Air Pollution and Infant Mortality - Has Tokyo Reached a Harmless Level of Air Pollution?

Table of Contents

I.  Introduction  1

II.  Air Pollution and the Impacts on Health  4

1.  Types of Air Pollution  4

2.  The Impact on Health  5

III.  Literature review  6

IV.  Air Quality in Tokyo  11

1.  Background  11

2.  Diesel-Powered Vehicle Control Regulations  12

3  .  15

4.  Post-Diesel Powered Vehicles Control Regulations  16

V.  Methodology  17

VI.  Data Sources  19

VII.  Regression Results  27

1.  Cross-Sectional Estimate Results  27

2.  Fixed-effects and Random-effects Estimates Results  30

VIII.  Discussions and Conclusion  33

  References  37

  Appendices  40

ii

  List of Tables  

Table 1.  Discriptive Statistics  25

Table 2.  Discriptive Statistis with CO  26

Table 3.  Pooled Cross-sectional Estimates (Both w/o CO and with CO)  41

Table 4.  Fixed-effects and Random-effects Estimates  43

Table 5.  Fixed-effects and Random-effects Estimates with CO  43

Table 6.  Fixed-effects and Random-effects Estimates by Grouping  44

Table 7.  Fixed-effects and Random-effects Estimates by Grouping with CO  44

  List of Figures  

  Figure 1. Excise difference between diesel fuel and gasoline  14

  Figure 2. Comparison of emission characteristics by fuel  15

  Figure 3. Transition in the number of LPG and CNG vehicles  16

  Figure 4. Maps of Tokyo  21

  Figure 5. A Trend of Air Pollutants (Mean Value)  23

  Figure 6. A Trend of Infant Mortality  24

  Figure 7. List of Other Studies  45

  Figure 8. Transition of PM and NOx regulations  46

  Figure 9. Ration of gasoline and diesel models among cargo vehicles  47

  Acknowledgements  

  Many thanks to Douglas Miller, my principle advisor on this thesis project, for his support and  

  assistance from discussing the theoretical implications to running the regressions  

  Also thanks to Guido Kuersteiner of UC Davis in economics department for answering my many  

  email questions regarding my econometrics modeling and STATA operations. And also thanks  

  to Y. Hossein Farzin of UC Davis, whose specialty is environmental economics, in agriculture  

  resources economics department for discussing many potential thesis ideas with me  

  Last but not least, thanks to Tsutomu Suzuki from National Institute Environmental Studies,  

  Kazuhiko Takemoto from Ministry of the Environment, and Sayaka Takemoto 1 for helping me to  

  obtain data on air pollution  

1  Sayaka Takemoto is a student of Waseda University in Tokyo whose father is Kazuhiko  

  Takemoto from Ministry of the Environment  

iii

I. Introduction

Among many of environmental issues we have today such as water pollution, nuclear waste, resources depletion and many others, air pollution is widely discussed publicly by people these days. One of many discussions about air pollution is the problems of increasing level of emission from automobiles releasing tremendous amount of carbon dioxide and others pollutants. Many countries in the world have been trying to reduce the amount of air pollution in order to provide cleaner air to its citizens. From current actions towards these issues taken by many countries in the world, people are aware that the air quality has been worsened and causing various health problems. However, the magnitude of risks to our health is not considered as much. Depends on the condition of individuals health, some are more vulnerable to possible adverse effects of air pollution. Air pollution can be severely dangerous and even can lead to a fatal disease or death.

The effects of air pollution on mortality have been intensively studies since the London pollution episodes in the 1950s (Logan, 1953). Since then, many of epidemiologic studies show a variety of severe effects among adults and children exposed to current levels of air pollution (Pope, 1989, Seaton, et al, 1995). Most studies of air pollution have been conducted in locations where the level of pollution is high such as China, Mexico, Brazil, US, and so on. In those areas, even before examining the associations, it is expected to find the definite associations between

1

air pollution and infant mortality due to the high level of air pollution. However, not many studies of air pollution-related health have conducted in the area with a moderate level of air pollution such as Tokyo in recent years, particularly since the end of 1980s. Therefore, this paper addresses this issue by examining the impact of air pollution on infant death in Tokyo from 1997 to 2006.

This study is based on a quantitative analysis with comprehensive data of air pollution and infant mortality from Tokyo Metropolitan Government (TMG) and National Institute of Environment Studies (NIE). The air pollution data is consisted of yearly-averaged level of four important air pollutants which many other studies typically considered in their studies, Nitrogen Oxide (NOx), Nitrogen Dioxide (NO 2 ), Suspended Particulate Matter (SPM), and Carbon Monoxide (CO), obtained by 44 districts over 10 years period of time. Many studies use time-series, cross section, or natural experimental methods, but only a few uses multiple methods together. My paper will use panel (time-series and cross section) model to examine changes in air pollution across Tokyo over time. Tokyo is broken into relatively small districts, which allow me to apply the similar methodology as ones of leading air pollution-infant mortality literature by Currie and Neidell (2004) and Chay and Greenston (2003). The comprehensive analysis of the associations is considered by applying various econometric models such as cross section with and without time effects, and fixed- and random-effects. Detailed description of my methodology is explained in Section V. As similar to study of Currie and Neidell (2004), my paper offers several innovations over the existing literatures. First, many studies only examine a few pollutants because of data limitations, while I examine four pollutants mentioned above that are commonly monitored in Japan. Thus it may allow me to analyze which pollutants are more harmful to infants.

2

Second, since Tokyo is believed to be relatively cleaner city than other metropolitan cities in the world, examining the associations between air pollution and infant mortality may give a signal to the existing literatures that even with a moderate level of air pollution, air pollution are still harmful to our health. From my knowledge, my paper would be the first study to answer this important question.

Third, many studies have looked at the effect of air pollution on health of adults more than that of infants. Recent epidemiologic studies point out that infant has unique characteristics that are clearer to show the association with air pollution than adults. In the case of infant death the link between cause and effect is immediate, whereas for adults, diseases today may reflect pollution exposure that occurred many years ago (Currie and Neidell, 2004). Also, Tokyo is a city where enormous numbers of people commute every day from neighboring cities. People are coming from different area with different level of air pollution. Therefore, it is hard to find the direct association between air pollution and deaths by focusing on adults death. Therefore, looking at infants death rather than adults would be able to give me better and possibly more accurate associations between air pollution and health.

Overall, my finding supports the conclusions that air pollution has possible causal effects on infant mortality. The findings are less compelling than the existing literatures because the estimated results are very sensitive to specifications. Yet, the cross-sectional estimates imply that reduction in NOx over the time interval in my sample saved roughly 500 infants lives and the panel fixed-effects estimates suggests that reduction in SPM saved about 200 infants in Tokyo. The rest of paper is laid out as follows: Section II provides necessary background information about air pollution and the way in which pollution may affect health. Section III introduces previous literatures and Section IV provides background information about the air

3

quality in Tokyo. Section V describes methods of my studies while my data are described in Section VI. Section VII offers results, and Section VIII ends with discussion and conclusions.

II. Air Pollution and the Impacts on Health

Before getting into any examinations on air pollution and infant mortality, I discuss in this section the general information about air pollution and its possible impacts on our heath. Types of Air Pollution

Air pollution is made up of many kinds of gases, droplets and particles that reduce the quality of the air, and can be found in both urban and rural settings. In the city, cars, buses and airplanes, as well as industry and construction may cause air pollution. While in the country, this pollution is caused by dust from tractors plowing fields, trucks and cars driving on dirt or gravel roads, rock quarries and smoke from wood and crop fires. Among pollution, there are several main types of and various well-known effects. These adverse effects include smog, acid rain, and the greenhouse effect, each of which have serious implications for our health and well-being as well as for the whole environment.

First, we can categorize air pollution into two groups; outdoor and indoor pollution each of which within contain various subgroups. In outdoor pollution, there are two main types; the release of particles and noxious gas. Particles are released into the air from burning fuel for energy. A good example of this particulate matter is diesel smoke, often referred it as PM. These particles are very small pieces of matter measuring about 2.5 microns or about .0001 inches (What is Air Pollution). Another type of outdoor pollution is the release of noxious gases, such as Sulfur dioxide (SO 2 ), CO, NOx, and chemical vapors. These gases are the main sources of environmental issue such as smog, acid rain, and ground-level ozone by chemical reactions once

4

they are in the atmosphere. Furthermore, pollution can also be considered inside our homes, offices, and schools.

We work, study, eat, drink and sleep in enclosed environments where air circulation may be restricted. For these reasons, some experts feel that more people suffer from the effects of indoor air pollution than outdoor pollution some of pollutants that I mentioned above can be created by indoor activities such as tobacco smoke, cooking and heating appliances, and vapors from building materials, paints, furniture. In the United States, we spend about 80-90% of our time inside buildings, and so our exposure to harmful indoor pollutants can be serious. (Indoor Air Pollution) It is therefore important to consider both indoor and outdoor air pollution. The negative effects of air pollution to environments are commonly known and discussed widely; however, many implications for our health are less known. Now let us discuss the impact of air pollution on our health.

The Impact on Health

Air pollution can affect our health in many ways with both short-term and long-term effects. Different groups of individuals are affected by air pollution in different ways. Some individuals are much more sensitive to pollutants than others. Young children and elderly people often suffer more from the effects of air pollution. People with health problems such as asthma, heart and lung disease may also suffer more when the air is polluted. The extent to which an individual is harmed by air pollution usually depends on the total exposure to the damaging chemicals, i.e., the duration of exposure and the concentration of the chemicals must be taken into account.

5

Examples of short-term effects include irritation to the eyes, nose and throat, and upper respiratory infections such as bronchitis and pneumonia (Health Effects). Other symptoms can include headaches, nausea, and allergic reactions. Short-term air pollution can aggravate the medical conditions of individuals with asthma and emphysema. Extreme example of short-term effects is the great "Smog Disaster" in London in 1952. Four thousand people died in a few days due to the high concentrations of pollution (Health Effects). Long-term health effects can include chronic respiratory disease, lung cancer, heart disease, and even damage to the brain, nerves, liver, or kidneys (Health Effects). Continual exposure to air pollution affects the lungs of growing children and may aggravate or complicate medical conditions in the elderly. It is estimated that half a million people die prematurely every year in the United States as a result of smoking cigarettes (Health Effects). In any case, both short and long-term effects can bring out serious health problems. It may be already everyones common knowledge that air pollution causes various adverse effects to our health. In the next section, to make these consensuses certain, I summarize many studies that empirically show the associations between air pollution and infants health.

III. Literature review

Epidemiological studies have shown an association between air pollution and not only exacerbate of illness in people with respiratory diseases but also rise in the numbers of deaths from cardiovascular and respiratory diseases. Different studies have attempted to investigate the effect of air pollution on human health from different angles. Looking at the changes in mortality rate on adults and infant, numbers of respiratory diseases hospital admissions and postneonatal health conditions such as its birth weight seem to be the common methods. Many studies have

6

looked at the effect of air pollution on health of adults more than that of infants. Most studies of air pollution have agreed upon that. However, in recent years, the effect on infant or children have been considered as much as adults in many studies. In fact, focusing on infants have become dominant ways of looking at the effects of air pollution because those studies that look specifically at infant mortality all agree that infant has unique characteristics that are clearer to show the association with air pollution than adults. In the case of infant death the link between cause and effect is immediate, whereas for adults, diseases today may reflect pollution exposure that occurred many years ago (Currie and Neidell, 2004). However, depends on the location of studies that are conducted, finding the definite associations between air pollution and infant mortality can still be difficult especially in developing countries. Many of studies of air pollution and mortality had been conducted in North America and Western European cities, while most of worlds children live in developing countries in extreme poverty and under conditions of health and environment quality distinctly different from those in high income countries. A few recent studies suggest that air pollution may pose greater risks for children in less affluent industrialized countries (Loomies et al, 1999). Studies in Brazil, Mexico, Taiwan, and the Czech Republic have reported associations between infant or child mortality and various measures of air pollution. It is plausible that children in less affluent countries, and poor, urban children, in particular, may be especially sensitive to the effects of air pollution. Numerous countries in Latin America, Asia, and Eastern Europe have undergone rapid industrialization without having fully resolved the environmental hazards of earlier stages of development. Children whose health is already compromised by infection and other conditions resulting from poor sanitation, inadequate nutrition, and deficient housing may be more likely to suffer acute, life-threatening events when the insult of air pollution is added. If they experience a life-

7

threatening event, these children may be more likely to die from lack of rapid, high-quality medical attention. Poverty and living conditions may interact with air pollution to exacerbate the risk of mortality among urban children exposed to high levels of air pollution. Therefore, when I attempt to find out the association between air pollution and infant mortality, other factors come into place significantly so that results may not be able to clearly show the direct and definite associations. However, children in Tokyo are relatively healthier than those in less affluent countries. Therefore, it may be somewhat easier to find the definite associations between air pollution and infant mortality. Hence, if my study appears to show the association, then it can add more confidences to those existing literatures.

Here are many different types of studies that look at the associations between air pollution and infants health. The summary tables of these studies are shown in Figure 7 in appendix. A time-series study of sudden infant death syndrome in Taiwan by Knobel, Chien-Jen, and Liang (1995; 96), suggested an association with visibility, a surrogate measure of air pollution that largely reflects fine particles. Another study in Sao Paulo, Brazil, by Saldiva at el (1994), also using a daily time-series design, showed associations of total and respiratory mortality among children under 5 years old with NOx and ozone, but not with inhalable particles or SO2. In an earlier time-series study of Mexico City in the years 1990 to 1992, by Loomis et al (1999), they observed patterns suggestive of an association of peak ambient ozone lagged 3 to 4 days with mortality among children under 5 years old, but in that study they did not examine other air pollution indices separately from ozone. Other studies based on analysis of geographical patterns of pollution and infant mortality have yielded additional supporting results. An ecologic analysis in Rio de Janeiro by Penna and Duchiade (1991) suggested an association between infant deaths from pneumonia and total suspended particles, but did not examine other air

8

pollutants. Annual infant mortality rates in the Czech Republic by Bobak and Leon (1999) were associated with annual average concentrations of inhalable particles, SO2, and NOx in an ecologic study based on 45 administrative districts. The effect of particles was observed for neonatal morality, total postneonatal mortality, and neonatal respiratory mortality, while the other pollutants were associated primarily with neonatal respiratory mortality. A study by Woodruff, Grillo, and Schoendorf (1997), using data from 86 cities in the United States reported an association of postneonatal infant mortality with the mean level of inhalable particles in the first 2 months of life. This latter study is important because it is the first to show that infant mortality may be associated with present-day air pollution in an affluent country with an established air pollution control program.

In general, ecologic studies are more susceptible than time-series designs to confounding by geographical differences in nonpollution factors that affect mortality, such as social class and medical attention; because they use average community- or neighborhood-level pollution levels as an estimate of chronic exposure to pollutants (Loomies et al, 1999). In contrast, time-series studies consider day-to-day changes in pollution within a geographic area, greatly reducing the potential for confounding by in essence allowing a single area to serve as its own control. This existing epidemiologic literature falls short of linking excess child deaths with a specific pollutant. The London air pollution episodes when excess infant deaths were observed in the past were characterized by high levels of particles and sulfates and some recent studies also point toward these pollutants, together or separately. Other studies suggest associations with NOx, SO2, or ozone. Few studies, however, have examined multiple pollutants simultaneously except the study of Currie and Neidell, and Lipfert, Zhang, and Wyzga.

9

In addition, some studies have looked at the exogenous shocks that might have altered the level of air pollution (Chay and Greenstone, Pope, Pope, Schwartz, and Ransom) such as economic shocks and environmental regulations and studied the before-after effects. In their analysis, they were able to conclude that those shocks had a significantly reduce the level of air pollution. For my analysis, No Diesel-powered Vehicle Campaign, which would be discussed in detail in next section, would be the equivalent shocks as their studies; however, the results from the campaign were not significant enough to be appeared.

Overall, most studies proves that there exists an association between air pollution and infant mortality. Some even shows ones are more harmful than the other. Most existing literatures and studies are conducted at the location where its air pollution level is relatively high. Then, they all seem to want to prove that there indeed exists an association. There are few studies that disprove the existence of associations. It may not make sense to conduct the studies at a location where the level of air pollution is low since the chances of having the association would be low. However, conducting these studies at a location like Tokyo where its air pollution level is moderate might shed more lights on the magnitude of effect of air pollution on infant mortality. If my study does not show the association, it can begin to have a sense of what level of air pollution is harmful or not harmful to our health. Not only to show the association between air pollution and infant mortality, finding out the threshold of concentration would be helpful to policymaker for a sake of setting the efficient regulations. Now before examining any associations between air pollution and infant mortality in Tokyo, it is important to understand the trend in air pollution and characteristics of Tokyo. In next section, I provide the information on an important regulation called Diesel-Powered Vehicle Control Regulations.

10

IV. Air Quality in Tokyo Background

Tokyo is the largest city in the world. An estimated 12.79 million people live in Tokyo with 8.653 million living within Tokyo's 23 wards. ("The Principal Agglomerations of the World.") Thirteen million might not sounds too exciting; however, considering the fact that Tokyo is the capital of Japan where millions of people commute from neighboring cities to work, the population increases to nearly 33 million once we include cities like Yokohama and Kawasaki ² . ("The Principal Agglomerations of the World.") Tokyo is known to have the best public transportation system in the world. Therefore, there must be fewer cars in Tokyo, which would lead to cleaner air quality. However, the numbers of cars in Tokyo are just as many or in fact more than that in other metropolitan cities in the world such as New York City or London. ("Tokyo Environment White Paper 2006.") It is known fact that the one of the main source of air pollution is automobile. Specifically, in Tokyo releasing significant amount of NOx and PM are coming from diesel powered vehicles. Since the late 1960s, various measures including control on factory emission of dust and smoke, have significantly improved the level of air pollution caused by SO2 and CO. As for SPM and NOx, the national government's delay in regulating diesel vehicles, which emit these kinds of pollutants, has kept their average concentrations high, leaving the nation with a low level of compliance to the national environmental quality standards. According to a study by the Environment Ministry, five of the ten air pollution monitoring stations that registered the worst PM concentrations, and 7 of those with the worst NO2 concentrations, were located in Tokyo, indicating that Tokyo's air pollution remains to be in a serious situation. ("Diesel Vehicle Control in Tokyo.")

² Yokohama and Kawasaki are in Kanagawa prefecture.

11

Diesel-Powered Vehicle Control Regulations (DPVC regulations)

Within a type of automobiles, large trucks contribute more to the release of air pollution than other types of automobile. In Tokyo, a number of diesel-powered trucks exceed that of gasoline truck since early 1990s. In order to achieve targeted reduction of NOx and PM, regulating and enforcing the Diesel Vehicle Control is important to the improvement of air quality in Tokyo. Tokyo Metropolitan Government (TMG) has compiled the brochure ³ to indicate the main six failures of national government towards improvement of air quality. Below I briefly summarize each failure.

1. Japan's PM regulation seriously lagging behind that of the U.S. and Europe ⁴ Japan introduced PM regulations in 1994, six years behind the United States and two years behind Europe. Yet, the regulation standard was over 5 times more relaxed than equivalent standards in the west. The current Japanese PM standard (Long-Term Regulation introduced in 1998) has only just caught up with the level western standards reached in the early 90s, a realterm delay of almost ten years. This national government's delay in introducing appropriate regulations has allowed diesel vehicles spreading a massive amount of PM to stay on roads throughout Japan.

2. Failure to ensure early distribution of "low sulfur diesel fuel", essential for PM reduction Low-sulfur diesel fuel is essential to ensure effective functioning of PM reducing devices (Diesel Particulate Filter or DPF). While EU, in preparation for a tight diesel emission control to go into effect in 2005, set out the schedule of introducing low sulfur diesel fuel in December 1998, around the same time, the Japanese Central Environment Council failed to even define the target year for introducing low sulfur diesel fuel in its set of recommendations to the government. EU

³ http://www2.kankyo.metro.tokyo.jp/jidousya/six-result/index.htm (In Japanese)

⁴ Graph for transactions of NOx and PM are in Figure 8 in Appendix.

12

countries have even provided tax incentives and other measures for low sulfur diesel fuel, thereby achieving its early distribution ahead of target years defined in their respective regulations. On the other hand, Japan has so far failed to even attempt to take such measures.

3. Turning its back on the primary source of air pollution -.

Many diesel vehicles, manufactured during the days of insufficient emission control, still remain on the road today, continuously spreading a large amount of PM and NOx. Action on such vehicles is an urgent task in order to improve air pollution in Tokyo. However, the national government has been reluctant to develop a DPF, which could effectively reduce the amount of PM in-use diesel vehicles emit. What's worse, the national vehicle registration system only checks for black exhaust smoke, does not even measure the level of PM / NOx emission, which is crucial in determining whether the vehicle has maintained its engine performance since it was new. In order to prevent diesel emission of existing vehicles from worsening further, it is imperative to improve the current vehicle registration / regular inspection systems.

4. Postponement of the revised Automotive NOxPM Law and letting off old diesel vehicles on the road.

The national government revised laws in 2001 at long last to make PM subject to control in addition to NOx released from in-use diesel vehicles. However, despite TMG's repeated opposition, it delayed the revision's enforcement by up to two and a half years from originally planned. The move consequently allowed approximately one million old diesel cargo trucks, not subject to PM control, to continue to drive in three major cities, releasing a massive amount of PM in their paths. This doubled the number of vehicles subject to TMG's diesel vehicle control from 94,000 to 202,000.

13

5. Preferential excise on diesel fuel increase diesel-powered vehicles ⁵ For compact and mid-sized vehicles, diesel models are more popular than petrol-fueled models because, although manufacturing costs are roughly the same between diesel fuel and gasoline,

prices", i.e. cheaper diesel fuel and more expensive gasoline. Figure 1 demonstrates the price differences in detail. The popularity and the expansion in fuel price differences increased the number of diesel vehicles. In view of the trend, TMG has consistently demanded that the preferential excise system be rectified, but the national government has not even explored any possibility.

Figure 1

6. Lack of action on "illicit diesel fuel", which facilitates serious tax evasion and threatens the health of Tokyo citizens.

"Illicit diesel fuel", prepared by mixing kerosene, diesel fuel, etc. with heavy oil, provides a hotbed for tax evasion, emits more PM and NOx than usual, and generates, in its manufacturing process, a substance called "sulfate pitch" that threatens health and the environment. TMG has actively cracked down on illicit diesel fuel, but the national government has taken almost no

⁵ Graph of Ratio of gasoline and diesel models among cargo vehicles is in Figure 9 in Appendix

14

action, leaving it to circulate freely. The national government must take fundamental measures, e.g. introducing a legal ban on the manufacturing of illicit diesel fuel.

TMGs Achievements

Those are the six failures of national government towards air quality measures. On the other hand, TMG has been taking initiative before the national government to reduce the amount of NOx and PM by correcting those failures of national government. I will not discuss all of TMGs achievements in detail here; however, there are two that are worth mentioning. First is, as

of October 1 ^st , 2003, TMG implemented and

enforced the ban of diesel vehicles that do not meet PM emission standards in Tokyo, Saitama, Chiba, and Kanagawa. Those four neighboring prefectures and four others have cooperated together to bring about this successful diesel vehicles control. The diesel vehicle control targeting 34 million

people in the Greater Metropolitan Area is an unprecedented initiative; unseen anywhere else in the world. The joint action of eight major prefectures and cities provides a model for pioneering environmental administration under the initiative of local governments. Second is promoting the proliferation of clean and low-pollution trucks. Starting 1999, as part of the "Say No to Diesel Vehicles" campaign, TMG has advocated replacing diesel vehicles with low-pollution vehicles. In 2000, the "New Market Creation Strategy Council" was established together with auto manufacturers, gas station operators and corporate diesel users in a bid to promote the proliferation of extremely low-pollution vehicles powered with LPG and CNG. Figure 2

15

demonstrates the significant differences of level of emissions among these vehicles. As a result, the number of LPG vehicles in Tokyo has increased 2.3 times to 4,000, while that of CNG vehicles has surged 13 folds to 3,100 over the past four years. Figure 3 shows the increase trend of LPG and CNG powered vehicles since 1999. It is apparent that Say No to Diesel Vehicles Campaign had a successful contribution to a shift of vehicles.

Post-Diesel Powered Vehicles Control Regulations

The air quality improved remarkably and steadily due to the compliance with environment standards for SPM and NO 2 in fiscal 2004, according to an announcement on June 24, 2005 by the TMG. The SPM level met the standard ⁶ at 33 out of 34 Roadside Air Pollution Monitoring Stations (RAPMS ⁷ ), as well as all of 47 Ambient Air Pollution Monitoring Stations

⁶ The Basic Environmental Law defines air quality environmental standards as those levels desirable for the protection of human health and conservation of living environment. For SPM, daily average of hourly values shall not be more than 0.10mg/cubic meter, and no single hourly value shall be more than 0.20mg/cubic meter. For NOx, daily average of hourly values shall be in a range between 0.04ppm and 0.06ppm, or less.

⁷ RAPMS refer to measuring stations set up along roads where large amounts of automobile emissions are present.

16

(AAPMS ⁸ ) in residential areas. The achievement rates in fiscal 2002 and 2003 were 0 and 12 percent for RAPMS, and 40 and 51 percent for AAPMS, respectively. Even at one station ⁹ where the level did not meet the standard this time, a remarkable improvement was seen. The NO2 level met the standard at 16 RAPMSs (47 percent) and all of the AAPMSs. The achievement rates in fiscal 2002 and 2003 were 37 and 53 percent for RAPMS, 93 and 98 percent for AAPMS, respectively. Since the No Diesel Vehicle Campaign began in Tokyo in August 1999, the TMG has been taking many pioneering initiatives in various fields to prevent air pollution in Tokyo, calling for cooperation of the national government and related industry groups. Tokyo attributes the above results to its diesel vehicle emission control regulations enforced in October 2003.

Due to the slow action of national government, the level of air pollution might have been even higher than other metropolitan cities in the world. However, as diesel-powered vehicles control regulations being strictly enforced, the dramatic improvements were seen.

V. Methodology

Now, I discuss the econometric models used to estimates the air pollution and infant mortality associations. I will use the methodology of Chay and Greenston (2003) as my role model and make adjustment so that it will fit into my data. Their analysis is predominantly crosssectional, but I would consider time-series and cross-sectional in order to take an advantages of my long panel data. A general cross-sectional model of infant mortality can be written as y t = fx ⁿ t , w t + t (1)

⁸ AAPMS refer to measuring stations set up in residential areas.

⁹ The one station is at Matsubara Bashi measuring station in Ota-ku

17

where y t is the infant mortality rate in district in year , x ⁿ t is the average air pollutants reading across all monitors in the districts ( n represents the air pollutant that are considered in a specification), w t is a vector of the other determinants of district-level infant mortality rates (such as district geographical characteristics and year dummies, etc), and is nonparametric conditional mean function of y t . Thus, and w t are defined so that [ t | x t, w t ] = 0. Empirical implementation of equation (1) is not practical for two reasons. First, w t could be of extremely high dimension, and many of its elements may not be observable to the researcher (e.g., maternal behaviors). Second, is a general conditional expectation that is a function of the joint distribution of y t , x ⁿ t , and w t . Consequently, estimation of the effects of interest is

A common approach to reducing the dimensionality of the problem is to assume that the (2) effects of the covariates are additive and linear. This results in the linear regression model:

y t = x ⁿ t n + w t + t , t = + u ^t This model would be my first approach to the analysis. The coefficient n is the true effects of air pollutants on infant mortality. For consistent estimation, the least squares estimator of n requires that E [x ⁿ t t ] = 0. If there are omitted permanent ( t ) or transitory (u t ) factors that covary with air pollutants levels, then the cross-sectional estimator will be biased. Any nonlinearities or interactions in the true effects will also lead to bias. The ideal research design is a controlled experiment in which different levels of pollution exposure are randomly assigned across mothers/infants. Random assignment ensures that variation in air pollution exposure is independent of other factors (i.e., w it ) that impact infant mortality. This reduces the dimensionality of the inference problem. However, using randomized clinical trials to study the

18

My next approach would be considering a model using a fixed-effects estimator applied to the pooled ten years of data, then compared that to cross-sectional estimator. . dy jt = dx 1 jt 1 + dx 2 jt 2 + + dx n jt n + d jt , d jt = sj + du jt (3)

, where si are district-fixed effects in infant mortality changes. Since (3) is based on firstdifferences, any permanent unobserved differences across districts, t , is controlled for. Then, I would consider a model using a random-effects estimator. j 1 + dx jt 2 + v j + jt y jt = + X (4)

Before proceeding, I note the potential for censoring bias in my estimates of the effects of air pollutants. The analysis is based on the population of live births. Since air pollution may damage the fetus before birth, it may also affect the likelihood of a miscarriage or stillbirth. In this case, our analysis will understate the impact of air pollutants on infant mortality by conditioning on fetuses that survive to a live birth. As Chay and Greenstone (2003), ideally I would like to consider Quasi-experiments with Say No to Diesel Vehicles Campaign; however, with limitations of controlled group data, I decided to forgo the model.

VI. Data Sources

To implement my evaluation strategy, I obtained data for 23 wards and 21 cities in Tokyo ¹⁰ on 1997 2006 period (once per a year), the most recent period for which the data were available. I combined infant mortality ¹¹ data from TMG Bureau of General Affairs with air pollution data from National Institute for Environmental Studies (NIE). Statistical information on

¹⁰ Tokyo is consisted of 23 wards, 26 cities, 5 towns, and 8 villages http://www.metro.tokyo.jp/ENGLISH/LINKS/links1.htm

¹¹ Infant mortality data are calculated by numbers of infant death divided by live birth in the same district and year.

19

infant death was obtained from vital statistics in Tokyo Statistical Yearbook. Infant is defined as a baby of between a birth and 1 year of age and new born as a baby of less than 4 weeks since birth. Infants born in some cities and towns were excluded from the analysis because for those areas air pollution data were not available. The vital statistics contained, not only the data on infant mortality, but also on overall birth/ death, stillbirth, total fertility rate, and marriage/divorce. Besides data on infant mortality rate, other earlier studies consider other important variables such as birth weight of the same infant from data, education level of mothers who gave birth to those infant they have data on and so forth. However, within the vital statistics, those information were not available; therefore, I was not able to find those data that match with my infant data. One of the difficulties with data on air pollution is inconsistency of monitoring stations over time. Monitoring stations tend to be located in the most densely populated areas of the districts, and also in those that are most polluted, which might potentially produce biased results, and the location of monitors may also change over time. Hence, in this analysis, I use only those monitors that existed continuously throughout the period. Air pollution data from NIE contained data sets for not only NO2, NOx, SPM, and CO, but also Ox and SOx; however, due to the limitations of monitoring stations, SOx and Ox are excluded from my analysis. NIE monitors the air pollutants at monitoring stations for at least 6,000 hours a year and disclose the year averaged value in their database. They also provide information on each monitoring stations geographical characteristics, which would be included in my analysis.

20

Tokyo. In the map for NOx and CO, their unit of measure is in ppb ¹³ as opposed to ppm which NOx, NO2, and CO in my data are measured in. The east side of Tokyo in the map is where 23 wards are located while west side is where the rest of cities are. As you can see, the levels of air pollution are relative low in the west side of Tokyo in all cases. Since CO was one of two most significant air pollutants in Currie and Neidells study, I decided to make another set of data with NO2, NOx, SPM and CO from 16 wards, 8 cities, and a village. However, a data set for CO is limited compared to the data sets of the rest of pollutants; therefore, I was not able to combine a data of CO to my master database to run regression together. Therefore, I separately run regression with those limited data. Descriptive statistics for the pollution variables are shown in the Table 1 and 2 below. Clearly, from both tables numbers of infant deaths in Tokyo are slightly lower than that of other

¹² I excluded NO2 from the map, since NOx includes NO2.

¹³ ppb is parts per billion.

21

studies. Infant mortality in Currie and Neidells study is 5.5 per 1000 births while mine is 3.17 per 1000 births. From 1997 to 2006, the numbers of births are relatively constant over time while the numbers of infant death varies moderately. Due to the limitations of CO data, there are more variations in both the numbers of birth and infant death overtime in Table 2 than that in Table 1. As far as the level of air pollution concern, the levels of all pollutants are almost the same level as Currie and Neidells. In their summary, they measure its level by ppb andg/m ³ , so it might be hard to compare mine to theirs. For example, for PM, the total mean of Currie and Neidells PM is 39g/m ³ , while mine is also 0.039 mg/m ³ , which is equivalent to 39g/m ³¹⁴ . The level of CO is largely smaller in my data than theirs. This information is interesting since I originally thought that the level of both air pollution and infant mortality would be much lower than that of other countries like US. However, I would still assume that on average the level of air pollution in Tokyo is rather moderate level.

If I look at the table more closely, Table 2 shows more variations in all air pollutants than that in Table 1. CO and SPM seem to vary larger than NOx and NO2 overtime. These two tables do not show the variations across districts; however, it is feasible to assume some variations exist across districts as well. Aside from variations, correlations among air pollutants are presented in both tables. As can be seen, all air pollutants are positively correlated with each others. From Table 1, as expected, NOx and NO2 have the highest correlation due to the way how NOx is calculated. However, from Table2, CO and NOx have a relative higher correlation than that of NOx and NO2. Since correlation coefficients for most air pollutants are higher than 0.5, it is a bit difficult to tell which one is not highly correlated with any other pollutants; however, NO2 seems to be the one with lowest correlations with all the pollutants.

¹⁴ 1g/m ³ = 0.001 mg/m ³ and 1ppb = 0.001ppm.

22

Figure 5. Source: National Institute Environmental Studies, graphed by author

Figure 5 presents the trends of yearly and district-wise averaged concentration of air pollutants from 1997 to 2006 on all the districts. Although the unit scale for CO is a lot bigger than the rest of air pollutants, all of them show decreasing trends in ambient air quality. Besides NO, they all seem to have the similar variations over time. They have relative larger reductions between 1998 and 1999, when No Diesel Vehicle Campaign has started. Considering the fact that most of air pollutants have met its standard level which TMG has set in most of monitoring station, ambient air quality in Tokyo might have reached the level which might be considered as clean. From Currie and Neidells study, the level is defined as a threshold for harmless air pollution level. Therefore, if my analysis ends up showing statistically insignificancy on air pollutants, it might be either due to the quality of data or the threshold level of concentration being already reached. Figure 6 shows the trend of infant mortality rate in Tokyo from 1997 to 2006. Slightly decreasing trends is shown; however, considering the unit scale, decreasing from 0.4% to 0.2% might be better to assume that the

23

change in infant mortality rate is constant over time, which is a problem since I would need some

variations on infant mortality to link with the variations of level of air pollution to show an

association. Therefore, hopefully this 0.2% decrease in infant mortality would give me some

interesting results.

VI. Regression Results

1. Cross-Sectional Estimate Results

I begin by replicating the conventional cross-sectional approach to estimate the association between air pollution and infant mortality across districts in Tokyo. For each cross section from 19972006, Table 3 shows estimates of model (2), where the dependent variables is the infant mortality rate per 100 livebirths. For comparison with previous work, I first estimate pooled cross sectional models for each pollutant separately and also together. In Table 3, mainly there are two parts, the one without CO ¹⁶ (Table 3.1 ~ 3.2) and the other with CO (Table 3.3 ~ 3.4). Then, there is subcategory within these models, which are Standard and Grouping. I define Standard as a normal regression with variables which contains true values of air pollutants. Grouping represents the separation of the level of pollutants concentration. I used data sets of Standard model and break the level of each air pollutants into two groups, low and high, using the information from descriptive statistics about 50 ^th percentile. Then I run new regressions with those dummy variables instead of using the variables that contains the actual concentration level. In addition, as can be seen in Table 3 I also created specifications with year effects, and then compared them with non-year effects specifications. Let us first take a look at Standard case without CO in Table 3.1. In terms of statistical significance of each coefficient, regardless of the sign of the coefficients, they all seem to have relatively large significances. Especially, NOxs coefficients might actually be meaningful. The single pollutant models shown in column (5) ~ (10) of Table 3.1 indicate that only exposure to NOx increases the infant mortality rate regardless of year-effects. Moreover, one ppm increase in NOx, holding other stuff constant, leads to an increase in 2.18 deaths per 100,000 live births ¹⁷ .

¹⁶ As stated in section V. specifications with CO has less districts.

¹⁷ Usually air pollutants increases by 0.01ppm instead of 1ppm; therefore, 1ppm increases in any air pollution leads to X deaths per 100,000 livebirths.

27

On the other hand, from column (7), exposure to SPM reduces the infant mortality rate, which is counterintuitive. Column (1) ~ (4) shows that if I run regression on some combinations of air pollutants, all of their magnitude of estimates becomes larger than that of single pollutants model. The estimated effects are much larger if I include all with year-effects, shown in column (2). Moreover, given the 941,190 births in areas where pollution could be assigned over my sample period, the estimated of 5.23 deaths per 100,000 suggests that the one unit decline in NOx that took place over the sample period saved 47 infant lives. (Note that I do not consider possible live saved in areas without pollution monitors. If these areas did not have monitors because they had little pollution and /or were sparely populated, then reductions in pollution could be expected to have relatively little effect, assumption established by Currie and Neidell) Also, if I consider year-effects for each specification, most of their estimates improve. Same inferences apply to the model with CO; however, the magnitude of those estimates is much larger, shown in Table 3.3. With this specification, reduction of one unit of NOx leads to 50 ¹⁸ infant lives saved in column (10) of Table 3.3.

Now let us look at the same models, but with different angels, looking at Grouping model. This model reveals the statistics about how much of each air pollutants from those districts that have high concentration contribute to infant mortality compared to those with low concentrations. From column (5) ~ (10) in Table 3.2, the effects on infant mortality rate from exposure to NOx in high concentration districts have statistically significant differences from that in low concentration districts. More precisely, high concentration districts have additional 5 deaths per 10,000 livebirths more than that of low concentration districts if I consider year-

¹⁸ Thecoefficient is much larger in Table 3.3, but with CO specifications, total birth is smaller than the one without CO, so the numbers of infant that were saved is not proportionate to the increase in the coefficient.

28

effects. Estimated 5 deaths per 10,000 suggest if a high concentration district converges to low concentration district, it can save 470 infants. Same inference applies to NO 2 case except that for NO 2 , estimates are significant regardless of year-effects. Once again like Standard case, magnitude of all estimates become larger if I run regression with some combinations of pollutants shown in column (1) ~ (4). A specification with all pollutants in column (1) show the largest magnitude, but only SPM and NO 2 are significant. NOx shows significant only with SPM. Once again, SPM has counterintuitive estimated effects, meaning high concentration districts have lower infant mortality rate than that of low concentration districts. With CO specifications shown in Table 3.4, NOx and NO 2 show statistically significances; however, CO shows counterintuitive results and SPM lost its significant that were there in without-CO model. With year-effects model, only exposure to NO 2 in high concentration districts has significant difference from its low concentration districts. Overall, in CO specification many of coefficients lose its significances, which is opposite results from standard regression. The magnitudes of coefficients become smaller than that of Non-CO specifications. To summarize, NOx appears to have the most significant effects on infant mortality. The estimated effect of NOx is remarkably robust to many changes in specification. The magnitude of coefficient becomes larger as being considered with other pollutants. It is quite new results that NOx have the most significant effects on infant mortality compared to other studies. The coefficient on SPM is also somewhat robust to many chances in specifications; however, it has negative effects. Unlike the results from the Currie and Neidell estimates, CO is not as significant as NOx or SPM. Not only CO is not significant, but also the coefficients on CO show negative effects. Also many other studies conclude to say that PM or TSP has a strong associations with infant mortality, while my data fail to support those results. However, it must

29

be kept in mind that the Currie and Neidell and others include many more control variables than mine, which could change the estimate results dramatically. Therefore, it is reasonable to assume that my cross sectional estimates might have many omitted variables that are not considered in the models but might affect the level of infant mortality rate. This raises the problems of biasness of cross sectional estimates. To eliminate this bias, in next section, I consider fixed-effects and random-effects estimates to see how the results changes.

2. Fixed-effects and Random-effects Estimates Results

Table 4 ~ 7 presents fixed-effects (FE columns) and random-effects (RE columns) regression estimates of the effect of air pollutants on the infant mortality rate within a year of birth per 100 live births. All specifications include year-effects, districts-effects, and geographical characteristic dummies but not shown in tables for the sake of limited spaces. As can be seen in Table 4, first all the significances that are there with cross sectional estimates are totally gone with both fixed and random effects. Single pollutant model in column (3) ~ (6) shows the opposite estimated effects, meaning now SPM has positive effect and NOx and NO 2 have negative effects. From column (1), 1 mg/m ³ increase in SPM leads to an increase in 1.5 deaths per 100,000 live births, while 1 ppm increase in NOx leads to a reduction in almost 3 deaths per 100,000. Regardless of its statistically significance, the magnitude of their coefficients is much smaller than cross sectional estimates. With random-effects shown in column (6) ~ (10), all coefficients are now positive regardless of their significant except when I include all pollutants shown in column (6); however, the magnitude of all the estimates are even smaller than that of fixed-effects estimates. Since all the coefficients lose their significant, it is hard for me to say which pollutants have stronger effects on infant mortality. However, what I

30

can say from these results is that the results from cross sectional estimates could highly be biased. Therefore, it is important to control for omitted variables. The estimated results on geographical characteristics ¹⁹ are excluded from the tables; however, in random-effects in Table 4, it shows that residential and commercial districts are relatively healthier than industrial districts as expected. They both show negative effects, meaning if a district is either residential or commercial, it will reduce the infant mortality rate. However, if a district is industrial, then it increases the infant mortality rate. Unfortunately, with random-effects estimates in Table 5 do not show such results.

Even though, all coefficients lost its significance, with fixed-effects I can discuss about the economical significances. From column (1) in Table 4 shows the reductions in SPM and NO ² can save 71 infant lives, which is still relatively large compared to cross-sectional analysis. However, with random-effects, most coefficients get so small that it loose economically significances as well.

Similarly, I run Grouping regression with fixed and random-effects, shown in Table 6 and 7. The interpretations on these results are the same as those on cross sectional estimates. NOx shows statistically significance at 90% level as opposed to 95 or 99%. In Table 6, in both fixed and random effects, only exposure to NOx has the significant different effects on infant mortality rate between high and low concentration districts. The magnitude of coefficient is about the same level as the cross sectional estimates with year effects. Table 6 shows that residential and commercial districts are once again healthier than industrial districts. Table 7 shows the estimated results with the specifications with CO. As can be seen in Table 7, in both fixed and random effects NOx gain more statistically significance than that in Table 6. NOx by

¹⁹ Geographical characteristics dummies get dropped when I run fixed-effects regression due to the multicolinearlity.

31

itself does not gain the significance; however, with all pollutants together, its magnitude increases so that it gains its significance. CO shows its significances for all specifications. However, the estimated effects are negative, which are similar results from cross sectional estimates. When I consider pooled cross section with Grouping, CO lost its significant when I include year-effects. However, with fixed and random effects, CO does not lose its significant at all. Comparing with cross section estimates, coefficients in Table 6 and 7 show that their magnitudes are much larger than that shown in Table 3.4. To summarize, with standard regressions, all coefficients show statistically insignificances, which prevents me from rejecting the null hypothesis that those coefficients are zero, meaning air pollution has no effects on infant mortality. With grouping regressions, estimated results are a slightly better than that of cross sectional estimate. This might proves that although I cannot say much about overall effects of air pollution on infant mortality since the estimated effects for all pollutants are sensitive to specifications and types of regressions, the high concentration districts have slightly higher infant mortality rate than that of low concentration districts. Even though the results from fixed- and random-effects are not statistically significant, in fixed-effects most coefficients maintain the same level of economical significance as cross sectional estimates.

Since the results differ a lot between cross sectional and fixed/random-effects estimates, it is hard for me to decide which results I should believe in. However, fixed/random-effects estimated results might be more believable as many literatures confirm that there exists bias due to omitted variables from cross sectional estimates. The results from fixed/random-effects are somewhat disappointing due to its inconsistency with existing literatures, yet the results might still be meaningful considering the economical significant of its estimated results.

32

VII. Discussions and Conclusion

This paper examined the effects of air pollution on infant mortality in Tokyo, using recent data from 44 districts on time period of 1997~2006. My models are consisted of three econometric models: cross-sectional, fixed effects, and random effects. Using within district level variation in pollution, I am able to control for unobservable fixed characteristics of districts a well as for detailed group of observable time-varying characteristics. (Currie and Neidell, 2003) The estimated results from each model are quite different; however, especially crosssectional and fixed-effects estimates show some economical significance of the associations between air pollution and infant mortality. In my cross section estimates, I find that only NOx increases infant mortality. Unfortunately, I cannot say much about the effect of SPM and CO since the results show the estimated effects of SPM and CO are more sensitive to specifications, and show negative effects on infant mortality. The cross section estimates imply that reduction in NOx over the time interval I study saved about 500 infant lives in Tokyo. However, my preferred model is fixed-effects estimates, which possibly eliminate the biasness of cross section estimates. Fixed-effects estimates show that SPM and NO 2 increase infant mortality, while the estimated effects of NOx and CO show negative effects. Even though fixed-effects estimates do not show statistical significances, yet it can still provide meaningful interpretations as some economical significances exist on some of pollutants. Although the magnitudes are still much a bit lower than cross sectional estimates, I find that reduction in SPM saved about 200 infants in Tokyo within 10 years.

My findings are interesting for two reasons. One is the finding can be useful in terms of policymaking. It can provide policymakers great motivations to establish air pollution related policies. For example, in a case of TMG, according to TMGs Tokyo Environment Master Plan,

33

based on data obtained in fiscal year 2006, the concentrations of NO 2 measured at all 43 AAPMS met the Environmental Quality Standards, while attainment was confirmed at 21 out of 34 RAPMS in the same year. The achievement of the Environmental Quality Standards regarding SPM has been confirmed at all 47 AAPMS and all 34 RAPMS for the first time since monitoring began. On the other hand, according to annual average data, the concentration of NO 2 has remained flat, while that of SPM has shown a downward trend over recent years. This decreasing tend has been clearly observed since 2003 when DPVC regulations took an effect. The Environmental Quality Standards regarding CO and SO 2 has been fulfilled at all monitoring stations; however, the standards regarding Ox have not yet been met at any of the stations. Clearly, DPVC regulations have brought a successful contribution to the air quality of Tokyo. However, DPVC is only targeted towards NOx and SPM; therefore, still some other air pollutants remained unattained in some monitoring stations. TMG believes necessary actions to obtain better air quality for Tokyo is needed. Considering what came out of TMGs effort to obtain cleaner air quality in Tokyo for ensuring safe environment for the residents of Tokyo, TMG has set up several goals established in their Tokyo Environment Master Plan. An environmental standard of a SPM and a NO2 is achieved at ALL monitoring stations by fiscal year 2010 and obtain more stable lower concentration status by 2016. High concentration pollution requires more dramatic improvement by fiscal year 2010. Establish a new standard level for NO 2 and PM ^2.5 Set a photochemical smog warning announcement day to 0 days by 2016. Secondly, my finding can contribute more certainty to the findings of existing academic literatures and also a new possible compelling results. Even though my estimates are a bit disappointing because some of estimates are not consistent with previous studies such as Currie

34

and Neidell (2003) or Chay and Greenstone (2004), yet it is not entirely hopeless. Considering the fact that my estimates are based on the best currently available information, but without many of important control variables which many other studies considered in their studies, the estimated results can be improved. The finding supports that the reduction on NOx and SPM possibly can save roughly 200 ~ 500 infants over my sample time period. This findings would mean a lot to the existing literatures because first not many studies proved that the associations of the level of NOx and infant mortality and second it confirms the finding of Chay and Greenstone and many others that SPM increases infant mortality. If overall statistical significances are a bit stronger and the estimated results are robust to specifications, my findings can say a lot more about the associations between air pollution and infant mortality with certainty. Yet, I believe that my findings are reasonable.

Currently, TMGs concern is the level of PM 2.5. The concentration of PM 2.5 in Tokyo is higher than standards set by the United States and the World Health Organization; however, environmental quality standards have not yet been set in Japan. TMG has called on the national government to set environmental quality standards and carry out pertinent measures as soon as possible. With the objectives of further improving air quality in Tokyo and protecting the health of local residents, conducting a study in the effect of PM 2.5 on health would be required as a next step. Many other epidemiological literatures confirm that PM 2.5 is harmful to our health; however, to determine the appropriate standard level of PM 2.5 , it is important to understand in detail how much PM 2.5 can potentially affects the resident of Tokyo. I conclude that air pollution-related infant mortality is a major public health problem. Particularly, PM 2.5 is much smaller particulate matter than SPM, which can be more harmful to the health. Therefore, the health damage to infants may be greater than my estimated results obtained by SPM. Therefore, based on my

35

estimated results and many other epidemiological literatures, TMG should consider the vulnerability of infant when it comes to setting policies regarding air pollution. Finally, to answer the question of Has Tokyo Reached a Harmless Level of Air Pollutions?, from my estimated results and the successful contribution from the DPVC it is apparent that the level of air pollution has been converging to a harmless level. I cannot say for sure that the level has already reached that level at this moment. However, considering the fact that I was unable to find a strong association between air pollution and infant mortality as existing literature, I may deduce that the level of air pollution in Tokyo might be less harmful than other metropolitan cities and more polluted locations. Yet, further research with more control variable added to regression or on new problematic pollutants in Tokyo such as PM ^2.5 would shed lights to answering the question even more.

36

References

Bobak M, Leon DA. The effect of air pollution on infant mortality appears specific for respiratory causes in the postneonatal period. Epidemiology 1999;10:665-669

Brunekreef, Bert. "Air Pollution Kills Babies...." Epidemiology 10(1999): 661 - 662.

Colin, Cameron A., and Pravin K. Trivedi. Microeconometrics Using Stata. Texas: STATA P, 2009.

Cross-Sectional Time-Series. release 8. Texas, USA: STATA, 2003. Print.

Currie, Janet, and Matthew Neidell. "AIR POLLUTION AND INFANT HEALTH: WHAT The Quarterly Journal of Economics (2005).

Mortality: Evidence from Geographic Variation in Pollution Shocks Induced

1167.

--- Paper No. 10053, 2003b. Dejmek, J. S.G. Selevan, I. Solansky, RJ. Sram. Fetal Growth and Maternal Exposure to

Particulate Matter During Pregnancy, Environmental Health Perspectives, 107, 1999, 475-480

"Diesel Vehicle Control in Tokyo." Septemer 2003. Bureau of Environment, Tokyo

Metropolitan Government. 12 Apr 2009

.

Hamilton, Lawrence C. Statistics with Stata (Updated for Version 7). New York: Duxbury P, 2001 "Indoor Air Pollution" Indoor and Outdoor Air Pollution. Lawrence Berkeley

National Laboratory. 25 Jan 2009

main.html>. "Health Effects" Indoor and Outdoor Air Pollution. Lawrence Berkeley

National Laboratory. 25 Jan 2009 .

Japan. Bureau of Environment TMG. Environment of Tokyo 2008. Bureau of Environment TMG, 2008. Web. 30 May 2009. . ---

37

Tokyo Environment Master Plan. Bureau of Environment TMG, 2008. Web. 30 May 2009. ---.

Lipfert, F. W., J. Zhang, and R.E. Wyzga. AInfant Mortality and Air Pollution: A Comprehensive Analysis of U.S. Data for 1990, Journal of the Air Waste Management Association, 50(8), August 2000, 1350-1566.

Logan, W.P.D. "Mortality In The London Fog Incident, 1952." The Lancet (1953): 336-37. Loomis, Dana, Margarita Castillejos, Diane Gold, William McDonnell, Victor Hugo Borja-Aburto. Air Pollution and Infant Mortality in Mexico City, Epidemiology Resources, 10 #2,1999, 118-123.

Loomis DP, Borja-Aburto VH, Bangdiwala SI, Shy CM. Ozone and daily mortality in Mexico City: A time-series analysis, Research Report 75. Cambridge, MA: Health Effects Institute, 1996.

Kaiser, Reinhard, Isabelle Romieu, Sylvia Medina, Joel Schwartz, Michal Krzyzanowski, and Nino Kunzli. "Air pollution attributable postneonatal infant mortality in U.S. metropolitan areas: a risk assessment study." Environmental Health:A Global Access Science Source 2004 05 May 2004 26 Jan 2009 .

Knobel HH, Chien-Jen C, Liang K-Y. Sudden Infant Death Syndrome in Relation to Weather and Optimetrically Measured Air Pollution in Taiwan. Pediatrics 1995;96:1106-1110 Pope, C. Arden III. Respiratory Disease Associated with Community Air Pollution and a Steell Mill, Utah Valley, American Journal of Public Health, 79, 1989, 623-628.

Pope, C. Arden, J. Schwartz, and Michael Ransom. Daily Mortality and PM10 Pollution in Utah Valley, Archives of Environmental Health, 47, 1992, 211-216.

Rabe-Hesketh, S. Handbook of statistical analyses using Stata. Boca Raton: Chapman &

Hall/CRC, 2007.

"Research Report on the health effects of Small Particulate Matter in the atmosphere ." Air pollution Research Report in Tokyo. Bureau of Social Welfare and Public Health, Tokyo Metropolitan Government. 6 Mar 2009

. [in Japanese]

Penna M-L F, Duchiade MP. Air Pollution and Infant Mortality from Pneumonia in the Rio de Janeiro Metropolitan Area. Bulletin of PAHO 1991;25:47-54

Saldiva PHN, Lichtenfels AJFC, Paiva PSO, Barone IA, Martins MA, Massad E, Pereira JCR, Xavier VP, Singer JM, Bohm GM. Association between Air Pollution and Mortality Due to Respiratory Diseases in Chil- dren in Sao Paulo, Brazil: A Preliminary Report. Environ Res 1994;65:218- 225

38

Seaton, Anthony et al. Particulate Air Pollution and Acute Health Effects,The Lancet, v354, Jan. 21, 1995, 176-178.

"The Principal Agglomerations of the World." City Population. 30 December 2008. THOMAS BRINKHOFF. 12 Apr 2009

.

"Tokyo Environment White Paper 2006." Tokyo Environment White Paper 2006. September 2003. Bureau of Environment, Tokyo Metropolitan Government. 18 Apr 2009 .

Wang, X., H. Ding, L. Ryan, X. Xu. Associations between Air Pollution and Low Birth Weight: A Community-Based Study, Environmental Health Perspectives, 105, 514-520, 1997.

between Selected Causes of Postneonatal Infant Mortality and Particulate (1997), 608612.

"What is Air Pollution?." Indoor and Outdoor Air Pollution. Lawrence Berkeley National Laboratory. 25 Jan 2009 .

Statistical data

The monthly value and the year value data on Air Quality in Environment numeric database at National Institute Environmental Studies http://www.nies.go.jp/igreen/index.html

Vital Statistics in TOKYO STATISTICAL YEARBOOK at Tokyo Metropolitan Government Bureau of General Affairs, Statistic Division Management and Coordination Section http://www.toukei.metro.tokyo.jp/tnenkan/tn-index.htm

39