Winter_Heating_2_17

Winter Heating or Clean Air?

Unintended Impacts of China’s Huai River Policy



Douglas Almond, Yuyu Chen, Avraham Ebenstein



Michael Greenstone, and Hongbin Li



Preliminary and Incomplete (Do not Cite)



February 2010







Abstract



This paper examines the impact of prenatal exposure to total suspended particulates (TSP)

on infant and adult health outcomes. We exploit variation in coal use patterns in China in-

duced by government regulations that provided free home heating north of the Huai River. We

demonstrate that cities north of the Huai River have persistently higher TSP levels, and the

difference is most severe during cold winters. By exploiting within-city variation in weather

over time, we find that average birth weight is 14 grams lower for each additional 100 g=m3

of prenatal TSP exposure. We also find that the impact is largest among births already at risk

of being low birth weight, such as births to lower educated mothers. Areas north of the Huai

also have higher adult mortality rates for diseases related to respiratory performance (e.g. heart

disease, lung cancer). In contrast, we find no effect for illnesses unrelated to respiratory per-

formance (e.g. accidents, violence). By exploiting the rise in TSP just north of the Huai river,

we estimate that an additional 100 g=m3 of long-term TSP exposure reduces life expectancy

by roughly 2 years.



Corresponding author is Ebenstein: Department of Economics, Hebrew University of Jerusalem, Mount Sco-

pus, Israel (email: ebenstein@mscc.huji.ac.il); Almond: Department of Economics and SIPA, Columbia Univer-

sity, International Affairs Building, MC 3308, 420 West 118th Street, New York, NY 10027, and NBER (e-

mail: da2152@columbia.edu); Chen: Applied Economics Department, Guanghua School of Management and

IPER, Peking University, Beijing 100871, China (e-mail: chenyuyu@gsm.pku.edu.cn); Greenstone: Department

of Economics, MIT, 50 Memorial Drive, E52–359, Cambridge, MA 02142, Brookings Institution, and NBER (e-

mail:mgreenst@mit.edu); Li: Department of Economics, School of Economics and Management, Tsinghua Univer-

sity, Beijing 10084, China (e-mail: lihongbin@sem.tsinghua.edu.cn). We thank Matthew Neidell for his careful and

insightful discussion at the 2009 AEA Meetings. Ilya Faibushevich, Joan Fang, Alison Flamm, Eyal Frank, Nir Regev

and Henry Swift provided outstanding research assistance.





1

1 Introduction



Air quality in China is notoriously poor. Ambient concentrations of Total Suspended Particulates

(TSP) from 1981-2000 were more than double China’s National Annual Mean Ambient Air Quality

Standard of 200 g/m3 (Bi et al. 2007) and five times the level that prevailed in the US before the

passage of the Clean Air Act in 1970. Furthermore, air quality is especially poor in northern China,

which is home to several of the world’s most polluted cities (World Bank 2007). Following a career

in the southern China city of Shanghai, Prime Minister Zhu Rongji reportedly quipped in 1999:

“If I work in your Beijing [located in northern China], I would shorten my life at least five years”

(The Economist 2004).

This paper assesses the role of a seemingly arbitrary Chinese policy in producing dramatic

differences in air quality within China, and resulting differences in health outcomes. During the

1950-1980 period of central planning, the Chinese government established free winter heating of

homes and offices via the provision of free coal for fuel boilers as a basic right. The combustion

of coal in boilers is associated with the release of air pollutants, and in particular, the discharge of

fine particulate matter that can be extremely harmful to human health. These particles, which are

small enough to penetrate deeper into the lungs than other pollutants, are primarily associated with

the combustion of fossil fuels, such as coal. Due to budgetary limitations, however, this right to

home heating was only extended to areas located in northern China, which is defined by the line

formed by the Huai River and Qinling mountain range (see Figure 1). As we will demonstrate, the

particulate levels in cities north of the line are significantly higher than in cities south of the line, a

difference which persisted for several decades – and continues to be relevant today.

China provides a useful context in which to examine the relationship between air qual-

ity and human health outcomes for several reasons. First, China’s air is extremely polluted and

there is dramatic regional variation in this pollution. Regional variation in TSP levels in China

is partly due to the aforementioned home heating policy, and partly due to meteorological factors

that make northern China more vulnerable to high TSP levels, holding emissions constant (World

Bank 2007). Second, for the period in question, mobility was extremely restricted in China, and



2

so for our sample most individuals will be observed where they lived, and it will be a location

pre-determined by household registration – overcoming a limitation present in many studies of the

effects of air quality on health outcomes, where selective migration may lead to biased estimates.

Third, China’s problems with air pollution in the 1970s encouraged policymakers to establish mon-

itoring stations throughout the country, and as a result we are able to construct a rich data set of

air quality readings from 90 cities in China from 1981-2000. During this period, we also have

access to monthly temperature readings, and so are able to exploit within-city variation in air qual-

ity driven by weather to assess the impact of the policy on both ambient air quality and health

outcomes. Therefore, data availability makes China a suitable context for analyzing the role of air

pollution on health outcomes, and the role of the policy in explaining regional variation in health

outcomes in China.

As demonstrated by Almond et al. (2009), we find that the winter heating policy led to dra-

matically higher TSP levels in northern China.1 We demonstrate this in a cross-sectional regression

discontinuity-style estimation approach, where we exploit the home heating policy and the differ-

ence in coal use just north of the line formed by the Huai River and Qinling Mountains. We also

demonstrate that this result holds in a panel data setting, where we compare the marginal effect

of winter temperature on TSP in northern and southern China, after controlling for all permanent

city-level determinants of TSP concentrations and transitory ones common to all Chinese cities.

As a result of the large first-stage impact of the home heating policy on TSP levels, we examine

the role of the policy on infant and adult health outcomes. We analyze a matched sample of births

and recorded air quality using a survey of children conducted by UNICEF in 1992. The UNICEF

survey provides detailed information on the characteristics particular to parents that could affect

birth weight (e.g. income) and also provides information on the over 26,000 births that occurred

in cities covered by China’s National Air Quality Monitoring system. We are able to exploit the

impact of the policy on within-city variation in TSP levels over time to identify a causal link be-

1

Almond et al. (2009) examine data for all cities with air quality monitoring stations in China between 1981-1993

and demonstrate the impact of the policy on TSP levels, but fail to find an impact of the policy on sulphur dioxide

or nitrous oxide levels. In this paper, we examine the health consequences of the TSP variation associated with the

policy.





3

tween ambient air quality and birth weight. In order to assess the impact of TSPs on adult health

outcomes and life expectancy, we analyze China’s 145 Disease Surveillance Points (DSP), which

form a nationally representative sample of adult mortality in China for 1991-2000.

Our examination reveals a large impact of China’s high TSP levels on average birth weight.

We find using OLS that an additional 100 g/m3 of TSP in a child’s year of birth reduces birth

weight by 8-9 grams. Using 2SLS exploiting the Huai River instrument and the severity of winter

during a woman’s pregnancy, we estimate that an additional 100 g/m3 is associated with a re-

duction in average birth weight of 14 grams. We also find that the impact is largest among births

already at risk of being low birth weight, with the largest impacts found for lower educated moth-

ers. Based on our estimates that the Huai River policy induced nearly a 500 g/m3 increase in

ambient TSP levels for births occurring in cities just north of the boundary, this would suggest

a 70 gram decrease in average birth weight. As a baseline, heavy smoking during pregnancy is

thought to reduce birth weight by roughly 300 grams (Evans and Ringel 1997; Wilcox 1993). In

combination, these estimates imply that China’s coal use and the associated particulate matter is

imposing a cost on human health from ambient inhalation that is nearly a quarter as large as if a

pregnant woman had smoked heavily instead. Our results are robust to several specification checks

and falsification exercises, including 2SLS results using a regression-discontinuity approach and

falsification exercises with alternative policy borders, such as the Yangtze river.

Our analysis of adult mortality rates also indicates that China’s high TSP levels are impos-

ing a cost on life expectancy by increasing rates of diseases thought related to respiratory perfor-

mance. Building on an existing literature linking these illnesses to ambient air quality (Dockery et

al. 1993, Pope et al. 2002), we find that long-term TSP exposure (1981-2000) in a city is correlated

with higher incidence of illnesses that have been previously associated with smoking and ambient

air quality, such as heart disease, stroke, and lung cancer. In contrast, we fail to find a link between

TSP and causes of death not thought related to air quality, such as accidents and violence. We

estimate that an additional 100 g/m3 lowers life expectancy by roughly 2 years, exploiting the

increase in TSP at the Huai river, consistent with a growing literature that ambient air quality plays





4

an important role in long-term respiratory outcomes (Pope et al. 2009). Our results also indicate

a pronounced impact of TSP on lung related illnesses throughout the life cycle, with the largest

impact observed from age 50 and beyond.

The paper is organized as follows. In the next section, we present background on China’s

air quality and the link between the home heating policy and increased air pollution north of the

Huai River. The third section presents the data and descriptive statistics on which our analysis

is based. The fourth section demonstrates our econometric strategy and estimates the impact of

the Huai River Policy on observed TSP. The fifth section reports our results of the impact of TSP

exposure on birth weight. In the sixth section, we examine the long-term impact of TSP exposure

on cause-specific mortality rates and life expectancy. We conclude in section 6.







2 China’s Heating System and the Huai River Policy



China’s heating system was established during the three decades of the planned regime (1950-

1980). During this period, heating was provided to residents at no cost by the government via

state-owned enterprises (SOEs). The state continued to provide free heating systems for residential

and office buildings until 1998, and commercialized heating did not appear until the mid-1990s.

Due to budgetary concerns, however, the Chinese government was forced to limit the free home

heating policy to northern areas. Northern China is designated by a border formed by the Huai

River and Qinling Mountains. Cities north of this border receive unlimited free heating between

November 15 and March 15. In contrast, heating has been largely non-existent in southern China

due to the lack of government-supplied heating facilities. A private sector for heating has appeared

only recently. Indeed, it is widely recognized that winters are cold and uncomfortable in cities that

are just south of the Huai River, such as Nanjing, Shanghai, and Chengdu.

The Chinese heating system is coal-based and technically inefficient. It uses either coal-

fired heat-only boilers or combined heat and power generators, both of which are inefficient com-

pared to the electric, gas and oil heating systems used in many industrial countries (Wang et al.







5

2000; Jiang 2007). Heat for a residential building typically comes from a coal-burning boiler lo-

cated either in the same building or in a heating factory. Heated water is sent to each household

through iron pipes, often traveling long distances and losing significant energy along the way, re-

quiring the burning of more coal. Adding to the inefficiency of the system is the lack of proper

pricing mechanisms. Although many urban residents in northern China have started to pay for

heating, they do so through a single lump sum payment, and no meter system exists to measure the

amount of heat used by a household. With an inefficient heating system, harsh winters in northern

areas, and continued robust economic growth, it is no surprise that China consumes 39% of the

world’s coal (British Petroleum 2009). This coal use has had severe consequences for ambient air

quality.

Coal consumption is associated with the emission of soot, the particulate matter that enters

the environment when coal is burned. The incomplete combustion of coal in these boilers leads to

the release of air pollutants. There is little doubt that this causes substantial particle emissions, and

large differences during the sample period are observed between northern and southern China. The

disparity in the level of total suspended particulates (TSP) is particularly large during the winter

months. Fan et al. (2004), Bi el al. (2007), and He et al. (2001) show that in cities across northern

China, TSP concentrations are significantly higher in the winter than the rest of the year. In the

capital city of Xinjiang, Wulumuqi province, 90% of pollutants in winter months are emitted by

the heating system (Tianshan 2006). Qiu and Yang (2000) find that, for 1980-1994, visibility in

five northern cities was much lower in winter than throughout the rest of the year.

We examine the impact of China’s Huai River policy on air pollution concentrations us-

ing two econometric strategies. First, we test whether concentrations are higher in northern cities

relative to southern ones, after adjustment for a polynomial in latitude. This test has some similar-

ities with a regression discontinuity-style approach that has become increasingly popular in recent

years (Cook and Campbell 1979; Greenstone and Gallagher 2008; Almond et al. 2008). We use

this plausibly exogenous variation in air quality to estimate the impact of air pollution on long-term

health outcomes. Second, we test whether concentrations are higher in northern cities, relative to





6

their long run average, after adjustment for the realized temperature in year. This approach takes

advantage of the substantial inter-annual variation in temperature to compare ambient concentra-

tions in northern and southern cities in years where the ambient temperatures are similar. We then

proceed to exploit this plausibly exogenous variation in air quality to estimate the impact of air

pollution on birth weight.







3 Data Sources and Descriptive Statistics



This paper utilizes four primary data sets. Air quality data is compiled from China’s national

monitoring system, which reports the recorded annual daily average concentration of TSP for the

cities in our sample for 1981-2000. The second data source necessary for the analysis is monthly

average temperature data, which is available for each city in which we have air quality monitor

data. These data are collected by the China Meteorological Administration and allow us to cal-

culate for each city the average winter temperature, which we define as the average temperature

for November-February. Our third data set is a large sample of births that overlap with our air

quality readings. We focus on the National Sample Survey on the Situation of Children (NSSSC),

a nationally representative survey sponsored by the State Statistic Bureau of China and UNICEF.

The sample was selected using a stratified two-stage, probability-proportionate-to-size approach,

with systematic sampling at each stage. The survey sampled a total of 570,274 children aged 0–14

years, is among the largest surveys of child welfare ever conducted in China, and provides detailed

information on both births and features of the mother and household. The air quality and weather

data are matched to the survey, and each reported birth is assigned the TSP and winter temperature

experienced during the woman’s pregnancy, given as T SPjt = AT SPjt 1 (1 Wjt )+AT SPjt Wjt .

This is necessary because TSP is only available annually, and we are interested in the pre-natal ex-

posure to air pollution. The fourth data set, the China Disease Surveillance Points (DSP), is used to

analyze patterns in adult mortality and cause-specific death rates, and will be discussed in section

5.







7

Table 1 reports summary statistics for the matched sample of births and TSP measures,

which consists of births between 1981 and 1991 where air quality monitoring readings are avail-

able. The most striking result is the average level of TSP’s observed for the period. We report that

the average birth in our sample is assigned a TSP g/m3 of 715. While TSP readings were high

in northern China, the readings were quite high during this period even in the south, which aver-

aged 325 TSP g/m3 . For comparison, prior to the Clean Air Act in the United States, the average

TSP recorded was 100 g/m3 (1964) among monitored cities and US standards2 require cities to

have concentrations below 75 g/m3 (US Environmental Protection Agency3 ). As another point of

reference, many observers worried about Beijing’s air quality for the Olympic Games on the basis

of the city’s poor environmental record in recent years. Between 1981 and 2003, however, Bei-

jing’s TSP levels had declined from 1,100 to 252 g/m3 – meaning that for the sample of births we

observe, particulate matter was roughly 4 times higher than today’s levels. The atrocious environ-

mental record for China during this window provides an interesting context in which to examine

the health consequences of air pollution using some of the highest recorded pollution readings in

history (World Bank 2007).

For each birth in the sample, we observe the mother’s age, education, and household in-

come, the sex of the birth, and importantly, the temperature during the winter months overlapping

with the pregnancy. The table reflects that the average southern birth was exposed to monthly tem-

peratures averaging 470 F, whereas northern births were exposed to more extreme conditions, with

the average birth being exposed to a monthly temperature of 280 F, below freezing and a temper-

ature that would induce extensive coal use. Since we are interested in exploiting the Huai River

Policy’s effect on coal use, we consider samples that are increasingly close to the border, which lies

at roughly 330 latitude. While the overall sample contains 24,088 births, in the sample restricted

to births very close to the border (between 300 and 350 ), we have only 6,466 births – but for those

births, the policy presumably has the most relevance and our identification strategy is most com-

2

The US standard was changed in 1987 to regulate particulate matter that was smaller than 10 micorons (PM10 ),

and again in 1997 to regulate particulate matter that is smaller than 2.5 microns (PM2:5 ). The 75 g/m3 is the most

recent available US TSP standard.

3

http://www.epa.gov/ttn/catc/cica/airq_e.html#3





8

pelling. In the next section, we examine how recorded TSP readings change near the Huai river,

and how the responsiveness of TSP to winter temperature is different on either side of the river at

these extraordinary TSP concentrations.







4 Impact of the Huai River Policy and Winter Temperature on



TSP Levels



One method for examining the impact of the home heating policy on TSP is to exploit the discon-

tinuity in the cost of coal for home heating at the river and the impact on TSP levels by estimating





T SPijt = 0 + 1 Nj + f (Lj ) + Xijt + t + ijt (1)





where T SPijt is the total suspended particulates in city j in year t experienced by fetus i. Nj is a

"North" dummy indicating whether city j is north of the Huai river, f (Lj ) is a polynomial function

of the latitude of city j, Xijt is a vector of the observable characteristics of the birth that might

influence birth weight other than air quality (e.g. parental income), t is a set of year fixed effects,

and ijt is a disturbance term. Under the assumption that a city’s TSP reading would be a smooth

function of latitude if not for the home heating policy, the North dummy identifies the causal effect

of being north of the river on TSP levels. In this circumstance, the impact of the policy would

be identified by Nj , and Nj would represent an instrument for air quality that could be used to

estimate the impact of air quality on other outcomes, such as birth weight. In fact, air quality does

rise discontinuously near the Huai river, and persistent differences are observed between cities

north and south of the border, and in particular, in places near the river.

Figure 2 previews these results visually. It plots the bivariate relation between the average

TSP concentration observed in a particular city and its latitude. The plots are done separately

for cities to the North and South of the Huai River/Qinling Mountains line and come from the

estimation of nonparametric regressions that use Cleveland’s (1979) tricube weighting function





9

and a bandwidth of 0.5. The figure presents dramatic evidence that northern cities have higher TSP

concentrations. An especially convincing feature of the graph is the evidence of a discontinuous

increase in TSP concentrations at latitudes just above the Huai River line. This jump is meaningful,

because it seems improbable that other determinants of air pollution change as discretely upon

crossing this line as does the heating policy.

As shown in panel 1 of Table 2, in the overall sample the North dummy is estimated to

be associated with 534 g/m3 increase in TSP. In restricted samples closer to the Huai river, the

result is somewhat larger: when the sample is restricted to births between 280 and 400 latitude, the

impact of being above the Huai is a 715 g/m3 increase in TSP. This is compelling evidence that

the rule affected TSP concentrations, but for inference regarding infant outcomes, we focus on a

strategy in which we account for time-invariant factors that might be particular to a city. It also

may be that observables change near the border, making the quasi-regression discontinuity strategy

inappropriate for inference regarding infant outcomes and air quality.

Our main identification strategy exploits both the discontinuity at the river and within-city

(across-year) variation in the severity of a winter. Specifically, we estimate the following equation:





T SPijt = 0 + 1 Tjt + 2 Nj Tjt + Xijt + j + t + ijt (2)



where Tjt refers to the average temperature during winter months in city j in year t, j is a set

of city fixed effects, and Nj Tjt is a variable that takes on the value zero for all southern cities,

and takes on the value of Tjt for northern cities. Under the assumption that in the absence of the

home heating policy, TSP levels would not be differentially responsive to temperature north and

south of the Huai river, we can identify the causal impact of temperature variation on air pollution

by this interaction term. This strategy can also be thought of as a "differences in differences"

approach, since we anticipate that air pollution will be worse in cold winters both above and below

the Huai river, but the impact will be larger above the Huai river. Since the models include city

fixed effects, we are essentially comparing the differential influence of the within-city change in

weather on pollution across the river. We expect that a colder (relative to a city’s own average)



10

winter will pollute a northern city more than a southern city because of the policy and the low

price of coal-driven indoor heating. In Figure 3, we present a visual preview of our identification

strategy, and plot the bivariate relation between TSP and winter temperature, separately for births

occurring north and south of the Huai river. The plot reflects that the gradient between winter

temperature and TSP levels is noticeably steeper in northern China than southern China. This plot

is very similar in spirit to our main identification strategy in which we exploit within-city variation

in temperature.

As reported in panel 2 of Table 2, the impact of winter temperature on TSP readings in

northern cities is large. Each additional degree of winter temperature in northern China lowers the

observed TSP by 73 g/m3 , and this estimate is significant at the 1% level. In the samples of births

closer to the Huai river, we observe a similar effect; for births between 280 and 400 latitude, each

additional degree of winter temperature lowers the TSP by 98 g/m3 . For births very close to the

river (between 300 and 350 latitude), we estimate that each degree lowers TSP by 49 g/m3 , but the

estimate is only statistically significant at the 10% level due to the small sample size. Note that this

result is found even after including city, year, and month fixed effects. Provided that cold winters

do not differentially affect other factors that would affect birth weight between north and south,

this provides a valid instrument for the TSP in utero experienced by the fetus. We will further

examine challenges to our identification assumption in the next section.





4.1 Is the Huai River Instrument Correlated with Observables?



In Table 3, we examine whether our two proposed instrumental variables strategies provide plau-

sibly exogenous variation in air quality. In particular, we are interested in the correlation between

the chosen instrument and the covariates of the birth parents. If the covariates are uncorrelated

with the instrument, it is reasonable to presume that unobservables are also similar, which is crit-

ical for the instrument to be valid. Balancing along the instrument also implies that the results

will not hinge on the functional form chosen for our estimation methods. In Panel 1 of Table 3,

we examine the correlation between a dummy for the birth occurring north of the Huai river (Nj )





11

and covariates, such as the sex of the child, household income, the education of the mother, and

other family attributes that could potentially affect birth weight. We find limited evidence that a

correlation exists between being born North of the Huai and family income. Note, however, that

our results indicate that parents North of the Huai appear wealthier and better educated, implying

that any estimated effect would be an understatement of the true effect. However, this does indicate

that for birth outcomes, the discontinuity approach may be compromised by a failure to account for

time-invariant variation to a location. In Panel 2 of Table 3, we examine the correlation between a

dummy for the birth occurring North of the Huai interacted with the temperature during the time

spent in utero. These results are quite strong, and they indicate that there is almost no correla-

tion between features of the birth and the chosen instrument (Nj Tjt ). While it is straightforward

to demonstrate that the instrument is essentially randomly assigned in the sense that parents who

have children in cold winters are similar to parents who have children in other winters, we are

also interested in evaluating whether the relationship between winter temperature and birth weight

we observe is in fact driven by air pollution, or potentially by other factors that would violate the

exclusion restriction. These concerns are dealt with further in the next section.







5 Empirical Results on TSP and Birth Weight



5.1 Models without City Fixed Effects



In Table 4 Panel A, we present the results of OLS models of the impact of air quality on infant

outcomes where we do not account for time-invariant factors that might affect birth weight within

a city. These models are of the form:





BWijt = 0 + 1 T SPijt + Xijt + t + ijt (3)





where infant i’s birth weight is modeled as a function of pre-natal exposure to particulates T SPijt ,

the covariates of the family Xijt , and fixed effects to account for variation in birth weight particular





12

to year t: The results indicate that an additional 100 g/m3 of measured TSP is associated with a 3.2

gram reduction in birth weight and is not statistically significant. Likewise, in the samples of births

closer to the Huai river, we fail to identify a robust effect; for births between 280 and 400 latitude,

an additional 100 g/m3 of measured TSP is associated with an 8.3 gram reduction in birth weight,

and this result is statistically significant at the 10% level. For samples closer to the river, we fail

to find an effect. These results indicate that simply comparing births by TSP exposure, without

accounting for city-specific effects, may not be an effective method of identifying an impact of

TSP on birth weight.

In Table 4 Panel B, we report the 2SLS estimates for the impact of air quality using the

quasi-regression discontinuity strategy. These models are of the form





BWijt = 0 + d

1 T SP ijt + Xijt + t + ijt (4)





d

where T SP ijt is taken from the first stage regressions in Table 2 (equation 1), where the excluded

regressor is whether the birth occurred north of the Huai The results reflect that areas just north of

the Huai river have lower birth weight. In the overall sample, we estimate that an additional 100

g/m3 of TSP are associated with a reduction of 7.4 grams in average birth weight. In the sample

near the Huai river, however, we estimate that an additional 100 g/m3 of TSP reduce birth weight

by 17.8 grams. In these specifications we do not include city fixed effects, however, and the models

do not fit as tightly, yielding larger standard errors. The results suggest the need for models where

we account for time-invariant factors that are specific to a location and affect birth weight.





5.2 Models with City Fixed Effects



In Table 4 Panel C, we present the results of OLS models of the impact of air quality on infant

outcomes where we account for time-invariant factors that affect birth weight and are particular to









13

a city. These models are of the form:





BWijt = 0 + 1 T SPijt + Xijt + j + t + ijt (5)





where infant i’s birth weight is modeled as a function of pre-natal exposure to particulates T SPijt ,

the covariates of the family Xijt , and fixed effects to account for variation in birth weight particular

to city j and year t: The results indicate that an additional 100 g/m3 of measured TSP is associated

with an 7.98 gram reduction in birth weight, which is statistically significant at the 1% level. In the

samples of births closer to the Huai river, we observe a similar effect; for births between 280 and

400 latitude, an additional 100 g/m3 of measured TSP is associated with an 9.17 gram reduction

in birth weight. These results are actually quite similar to other estimates for China, such as Wang

et al. (1997) who found a 6.9 gram reduction in birth weight in a sample of mothers in several

neighborhoods in Beijing. These results, however, are subject to the normal caveats with OLS

models. It may be that TSP variation within a city across time is associated with other confounding

factors that may either improve or worsen infant health outcomes. A potential endogeneity may

exist if, for example, demand-side shocks for manufactured goods induces both higher TSP levels

but increases in income. In this case, OLS would potentially understate the true cost in birth weight

if factors that induced air pollution were associated with temporary increases in wealth, and the

ability to offset the negative consequences of air pollution with other pre-natal forms of care. We

proceed with our IV models which attempt to address the potential endogeneity of TSP exploiting

the Huai River Policy and plausibly exogenous variation due to within-city variation in temperature

over time.

In Table 4 Panel D, we present the 2SLS estimates of the impact of TSP on birth weight

using the interaction of winter temperature and a North dummy as an instrument for TSP levels.





BWijt = 0 + d

1 T SP ijt + 2 Tjt + Xijt + j + t + ijt (6)





d

where T SP ijt is the fitted value for TSP following from our first stage reported in panel 2 of





14

Table 2 (equation 2). In column 1 we report the 2SLS results for the overall sample, and we

estimate that a 100 g/m3 increase in TSP is associated with a 14.3 gram decline in birth weight,

statistically significant at the 5%. Relative to the OLS estimate of 7.96 grams, and this indicates

that 2SLS produces coefficients that are nearly twice as large. In samples closer to the river, the

point estimates are even larger. As shown in column 2, among births between 280 and 400 latitude,

an additional 100 g/m3 of TSP reduces birth weight by 20.5 grams. For births within the narrow

range of 300 and 350 , the coefficient indicates that a 100 g/m3 increase reduces birth weight by

39.9 grams, but this is measured imprecisely and is not statistically significant. These results are

compelling evidence that the Huai river policy is adversely affecting birth outcomes. Our results

are identified off variation in air quality due to the steeper TSP-temperature gradient observed in

northern China, and indicate that infants born following severe winters have lower birth weight.

We explore in the next section the robustness of our main result to alternative specifications, and

subject our empirical strategy to a set of falsification exercises.

In Table 4 Panel E, we present the results of the reduced form relationship between the

instrument for air quality and birth weight. These models are of the form:





BWijt = 0 + 1 Tjt + 2 Nj Tjt + Xijt + j + t + ijt (7)





where we are interested in the coefficient on the interaction between the winter temperature and the

dummy for North of the Huai. A graphical (and simplified) version of this regression is presented

in Figure 4, where we regress birth weight directly on temperature and control for year and city

fixed effects, as well as demographic controls. The figure reflects that extremely cold winters are

associated with lower birth weights. We proceed in Table 4 with more rigorous analysis, which

exploits the North-South comparison. As shown in column 1, in the overall sample, births north of

the Huai river in cold winters have lower birth weights. Specifically, an additional 10 increase in

temperature is associated with a 10.5 gram increase in recorded birth weight, which is statistically

significant at the 10% level. These models include a rich set of controls for the features of the







15

parents as well as city and year fixed effects. Note that the other coefficients reflect that birth

weight is correlated with other measures of wellbeing and resources. birth weights are higher for

older mothers, higher income families, and for mothers with more education. In columns 2-5, we

report the results of the same regressions in samples increasingly close to the Huai river border.

Among births between 280 and 400 latitude, the impact of an additional degree is to increase the

birth weight by 18.99 grams, which is statistically significant at the 1% level. The results in the

sample very close to the river, where we include only births between 300 and 350 , indicate that a 10

increase in temperature north of the Huai is associated with a 19.6 gram increase in birth weight,

but this is not statistically significant due to the smaller sample size and larger standard error of the

estimate.





5.3 Robustness Checks and Falsification Exercises



In Table 5, we present a set of robustness checks and falsification exercises with respect to the use

of temperature and the Huai river policy as an instrument for TSP levels. In Panel A, we report

the result for a baseline first stage and 2SLS estimate from Table 2 and 4, which reflect that an

additional degree of temperature in the North reduces TSP by 73 g/m3 , and that an additional

100 units of TSP lower birth weight by 14.3 grams. A potential challenge to our identification

strategy is that our result is missing the potential impact of it being very cold indoors in the South

since the policy affected not only ambient air quality but also people’s access to sufficient indoor

heating, implying a violation of the exclusion restriction. As an attempt to rule out this effect, in

Panel B we include controls (row b) for whether the average monthly temperature in the winter

was below 400 F, and the interaction of this dummy for extreme weather with the North dummy.

We continue to include in this specification all the regressors from our baseline, including a main

effect for winter temperature. If temperature only affects infant health in extreme conditions, but

TSP rises linearly with the use of coal, this specification should allow us to control for the direct

impact of very cold weather in either the North or South on birth weight that is separate from the

impact of TSPs, which are also affected by extreme temperature. The results are very similar to our





16

baseline results but with slightly larger point estimates for 2SLS (14.3 versus 16.6). As a second

robustness check, we include instead a full set of controls for the temperature in non-winter months

(March-October) and this variable interacted with the North dummy. The result is very similar to

the base case result (14.3 versus 15.0), and provides further support for the claim our identification

is being driven by responsiveness to cold winters, rather than the variation in temperature for the

other months.

In Panel C, we present the results from a set of falsification exercises to further explore

the Huai river instrument’s validity and our empirical results. In a first specification, we examine

whether other demarcations in China have relevance for air quality. We examine whether a first

stage and statistically significant 2SLS result exists for a set of alternative demarcations. Using

a 250 border, a 400 border, or a Yangtze border, we find that none of these are able to produce

a statistically significant first stage and 2SLS result.4 This is compelling evidence that the air

pollution induced by the coal policy associated with the Huai River-Qingling Mountains border

is in fact responsible for the results in our main sample. In our last robustness check, we assess

whether air quality affects post-natal birth weight. Note that in our regressions, our measure of

TSP is a weighted average of air quality for each birth. As a falsification exercise, we examine

whether post-natal air quality has a statistically significant relationship with birth weight. This

is actually a high-powered falsification exercise, because these measures are correlated, since our

TSP measures are annual and many births will have highly correlated natal- and post-natal TSP

readings. However, the result shows that TSP in the post-natal window is much more weakly

correlated with the birth weight, indicating that TSP levels are affecting birth weights.





5.4 Heterogeneity



In Table 6, we examine the heterogeneity in the impact of air pollution on birth weight. In panel

1, we stratify the sample of births by the mother’s education level. Among mothers in the bottom

4

The Yangtze falsification is composed of all observations south of the Yangtze, and north of the Yangtze but south

of the Huai.









17

third of the distribution (primary education or less), we estimate using 2SLS that an additional

100 g/m3 of TSP lowers birth weight by 36 grams, more than twice as large as our overall effect.

In contrast, among mothers with an intermediate level of education (junior high) or in the highest

category (high school and beyond) we find that an additional 100 g/m3 of TSP lowers birth weight

by 11 grams, and these results are not statistically significant. In panel 2, we consider how TSP af-

fects the chance a birth has either very low birth weight (<2,750 grams), low birth weight (<3,000

grams), or below median birth weight (<3,250 grams). The outcome variable in these regressions

is a binary for whether the birth weight was lower than the specified level, and the results reflect

that exposure to air pollution has the greatest impact at the tail of the distribution. We estimate

that an additional 100 g/m3 of TSP increases the probability of the birth having a very low birth

weight by 8%, significant at the 5% level. In contrast, the impact of TSP is negligible at affecting

the chance of a birth being in the bottom quarter or below the median birth weight.

These results have important implications for the expected health consequences of our es-

timate of the average impact of TSP on birth weight. Since air quality appears to affect those of

lower means (education) and those births which are already at risk of being low birth weight, the

health consequences of the pollution are likely to be large since these parents are less able to cope

with a more vulnerable infant. The fact that the impact is largest in inducing birth weights in the

bottom 10% of the distribution suggests that infants exposed to China’s air pollution might have

significantly lower endowments of health at birth. In light of evidence of non-linear impacts of

birth weight on health outcomes, where large benefits are observed from avoiding births with very

low birth weight (Behrman and Rosenzweig 2004), this amplifies the relevance of our estimate of

the effect on average birth weight. Future research should examine how these adverse outcomes

affect infant health later in life using our instrument and the plausibly exogenous variation in birth

weight it induces, in light of the academic debate over the relevance of birth weight in causing

adult health problems (Almond, Chay, and Lee 2005; Black 2007). Our identification strategy also

provides variation in birth weight that could be used to examine how parents respond to a negative

health "shock". It is an open research question whether parents compensate or reinforce the en-





18

dowment associated with a birth outcome, and our analysis presents novel variation in birth weight

(endowment) that could, in combination with measures of parental investment, shed light on this

question that has drawn attention from recent research (e.g. Rosenzweig and Zhang 2009).







6 Adult Mortality and the Huai River Policy



In the following section, we examine whether the high TSP levels in northern China have had

consequences on adult mortality, and we focus on diseases that are thought to be related to air

quality. Diseases that have been linked to ambient air quality include heart disease (Laden et al.

2009, Pope et al. 2002, Dominici et al. 2006), stroke (Hong et al. 2002a, Hong et al. 2002b,

Wellenius 2005), lung cancer (Laden et al. 2009, Pope et al. 2002, Kabir et al. 2007), and

respiratory illnesses (Laden et al. 2009, Pope 1989, Dab et al. 1996). Diseases presumably

unrelated to air quality include other cancers, accidental or violent deaths, and various stomach

ailments.

Estimates of the health consequence to long-term TSP exposure in China are few in num-

ber, but in fact China is an almost ideal context to analyze the long-term consequences of air

pollution. In many of the referenced studies, there is no attempt to identify the factors that are

causing the increase in air pollution, and so it is difficult to disentangle the direct effect of air pol-

lution from confounding factors that might be associated with air pollution. Another issue arises

in research among populations in Western countries where mobility is unrestricted; free mobility

results in a selected sample of people located near air pollution, who may have less means or be

less healthy regardless of the pollution-induced health problems. Another reason to focus on China

is that its TSP levels are so high and without historical precedent that parameter estimates based on

the relatively clean air in the United States and Western countries may not be that useful for China

to appropriate policy decisions in the context of air pollution. The Huai River Policy provides a

natural experiment to examine long-term health consequences of China’s air pollution. The popu-

lation in China that we focus on had very little potential for migration under the hukou system, and







19

so ambient air quality at each site is presumably an accurate measure of the population’s exposure.

We also improve on existing studies by analyzing a nationally representative data set for China,

which is described in the next section.

Note that unlike the analysis on infants, this is a longitudinal analysis and so it does not

include fixed effects that account for location.5 While short-term air quality fluctuations have been

shown to affect health outcomes (Katsouyanni et al. 1997), our analysis here aims to assess whether

there are long-term impacts in China of poor air quality. Models with location fixed effects will

only identify how air pollution in a given year affects mortality in that year – ignoring the potential

effect of air pollution in previous years on life expectancy. These models may also capture the

"harvesting" effect, where the mortality increase in years of high pollution are lowering mortality

rates in later years, and so it may not accurately reflect the total loss in life expectancy (Rabl 2005).

As such, we proceed with our analysis of long-term TSP exposure and mortality using a unique

data set for China, described in the next section.





6.1 Data



The analysis of adult mortality patterns in China is based on China’s Disease Surveillance Point

system (DSP). The DSP is a set of 145 sites chosen to form a nationally representative sample of

China’s population, and selects sites across different levels of wealth and urbanization. The cover-

age population was also chosen to reproduce geographic dispersion in China’s population, relative

to patterns in China’s 1990 census. The DSP records all deaths among the coverage population

of 10 million residents at the points and, due to careful sample selection of the DSP sites, yields

an annual sample of deaths that mirrors patterns in the country nationwide (Yang 2005). This

paper relies on the data taken from roughly 500,000 deaths recorded at DSP sites between 1991

and 2000, and population counts by age and sex that are used to convert the recorded deaths into

death rates. Each site is identified as being either north or south of the Huai river, and assigned

5

Note however that Northern and Southern China have similar smoking rates and dietary differences mostly relate

to the Northern wheat culture versus the Southern rice culture. This should not have a confounding effect on the causes

of death we examine. Results on diet and smoking patterns by region in China available from the authors upon request.





20

the long-term TSP average from 1981-2000. The long-term average for each year at each site is

calculated through a regression of TSP on year and city dummies for all years prior to the DSP

observation. The assignment of TSP in a particular year and city is calculated with a regression

of TSP on year and city dummies for data from all years prior to the DSP observation. The TSP

assigned to the observation is the value for the city fixed effect. This is done to overcome missing

data in our sample, since we have an unbalanced panel of air quality readings. The 145 DSP sites

with an air quality monitoring station within 150 kilometers for the ten available years yields a

sample of 1,139 observations, where each observation is a DSP site in a given year.6 Summary

statistics on the DSP and matched air quality data are shown in Table 7.





6.2 Cause-Specific Death Rates and the Huai River Policy



In this section, we examine how both particulates and cause-specific mortality rates change around

the discontinuity formed by the Huai river.





T SPjt = 0 + 1 Nj + 2 f (Lj ) + Xj + jt (8)



ln(DRjt ) = 0 + d

1 T SP jt + 2 f (Lj ) + Xj + jt (9)





where T SPjt is the total suspended particulates average at DSP site j in years prior to t and DRjt is

the log of the death rate at DSP site j. This is similar to the RD approach discussed in Section

3, where we anticipate that air quality should decline continuously with respect to latitude, and

be captured by f (Lj ). Under this assumption, Nj represents the difference in air quality that can

be attributed to the Huai River Policy and will represent a proxy for the discontinuous increase in

particulate matter near the Huai river.

In Table 7 Panel A, we report the coefficient on Nj which indicate that north of the Huai

the average TSP was 227 g/m3 at DSP points north of the Huai. After accounting for a cubic in

latitude, the difference between North and South remains 195 g/m3 , implying Northern residents

6

Results where we restrict the sample to DSP sites at closer ranges to air quality monitoring stations are very

similar to the results we report. Results available upon request.





21

were on average exposed to 195 g/m3 additional units of TSP due to the Huai River Policy.7 In

Panel B, we demonstrate that DSP sites south and north of the Huai river are reasonably similar

along observable dimensions: education, share in farming or manufacturing, income, and rates of

urbanization are similar and differences are statistically insignificant. In Panel C, we report the

differences in heating and cooling degree days; degree days are the sum of the difference between

the temperature and 650 F. The results reflect that while northern China and southern China have

very different climates, the difference is statistically insignificant after accounting for the cubic in

latitude. The similarity between southern and northern China along all dimensions after controlling

for a cubic in latitude – with the exception of TSP – is compelling evidence that the Huai River

Policy affected the TSP exposure of inhabitants on both sides of the Huai river, and presents an

opportunity to estimate the causal impact of air pollution on health outcomes.

In Table 8, we present the results of regression analysis where we estimate models of the

death rate disaggregated by cause at each DSP site and year. Each cell in the table represents the

coefficient from a separate regression. All models have TSP ( g/m3 ) as the independent variable

of interest. These models include the controls shown in Table 7. In column 1, we report results

without controlling for latitude or the observable dimensions listed in Table 7. In column 2 we

add a cubic in latitude, and column 3 is our main specification with the cubic in latitude as well

as the observables. The results indicate that air pollution increases lung related illnesses in a

statistically significant manner but other causes of death are uncorrelated with air pollution. In

Panel A, we report OLS estimates of the impact of TSP on lung- and non-lung related illnesses.

While an additional 100 g/m3 of TSP raise the log of the death rate by 1.87%, this is composed

of a 3.24% increase in the death rate for lung related illnesses, and a 0.3% increase for all other

causes of death. In Panel B, we report 2SLS estimates using the Huai River instrument and the

coefficients are much larger than the OLS estimates. An additional 100 g/m3 raises the log of

the death rate for all causes by 17.5% and lung related illnesses by 26.5% - but non-lung related

7

Our regression strategy for accounting for an unbalanced panel yields a negative average for TSP due to the

structure of the values for year fixed effects. For both the South and North, we report the average value for the city

fixed effects plus 450, so that the average is near the actual average value. This does not affect any of the regression

analysis.





22

illnesses only rise by 3.2%. The results suggest that long-term TSP exposure is reducing the life

expectancy of inhabitants north of the Huai river. While the 2SLS results are much larger than the

OLS estimates, this may be due to wealth creation associated with air pollution that mitigates the

health consequences of pollution. Both sets of results suggest that overall death rates, and lung

related illnesses in particular, are responsive to TSP exposure in China.

In order to further examine how TSP is affecting mortality rates over the life-cycle, in Fig-

ure 5 we analyze the relationship between age-specific death rates from lung and non-lung related

illnesses with TSP exposure. The regressions are estimated using the sames specification reported

in column 3 of Table 8, where the log of the death rate of the listed cause is the dependent variable,

long-term TSP average is the independent variable, and North is the excludable instrument for the

2SLS models. Each point in the plot represents the coefficient from a separate regression, where

we examine how age-specific death rates are affected by long-term exposure to TSP ( g/m3). The

95% Confidence Intervals are drawn around the point estimates. The figures indicates that the

impact of air pollution on lung related illnesses is evident on death rates among persons age 50

and beyond, and relative to the impact of pollution on non-lung related illnesses, the lung related

illnesses are much more sensitive to TSP. As in the overall sample, our analysis of age-specific

death rates indicates that OLS may be understating the true cost of air pollution relative to 2SLS,

possibly due to confounding variables that are correlated with air pollution and factors that allow

people to mitigate the consequences of air pollution (e.g. wealth). The results suggest that from

age 50 and beyond, high TSP exposure leaves individuals at an elevated risk of dying form a lung

related illness.

In a final exercise, we examine Zhu Rongi’s claim that life expectancy is lower in northern

China due to poor air quality. We calculated life expectancy at birth in each of the DSP sites by

year by calculating age-specific death rates and creating a life table presuming an individual had

been exposed to age-specific mortality rates observed at a particular DSP site in a particular year.

In Table 8 Panel C, we report both OLS and 2SLS estimates of the life expectancy cost of an

additional 100 g/m3 : Our OLS and 2SLS estimate are that life expectancy is .42 and 2.1 years





23

lower respectively at DSP sites north of the Huai river. These estimates are lower than those found

by Pope et al. (2009) who estimate that 100 g/m3 reduces life expectancy by roughly 6 years

using data from the United States. Their study identified causal estimates using a first-difference

regression of city-level improvements in life expectancy on improvements in air quality in the

wake of the Clean Air Act. Our results may differ from theirs for a variety of reasons; it may be

the effect of air pollution is non-linear, and Chinese levels are much higher than US levels. Other

possibilities include our different identification strategy and our focus on the Chinese population.

Interestingly, Zhu Rongi posited that Beijing’s air pollution could cost an inhabitant 5 years of

life expectancy. Since the pollution was roughly 200 g/m3 higher north of the Huai, the estimate

in our data would be 4.2 years, implying that Rongi’s estimate may in fact be reasonable and

that inhabitants of Northern China are experiencing long-term health consequences due to their

exposure to high TSP levels.







7 Conclusions



This paper examines the impact of neonatal exposure to total suspended particulates (TSPs) on

birth weight and adult mortality using the extraordinary levels of ambient particulate matter ob-

served in Chinese cities during the 1980s and 1990s. Using a unique data file on air pollution con-

centration in cities across China, this paper has demonstrated that the Huai River-Qinling Moun-

tains heating policy leads to dramatically higher TSP levels in northern China. This result holds in

a cross-sectional regression discontinuity-style estimation approach and in a panel data setting that

compares pollution concentrations in northern and southern cities in years when they have similar

winter temperatures. In a matched sample of births and TSP readings, we have demonstrated that

these high levels of particulate matter are leading to lower birth weights in China, and that children

who had the misfortune of being in utero during a harsh winter are suffering the consequences of

China’s extensive coal use. In combination with the large body of evidence demonstrating that

lower birth weight may lead to serious health problems later in life, this suggests that the health







24

consequences of China’s air pollution are being felt by the most vulnerable population. The health

consequences of China’s high TSP levels on infant outcomes and the potential enduring impact

may help explain why life expectancy in China has not risen more quickly, in spite of the country’s

rapid economic progress. We investigate the direct impact of long-term exposure to TSP as well,

which reveals a large cost in life expectancy to TSP exposure. This evidence further strengthens

the body of evidence that China’s air pollution is a severe public health problem, and the need for

China to wean itself from its reliance on coal.

Coal remains China’s primary energy source, and it remains a national challenge to switch

to cleaner forms of energy in light of the large economic costs of doing so. In this paper, we

demonstrate the health consequences associated with the choice of coal, and hope to provide further

support for those who are encouraging China to invest in cleaner energy. We have shown birth

weights and adult mortality rates are highly responsive to particulate matter at the high levels

observed in China, and the results further bolster the case for stronger environmental regulation in

China as the country grapples with its environmental problems. While China’s State Environmental

Protection Agency (SEPA) has gained clout in recent years, and air quality in China improved

markedly during the 1990s, there are still challenges in China’s political arena in dealing seriously

with the environmental consequences of the country’s economic growth – and China remains home

to many of the world’s dirtiest cities. Our results represent further evidence supporting the call to

enforce tougher environmental standards in China, and establish regulations that will encourage

individuals and firms to adopt alternative sources of energy.









25

References

[1] Almond, Douglas, Kenneth Y. Chay, and David S. Lee. 2005. “The Costs of Low Birth

Weight”. Quarterly Journal of Economics 120(3):1031-1083.



[2] Almond, Douglas, Joseph J. Doyle, Jr., Amanda E. Kowalski and Heidi Williams. 2008.

“Estimating Marginal Returns to Medical Care: Evidence from At-Risk Newborns.” NBER

Working Paper No. 14522.



[3] Almond, Douglas, Yuyu Chen, Michael Greenstone, and Hongbin Li. 2009. “Winter Heat-

ing or Clean Air? Unintended Impacts of China’s Huai River Policy.” American Economic

Review:Papers and Proceedings, 99(2):184-190.



[4] Bi, Xiaohui, Yinchang Feng, Jianhui Wu, Yuqiu Wang and Tan Zhu. 2007. “Source Appor-

tionment of PM10 in Six Cities of Northern China.” Atmospheric Environment 41(5):903-

912.



[5] Black, Sandra E., Paul J. Devereux, Kjell G. Salvanes. 2007. “From the Cradle to the Labor

Market? The Effect of Birth Weight on Adult Outcomes.” Quarterly Journal of Economics

122(1):4094-39.



[6] British Petroleum. 2009. http://www.bp.com/productlanding.do?categoryId=6929ontentId=7044622.



[7] Chay, Kenneth and Michael Greenstone. 2003. “The Impact of Air Pollution on Infant Mor-

tality: Evidence from Geographic Variation in Pollution Shocks Induced by a Recession.”

Quarterly Journal of Economics 118(3):1121-1167.



[8] Chay, Kenneth and Michael Greenstone. 2005. “Does Air Quality Matter? Evidence from the

Housing Market.” Journal of Political Economy 113(2):376-424.



[9] Cook, Thomas D. and Donald T. Campbell. 1979. Quasi-Experimentation: Design and

Analysis Issues for Field Settings. Boston, MA: Houghton Mifflin.



[10] Cleveland, W. S. 1979. “Robust locally weighted regression and smoothing scatterplots.”

Journal of the American Statistical Association 74, 829–836.



[11] Dab, W, S Medina, P Quénel, Y Le Moullec, A Le Tertre, B Thelot, C Monteil, P Lameloise,

P Pirad, I Momas, R Ferry, B Fesy. 1996. "Short term respiratory health effects of ambient air

pollution: results of the APHEA project in Paris." Journal of Epidemiology and Community

Health 50(1):S42-S46.



[12] Dockery, Douglas, C. Arden Pope, Xiping Xu, John D. Spengler et al. 1993. “An Associa-

tion between Air Pollution and Mortality in 6 US cities.” New England Journal of Medicine

329(24):1753-1759.



[13] Evans, William M. and Jeanne S. Ringel. 1997. “Can Higher Cigarette Taxes Improve Birth

Outcomes?” NBER Working Paper No. 5998.







26

[14] Fan, Shuxian, Youfei Zheng, Guoxing Jin and Shigong Chen. 2004. “The Physical and Chem-

ical Studies of TSP in Yinchuan City.” Journal of Nanjing Institute of Meteorology 27(1):20-

28. (In Chinese)



[15] Greenstone, Michael and Justin Gallagher. 2008. “Does Hazardous Waste matter? Evidence

from the Housing Market and the Superfund Program.” Quarterly Journal of Economics

123(3):951-1003.



[16] He, Kebin, Fumo Yang, Yongliang Ma, Qiang Zhang, Xiaohong Yao, Chak K. Chan, Steven

Cadle, Tai Chan and Patricia Malawa. 2001. “The Characteristics of PM2:5 in Beijing, China.”

Atmospheric Environment 35(29):4959-4970.



[17] Hong, Yun-Chul, Jong-Tae Lee, Ho Kim, and Ho-Jang Kwon. 2002. "Air Pollution a New

Risk Factor in Ischemic Stroke Mortality." Stroke 33:2165-2169.



[18] Hong, Yun-Chul, Jong-Tae Lee, Ho Kim, Eun-Hee Ha, Joel Schwartz, and David C. Chris-

tiani. 2002. "Effects of Air Pollutants on Acute Stroke Mortality." Environmental Health

Perspectives 110:187-191.



[19] Jiang, Yi. 2007. “Promoting Chinese Energy Efficiency.” China and the World Discuss the

Environment, working paper.



[20] Kabir Z, Bennett K, Clancy L. 2007. "Lung cancer and urban air-pollution in Dublin: a

temporal association?" Irish Medical Journal 100(2):367-369.



[21] Katsouyanni, K. 1997. “Short term effects of ambient sulphur dioxide and particulate matter

on mortality in 12 European cities: results from time series data from the APHEA project.”

British Medical Journal 314:1658.



[22] Laden, Francine, Joel Schwartz, Frank E Speizer, Douglas W. Dockery. 2009. "Reduction

in fine particulate air pollution and mortality: extened follow-up of the Harvard Six Cities

Study." American Journal of Respiratory and Critical Care Medicine 173:667-672.



[23] Pope, C. Arden, Richard T. Burnett, Michael J. Thun et al. 2002. “Lung Cancer, Cardiopul-

monary Mortality, and Long-term Exposure to Fine Particulate Air Pollution,” Journal of the

American Medical Association 287(9):1132-1141.



[24] Pope, C. Arden, Majid Ezzati, and Douglas Dockery. 2009. “Fine Particulate Air Pollution

and Life Expectancy in the United States.” New England Journal of Medicine 360(4):376-

386.



[25] Qiu, Jinhuan and Liquan Yang. 2000. “Variation characteristics of atmospheric aerosol op-

tical depths and visibility in North China during 1980-1994.” Atmospheric Environment

34(4):603-609.



[26] Rabl, Ari. 2005. "Air Pollution Mortality: Harvesting And Loss Of Life Expectancy", Journal

of Toxicology and Environmental Health, Part A, 68:13, 1175-1180.







27

[27] Rosenzweig, Mark and Junsen Zhang. 2009. “Do Population Control Policies Induce More

Human Capital Investment? Twins, Birth Weight, and China’s ‘One Child Policy’,” Review

of Economic Studies, 76:1149-74.



[28] “Special Report: A Great Wall of Waste - China’s Environment.” The Economist. London:

August 21, 2004. 372(8389):64.



[29] Tianshan Webpage. 2006. "90% of the Pollutants Are from Resident Heating in Wurumuqi

City.” December 25. (In Chinese)



[30] United States Environmental Protection Agency. 2007. http://www.epa.gov/ttn/catc/cica/airq_e.html#3.

Downloaded September 30th, 2009.



[31] Wang, Xiaobin, Hui Ding, Louise Ryan, Xiping Xu, 1997. “Association between air pollu-

tion and low birth weight: a community-based study.” Environmental Health Perspectives

105(5):514–520.



[32] Wang, T.J., L.S. Jin, Z.K. Li and K.S. Lam. 2000. “A modeling study on acid rain and recom-

mended emission control strategies in China.” Atmospheric Environment 34(26):4467-4477.



[33] Wilcox, Allen J. 1993. “Birth Weight and Perinatal Mortality: The Effect of Maternal Smok-

ing.” American Journal of Epidemiology 137(10):1098-1104.



[34] World Bank. 2007. “Cost of Pollution in China: Economic Estimates of Physical Damages.”

Conference Edition, The World Bank, Washington, D.C.



[35] Yang, Gonghuan. 1997. “1996 National Prevalence Survey on Smoking Patterns.” China Sci-

ence and Technology Press.









28

Table 1

Sample Means among Births in Air Quality and Natality Sample (1981-1991)

Statistic North South Differences

(1) (2) (3)

Total Suspended Particulates (µg/m3) 714.6 324.7 389.87***

Birth weight (grams) 3,286 3,192 93.25***

Sex (Male=1) 0.52 0.52 0.01

Mother's Age 33.29 33.36 -0.07

Log of Household Income 7.07 7.21 -0.14***

Mother's Years of Education 8.98 8.83 0.16***

Latitude 39.59 29.56 10.03***

Year of Birth 1985.63 1985.66 -0.04

Month of Birth 6.21 6.00 0.21***

o

Winter Temperature ( F) 27.57 47.16 -19.59

Sample Sizes Total

Births between 23o- 47o (ALL) 17,592 6,496 24,088

Births between 27o- 42o 13,722 5,387 19,109

Births between 28o- 40o 12,070 5,387 17,457

Births between 29o- 37o 2,576 4,934 7,510

o o

Births between 30 - 35 1,540 4,926 6,466



Source : China Environmental Yearbooks (1982-1992), China Meteorological

Administration (1981-1991), National Sample Survey on the Status of Children (1992)



Note : The sample is all births in cities where air quality data is available from China's Air

Quality Monitoring Sites, and infant outcome data is available from the UNICEF fertility

survey. China is divided into North and South by a line formed by the Huai River and

Qingling mountain range (approximately 33o latitude). Winter temperature is the average

monthly temperature recorded for November, December, January, and February. The TSP

level for each birth is calculated as the weighted average of the annual TSP reading for year

of the child's birth and the preceding birth, with the weights determined by birth month.

Table 2

Estimated Effect of Home Heating Policy on Total Suspended Particulates (TSP)

(1) (2) (3) (4) (5)

Geographic Sample All 27o- 42o 28o- 40o 29o- 37o 30o- 35o



Panel A: Quasi Regression Discontinuity using North-South Border

533.5*** 597.6*** 714.9*** 424.60 324.20

1(North)

(112.20) (134.00) (238.20) (300.70) (409.90)

Cubic Latitude Y Y Y Y Y

Demographic Controls Y Y Y Y Y

Year FE Y Y Y Y Y

Month FE Y Y Y Y Y

City FE N N N N N

R Squared 0.52 0.59 0.60 0.67 0.68

N 22,164 17,231 15,579 6,937 5,893

Panel B: City Fixed Effects Strategy exploiting North-South Border and Severity of Winter



1(North) *Winter -73.43*** -93.73*** -98.03*** -53.98** -49.04*

Temperature (oF) (22.60) (22.68) (24.59) (22.52) (23.54)

Demographic Controls Y Y Y Y Y

Year FE Y Y Y Y Y

Month FE Y Y Y Y Y

City FE Y Y Y Y Y

R Squared 0.76 0.79 0.80 0.81 0.81

N 22,164 17,231 15,579 6,937 5,893

* significant at 10% ** significant at 5%. *** significant at 1%.



Source : China Environmental Yearbooks (1982-1992), China Meteorological Administration

(1981-1991), National Sample Survey on the Status of Children (1992)



Note : The dependent variable in each regression is the Total Suspended Particulates assigned

to the birth, as described in Table 1. The variable "1(North)" is a dummy for whether the birth

occurred in a city covered by the home heating policy, which is all cities north of the line

formed by the Huai River and Qingling mountain range (approximately 33o latitude). The

final four specifications restrict the sample to births occurring between the listed latitude

range. Winter temperature is the average monthly temperature recorded for November,

December, January, and February. All models in panel 2 include a control for the winter

temperature in the city and year. The demographic controls are listed in Table 1.

Table 3

Testing the Relationship Between Observable Characteristics of the Birth and the

Instrumental Variables

(1) (2) (3) (4) (5)

Geographic Sample All 27o- 42o 28o- 40o 29o- 37o 30o- 35o

Panel A: Correlation between Observable Characteristics of the Birth and North of the Huai River

Sex (Male=1) 0.010 0.0182 0.0298 -0.0359 0.098

(0.02) (0.02) (0.02) (0.11) (0.11)

Mother's age -0.001 -1.420* -0.889 -0.904 -2.837

(0.40) (0.74) (1.17) (1.01) (1.61)

Log (Income) 0.259*** 0.384** 0.33 -0.608 -1.465***

(0.09) (0.17) (0.24) (0.35) (0.15)

Years of Education 2.020*** 1.946** 2.842** -0.303 -1.307

(0.60) (0.88) (1.10) (2.24) (1.85)

Predicted Birthweight -3.60 -2.04 -4.79 -10.58 -3.69

(4.08) (4.99) (5.64) (19.18) (24.32)

Panel B: Correlation between Observable Characteristics of the Birth Sample and Winter

Temperature Interacted with North of the Huai

Sex (Male=1) -0.009 -0.004 -0.006 -0.004 -0.004

(0.01) (0.01) (0.01) (0.01) (0.01)

Mother's age -0.001 -1.420* -0.889 -0.904 -2.837

(0.40) (0.74) (1.17) (1.01) (1.61)

Log (Income) -0.01 -0.03 -0.06 0.02 0.05

(0.07) (0.07) (0.07) (0.07) (0.06)

Years of Education 0.00 (0.00) (0.00) 0.0123* 0.0146*

(0.00) (0.00) (0.00) (0.01) (0.01)

Predicted Birthweight -0.61 0.11 -0.30 0.51 0.40

(0.85) (0.76) (0.72) (1.27) (1.40)

N 22,163 17,230 15,578 6,936 5,892

* significant at 10% ** significant at 5%. *** significant at 1%.

Source: China Environmental Yearbooks (1982-1992), China Meteorological Administration (1981-

1991), National Sample Survey on the Status of Children (1992)

Note : Each cell in this table represents a separate regression. In Panel A, we report the partial

correlation between the demographic characteristics of the birth sample and the first instrument: a

dummy for whether the birth was north of the Huai river. Birthweight is predicted using the other

covariates by OLS. All models include year and month fixed effects, and a cubic in latitude. In

Panel B, we report the partial correlation between the demographic characteristics of the birth

sample and the instrument: a dummy for whether the birth was north of the Huai river interacted

with the winter temperature during the pregnancy. All models include year, month, and city fixed

effects. A control for the winter temperature is included in all results (not shown). Standard errors

are robust and clustered at the city level.

Table 4

OLS and 2SLS, and Reduced Form Models of the Effect of Total Suspended

Particulates (TSP) on Birth Weight (grams)

(1) (2) (3) (4) (5)

Geographic Sample All 27o- 42o 28o- 40o 29o- 37o 30o- 35o

Models without City Fixed Effects

Panel A: OLS Models of TSP impact on Birth Weight (grams)

TSP (µg/m3) -0.032 -0.060 -0.0829* 0.051 -0.048

(0.03) (0.04) (0.05) (0.07) (0.06)



Panel B: 2SLS using North as an Instrument for TSP impact on Birth Weight (grams)

TSP (µg/m3) -0.074 -0.116 -0.060 -0.892 -0.178

(0.10) (0.08) (0.09) (0.58) (0.48)



Models with City Fixed Effects

Panel C: OLS Models of TSP impact on Birth Weight (grams)

TSP (µg/m3) -0.0798*** -0.0917*** -0.0883** -0.04 -0.04

(0.02) (0.03) (0.03) (0.04) (0.04)

o

Panel D: 2SLS using Winter Temperature ( F) Interacted with North as an Instrument for TSP

TSP (µg/m3) -0.143** -0.203*** -0.205*** -0.475 -0.399

(0.07) (0.07) (0.07) (0.33) (0.45)

o

Panel E: Reduced Form of Winter Temperature ( F) Interacted with North as an Instrument for TSP

1(North) *Winter 10.51* 18.99*** 20.08** 25.64** 19.56

o

Temperature ( F) (6.24) (6.67) (7.54) (10.11) (14.38)

N 22,208 17,231 15,579 6,937 5,893

* significant at 10% ** significant at 5%. *** significant at 1%.



Source : China Environmental Yearbooks (1982-1992), China Meteorological Administration

(1981-1991), National Sample Survey on the Status of Children (1992)



Note : The dependent variable in each regression is the birth weight recorded for the birth. The

sample is all births in cities where air quality data is available, and infant outcome data is available

from the NSSSC (1992). The results of the first stage of the 2SLS results are reported in Table 2.

The demographic controls are mother's age, education, household income, and the sex of the child.

All models include birth year and month fixed effects. A main effect for temperature is also

included. Standard errors are robust and clustered at the city level.

Table 5

Robustness Checks and Falsification Exercises: 2SLS Models of the Effect of Total

Suspended Particulates (TSP) on Birth Weight (grams)

First Stage: 2SLS:

TSP=f(North x Temp) Birthweight=f(Fitted TSP)



(1) (2)

Panel A: Actual Sample

(a) Huai River Instrument (base case) -73.43*** -0.143**

(22.60) (0.066)

Panel B: Robustness Checks using Additional Weather Controls

(b) Add Extreme Weather Controls -73.87*** -0.166**

(22.95) (0.069)

(c) Add Summer Temperature Controls -74.11*** -0.150**

(18.10) (0.071)

Panel C: Falsification Tests

o

(d) 25 Latitude -27.35 0.158

(17.81) (0.227)

o

(e) 40 Latitude -19.66 0.180

(21.53) (0.405)

(f) Yangtze River 15.45 -2.132

(12.57) (1.58)

(g) Post-Natal TSP -98.46** -0.095

(41.45) (0.061)

* significant at 10% ** significant at 5%. *** significant at 1%.

Source : China Environmental Yearbooks (1982-1992), China Meteorological Administration

(1981-1991), National Sample Survey on the Status of Children (1992)

Note : Each cell in the table represents the coefficient from a separate regression. In the first

column, the dependent variable in each regression is TSP and the regression coefficient is reported

for the interaction term of the winter temperature and the policy. In the second column, the

regressor shown is the 2SLS coefficient estimate for TSP, using the instrument from column 1. In

the first panel, the regression for the full actual sample is reported. In the second row, dummies are

added for the winter being colder than 40o, and this dummy interacted with being north of the Huai

river. In the third row, a term is added for the average temperature in the city in summer (March-

October) months interacted with a north dummy. In the third panel, falsification exercises are

presented. The first and second rows present results where the policy is presumed to have been

o o

instituded along the border defined by 25 and 40 latitude respectively. The third specification

presumes the policy was instituted along the Yangtze river. The last specification uses post-natal

TSP as the RHS variable instead of pre-natal TSP.

Table 6

Heterogeneity in the Impact of Total Suspended Particulates on Birth Weight

(1) (2) (3)

Panel A: Impact of TSP on Mean Birth Weight for Subsets of Mothers by Education

Lower Education Middle Education Upper Education

(Primary or Less) (Secondary) (High School or more)

TSP (µg/m3) -0.355** -0.107 -0.108

(0.143) (0.095) (0.145)

N 2,878 9,618 9,799

Panel B: Impact of TSP on Probability of having Birth Weight below a Percentile

Bottom 10% Bottom 25% Below Median

(BW<2,700g) (BW<3,000g) (BW<3,250g)

TSP (µg/m3) 0.0799** -0.0125 0.0622

(0.030) (0.083) (0.094)

N 22,295 22,295 22,295

* significant at 10% ** significant at 5%. *** significant at 1%.



Source : China Environmental Yearbooks (1982-1992), China Meteorological

Administration (1981-1991), National Sample Survey on the Status of Children (1992)



Note : Each cell in the table represents the coefficient from a separate regression. All

regressions are 2SLS models using 1(North) * Winter Temperature as the instrument. In

panel A, we estimate the impact of TSP on mean birth weight for different subsamples of

mothers stratified by education. In panel B, we examine the impact of TSP on the

probability of having a birth weight below 2,700 grams (10th percentile), 3,000 grams (25th

percentile), and 3,250 grams (median). The dependent variable in these regressions is a

dummy for having a birth weight lower than the listed percentile. The coefficient in panel B

is reported per 1,000 units of TSP. Low education in panel A referes to mothers with

primary or less, middle education is mothers with more than primary but less than high

school degree, and high education is high school graduates and beyond.

Table 7

Summary Statistics Among Deaths in Air Quality and Mortality Sample (1991-2000)

Adjusted

Difference In Difference

South North Means In Means p-value



(1) (2) (3) (4) (5)

Panel A: Air Pollution Exposure at China's Disease Surveillance Points

Total Suspended 210.96 437.65 226.69*** 195.00*** <.001/<.001

3

Particulates (µg/m ) (169.1) (196.8) (27.59) (48.49)

Panel B: Demographic Characteristics of China's Disease Surveillance Points

Years of Education 4.31 3.98 -0.32 -1.21 0.371/0.144

(2.09) (2.25) (0.36) (0.82)

Share in Farming 0.64 0.64 0.01 0.19 0.925/0.198

(0.39) (0.40) (0.08) (0.15)

Manufacturing 0.09 0.08 -0.01 -0.07 0.698/0.113

(0.14) (0.13) (0.02) (0.05)

Share Han 0.92 0.95 0.03 0.00 0.369/0.923

(0.21) (0.16) (0.03) (0.04)

Rural, Poor 0.22 0.25 0.02 -0.43** 0.797/0.029

(0.42) (0.43) (0.09) (0.19)

Rural, Middle Income 0.34 0.30 -0.04 0.25 0.707/0.315

(0.48) (0.46) (0.11) (0.25)

Rural, High Income 0.21 0.19 -0.02 0.33* 0.843/0.088

(0.41) (0.39) (0.09) (0.19)

Urban 0.23 0.26 0.03 -0.15 0.672/0.333

(0.42) (0.44) (0.08) (0.16)

Panel C: Weather Patterns at the Disease Surveillance Points

Total Heating 2927.4 6145.9 3,218.49*** 281.60 <.001/0.475

Degree Days (000s) (1595.35) (2216.10) (432.40) (393.20)

Total Cooling 2008.4 1144.4 -864.02*** -136.90 <.001/0.522

Degree Days (000s) (864.25) (581.27) (151.80) (213.40)



* significant at 10% ** significant at 5%. *** significant at 1%.

Source : China Disease Surveillance Points (1991-2000), China Environmental Yearbooks (1981-2000),

World Meteorological Association (1980-2000).

Note : N=1,139. The sample is comprised of annual observations of all DSP points within 150 kilometers

of an air quality monitoring station by year for 1991-2000. TSP (µg/m3) in the years 1981-2000 prior to

the DSP period is used to calculate city-specific averages. See the main text for details. Degree days are

the absolute value of the deviation of each day's average temperature from 650 F, averaged over the years

1980-2000 prior to the DSP period. The results in column (4) are adjusted for a cubic in latitude. The

differences in means reported in columns (4) and (5) are weighted by the population at the DSP site. The

standard errors of the differences are clustered at the DSP site level.

Table 8

Impact of Total Suspended Particulates on Cause-Specific Death Rates

Cubic in

Cubic in Latitude &

No Controls Latitude Controls

(1) (2) (3)

3

Panel A: Impact of TSP (µg/m ) on Death Rates by Cause, Ordinary Least Squares

ln(Total Death Rate) 0.0126 0.0189** 0.0187**

(0.010) (0.009) (0.009)

ln(Lung Related Illnesses) 0.0275* 0.0326*** 0.0324***

(0.015) (0.012) (0.012)

ln(Non Lung Related Illnesses) -0.003 0.005 0.003

(0.011) (0.009) (0.008)

3

Panel B: Impact of TSP (µg/m ) on Death Rates by Cause, Instrumental Variables

ln(Total Death Rate) -0.0018 0.105* 0.175*

(0.025) (0.063) (0.095)

ln(Lung Related Illnesses) 0.0382 0.180** 0.265**

(0.033) (0.081) (0.127)

ln(Non Lung Related Illnesses) -0.0610** -0.00717 0.0316

(0.028) (0.054) (0.067)

Panel C: Impact of TSP (µg/m3) on Life Expectancy

Life Expectancy (OLS) (0.2160) -0.558*** -0.415***

(0.138) (0.129) (0.118)

Life Expectancy (IV) 0.536 -1.654* -2.100

(0.359) (0.882) (1.268)



* significant at 10% ** significant at 5%. *** significant at 1%.

Source : China Disease Surveillance Points (1991-2000), China Environmental Yearbooks

(1981-2000), World Meteorological Association (1980-2000).



Note : N=1,139. Each cell in the table represents the coefficient from a separate regression.

The lung-related illnesses are heart disease, stroke, lung cancer and other respiratory

illnesses. The non-lung related illnesses are violence, cancers other than lung, and all other

causes. Models in columns (2) and (3) include a cubic in the latitude of the DSP site.

Models in column (3) include controls for average education, share employed in farming,

county income, and heating and cooling degree days between 1980 and 2000 prior to the

year being analyzed. Degree days are the sum of the difference between the temperature

and 650F. The IV models are estimated using 2SLS with 1(North) as the instrument for

TSP. Coefficients are reported per 100µg/m3 of TSP.

Figure 1

North and South China Denoted by Huai River/Qinling Mountains 0° Celsius Line







Yicun

Hegang

Qitaihe

Daqing

Harbin



ChangchunJilin Tumen

Siping

Fushun Jian

Benxi

Urumqi Anshan





BaotouHohhot Beijing Dalian

Datong Qinghuangdao

Baoding

Jiayuguan Tianjin

Yinchuan Taiyuan

JinanZibo Qingdao

Shizhuishan

Yanan Anyang

Xining Yichang

Jiaozuo Xuzhou

Lanzhou Yuncheng Kaifeng

Baoji Xian Zhengzhou

Nantong

Suzhou

Hanzhong XiangfanLuoyangHefei

Anqing Ningbo

Wuhan Hangzhou

Wanxian Jiujiang

Chengdu Wenzhou

Nanchong Nanchang

Lasha Leshan Chongqing Changsha

Yibin

Zigong Huaihua SanmingFuzhou

Ganzhou

LiupanshuiGuiyang Hengyang Xiamen

Guilin

Kunming Hechi

Baise Wuzhou

Shenzhen

Gejiou Nanning Guangzhou



Zanjiang

Haikou









Note : The cities shown are the locations of the air quality monitoring stations in China. Cities

north of the solid line were covered by the home heating policy.

Figure 2

Mean Total Suspended Particulate Concentration from 1981-1991 by City Latitude

1200

Total Suspended Particulates (mg/m^3)

400 600 200 800 1000









20 25 30 35 40 45

City Latitude (degrees north)





Note : The vertical line is drawn at 33.6°, which is in the middle of the latitude range covered by

the Huai River/ Qinling Mountains line. Each observation is the average TSP reading in a

particular city in our matched sample of air quality and infant outcomes. Infant outcome data are

taken from the UNICEF fertility survey (1992) and TSP readings are taken from China's

Environmental Yearbooks (1982-1992).

Figure 3

Relationship between Winter Temperature and Total Suspended Particulates



South North

400









800

350









750

300









700

250









650









30 40 50 60 70 0 10 20 30 40

Winter Temperature (F)

Total Suspended Particulates Fitted Values





Note : The results above are scatterplots and fitted values from models where the TSP for a

particular birth is predicted by a linear regression using the average winter temperature (in

Fahrenheit) for the birth as the dependent variable. Each observation is the TSP reading and

winter temperature for a birth in our sample, and we plot 5% of the births. The regression is

estimated over all births in which we observe weather and a TSP reading, separately for northern

and southern China. Infant outcome data are taken from the UNICEF fertility survey (1992), TSP

readings are taken from China's Environmental Yearbooks (1982-1992), and weather data is

taken from China's Meteorological Administration data (1981-1991).

Figure 4: Birth Weight by Winter Temperature During Pregnancy

China UNICEF Fertility Survey (1981-1991)

3,450





3,400





3,350

Birth weight (grams)









3,300





3,250





3,200





3,150





3,100 Coefficient of Winter Temperature Bin

Upper (1 standard deviation)

Lower (1 standard deviation)

3,050

5 10 15 20 25 30 35 40 45



Temperature (oF)



Note : The figure above plots the predicted birth weight associated with each temperature bin. The prediction is based on a

regression of birth weight on dummies for the temperature bin, year and city fixed effects, and demographic control

variables (sex of birth, mother's education, household income). Temperature refers to the average monthly temperature

overlapping with the winter during the year of the child's birth.

Figure 5

Long-Term TSP exposure (1981-2000) and Log of Death Rates by Cause



Beta OLS (non-lung) Beta OLS (lung)

.0015 .003









.0015 .003

TSP



0









0

-.003 -.0015









-.003 -.0015

30 40 50 60 70 80 30 40 50 60 70 80







Beta IV (non-lung) Beta IV (lung)

-.015-.01-.005 0 .005 .01 .015









-.015-.01-.005 0 .005 .01 .015

TSP









30 40 50 60 70 80 30 40 50 60 70 80









Note : Each point in the plot represents the coefficient from a separate regression, where we

examine how age-specific death rates are affected by long-term exposure to TSP (µg/m3). The

95% Confidence Intervals are drawn around the point estimates. The regressions are estimated in

the manner described in Table 7, where the log of the death rate of the listed cause is the

dependent variable and long-term TSP average is the independent variable. The 2SLS models

are estimated using 1(North) as the instrument for average TSP.