
(0.15)
0.06***
(0.01)
-0.22
(0.21)
-0.26
(0.25)
-0.21***
(0.04)
-0.07
(0.09)
0.06
(0.14)
-0.06***
(0.02)
-0.15
(0.09)
(0.25)
0.02
(0.05)
-0.33
(0.49)
0.77
(0.87)
-0.04***
(0.01)
1.40**
(0.70)
-3.05***
(0.70)
0.11
(0.23)
0.15
(0.51)
(0.06)
0.00
(0.02)
-0.35
(0.25)
-0.40***
(0.15)
0.07***
(0.01)
0.25
(0.26)
-0.57***
(0.16)
-0.27***
(0.06)
-0.04
(0.15)
0.06
(0.18)
-0.10***
(0.01)
-0.16
(0.18)
(0.16)
-0.05
(0.05)
1.58***
(0.51)
-0.3
(0.24)
0.00
(0.01)
-0.04
(0.55)
-0.32
(0.52)
0.19
(0.12)
-0.48**
(0.20)
(0.05)
-0.01
(0.02)
-0.38*
(0.22)
-0.08
(0.15)
0.06***
(0.01)
-0.22
(0.21)
-0.26
(0.25)
-0.21***
(0.04)
-0.07
(0.09)
0.06
(0.14)
-0.06***
(0.02)
-0.15
(0.09)
-1121.91
-638.08
-1162.73
13183
11142
13183
Source: CHNS household survey data (1991-2006). Notes: Robust errors are in the brackets. Standard errors clustered at the village level.
30
(0.29)
0.02
(0.05)
-0.28
(0.55)
0.87
(1.01)
-0.05***
(0.01)
1.66**
(0.82)
-3.51***
(0.83)
0.14
(0.26)
0.23
(0.61)
-659.78
11142
(0.06)
0.00
(0.02)
-0.35
(0.25)
-0.40***
(0.15)
0.07***
(0.01)
0.25
(0.26)
-0.57***
(0.16)
-0.27***
(0.06)
-0.04
(0.15)
0.06
(0.18)
-0.10***
(0.01)
-0.16
(0.18)
Appendix V Spatial Econometric Estimation
Spatial econometric methods are employed to separately account for own pressure in the marriage
market (the direct effect) and peer pressure in the marriage market (the indirect effect). Specifically, we
estimate the Spatial Durbin Model (SDM) with individual fixed effects. Quasi-maximum likelihood (QML)
estimations are conducted. Following Elhorst (2012), we define Spatial Durbin Model (SDM) as
ρWyt + X it β + WX tϕ + µi + ε it
y=
it
The adjacency matrix W represents an N X N matrix of spatial weights. The Guizhou survey collected
rich information on lineal relationship networks as well as gift exchange networks to construct the
adjacency matrix W (Appendix Figure 3). X it is an 1 X k row vector of independent variables for panel
unit i at time t. yt is an N X 1 column vector of the value of y for all panel units in period t. Compared to
the Spatial Autoregressive Regression (SAR) model, the SDM further includes WX t , which denotes
spatially weighted independent variables that captures the interactions via independent variables. Next
we show how to derive the direct effect and the indirect effect.
The direct effect
Rewriting the SDM model in matrix form, yt =
[ I − ρW ]−1 ( X t β + WX tϕ + µ ) . Differencing with respect
to X t and averaging the result over all households generates the direct effect and the total effect as
follows:
M dir (k ) =
trace([ I − ρW ]−1[ I N β k + W ϕ k ])(1/ N )
M tot (k ) =
iN' ([ I − ρW ]−1[ I N β k + W ϕ k ])iN (1/ N )
where iN is an N X 1 vector of 1’s. The direct effect measures the impact of a unit change in variable X k
in household i on outcome in household i average over all households i=1…N. In the language of matrix,
the direct effect is defined as the average of the diagonal elements of the matrix. In contrast, the total
effect measures the impact of the same unit change in variable X k in all households on outcome in
household i and averages over all households.
The indirect effect
M tot (k ) − M dir (k ) . In the language of matrix, the indirect effect is
The indirect effect is M=
ind ( k )
defined as the average of either the row sums or the column sums of the non-diagonal elements of
these matrices. Since the numerical magnitudes of these two calculations of the indirect effect are the
same, it does not matter which one is used.
31
Figure 3 Lineal Relationship Networks and Gift Exchange Networks in One Village
Source: Author’s Guizhou household survey data.
Notes: Dots to the boundaries show households from other villages. Those bigger dots of the same color show
households in the same clan. This social network map shows one of the five villages in which complete information
on both lineal relationship networks and gift exchange networks was collected. This rich network data was
collected based on spontaneous written gift records each household kept for the past ten years. For more details
about the data, please refer to Chen (2014).
Table 4 Spatial Econometric Estimations: Marginal Effects
Spatial Durbin Model
Ln(Tobacco Consumption
Ln(Alcohol Consumption
Expenditure)
Expenditure)
Village sex ratio for age
0.43***
-0.03
0.42***
0.01
cohort 5-19
(0.11)
(0.28)
(0.12)
(0.31)
Sex ratio*first child being a
0.77*
0.71*
son
(0.39)
(0.42)
adjusted R2
0.01
0.02
0.01
0.03
N
552
552
552
552
Source: Author’s Guizhou household survey data (2004-2009)
Notes: Robust errors clustered at the village level are in brackets. Control variables included but not reported here.
Table 5 Spatial Econometric Estimations: Direct and Indirect Effects
Ln(Tobacco Consumption
Expenditure)
IDV: Sex ratio*first child being a son
Direct
Indirect
Non-spatial panel
0.73
Spatial Durbin model (SDM)
0.77
0.19
Ln(Alcohol Consumption
Expenditure)
Direct
Indirect
0.69
0.71
0.24
Source: Author’s Guizhou household survey data (2004-2009)
Note: Lineal relationship networks are adopted in the spatial network estimations.
32

Download Report