Three essays in environmental economics

... Professor Robert Stavins (Chair) Professor Dale Jorgenson Professor Cary Coglianese Fei Yu Three Essays in Environmental Economics This dissertation presents three essays in environmental economics They... establishing and maintaining a comprehensive environmental management system (EMS), (2) going beyond legal requirements by making commitments to continuous environmental improvement, (3) informing... selection criteria for including firms in the study18; estimating normal performance within the estimation window and predicting normal returns during the event window in the absence o f the

HARVARD UNIVERSITY G raduate School o f A rts and Sciences DISSERTATION ACCEPTANCE CERTIFICATE The undersigned, appointed by the Committee on Public Policy have examined a dissertation entitled Three Essays in Environmental Economics presented by FeiYu candidate for the degree of Doctor of Philosophy and hereby certify that it is worthy of acceptance. Signature_______ Typed name: Signature Robert N. Stavins J (_ A y Typed n a m e Dale WHbrgenson Date: 'J L ch jt / ZOO ~f~ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Three Essays in Environmental Economics A dissertation presented by Fei Yu to The Committee on Higher Degrees in Public Policy in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the subject of Public Policy Harvard University Cambridge, Massachusetts July 2007 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3285571 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3285571 Copyright 2008 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. © 2007 - Fei Yu All rights reserved. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Advisors: Professor Robert Stavins (Chair) Professor Dale Jorgenson Professor Cary Coglianese Fei Yu Three Essays in Environmental Economics This dissertation presents three essays in environmental economics. They address issues o f environmental economics from macroeconomic, financial markets and program evaluation perspectives, respectively. The first essay examines whether air pollutants in the US states converge to predictable patterns. US data concerning five types o f air pollutant emissions for the period 1985 to 1999 were analyzed to test whether or not there is convergence in the manner hypothesized by Stokey’s optimal growth model. The empirical tests show evidence supporting conditional convergence o f emissions, where each o f the 51 states converge to different steady states or balanced growth paths. The implied rates o f convergence are in the comparable range with that o f income convergence. The second essay employs an event study method to analyze the stock market response to news o f firms having been awarded membership in the EPA’s National Environmental Performance Track (NEPT) Program. The results indicate that participating firms experienced positive cumulative abnormal returns over ten and fifteen day event windows following release o f news on new memberships. Significant determinants o f cumulative abnormal returns include R&D, total assets and sales. The third essay analyzes the effectiveness o f a recently completed World Bank project in China, where 5,500 rural households were subject to different combinations o f iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. improved stoves and behavioral interventions to reduce indoor air pollution (IAP) exposure. Fixed effects models, random effects models, linear probability models and matching estimators were used in the analysis. There is significant evidence that the interventions were effective in reducing indoor air pollution levels, and in reducing acute respiratory infection (ARI) risks among children under five years o f age. Cost-benefit analysis shows that both the combination o f stove and behavioral interventions and behavioral interventions alone generate health benefits far exceeding the costs. Behavioral interventions alone appear to be more cost effective. iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table of Contents Acknowledgements................................................................................................................ vi Chapter 1. Revisiting the Environmental Kuznets’ Curve: Do Pollution Levels Tend to Converge to Predictable P attern s..............1 Chapter 2. Participation o f Firms in Voluntary Environmental Protection Program: An Analysis o f Corporate Social Responsibility and Capital Market Perform ance................................................................................................... 27 Chapter 3. Measuring Benefits from Interventions to Reduce Indoor Air Pollution in Rural C h in a .................................................................................................... 60 v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgements This has been a long journey, from undergraduate studies at the Dalian University o f Technology to a doctorate from what in the People’s Republic o f China, my home country, is considered the world’s most prestigious university. Each step o f the way, including the Universities o f London and Princeton, has been most challenging and inspiring, culminating in five-years o f intensive study at the Kennedy School o f Government, Harvard University, leading to a Doctor o f Philosophy in Public Policy. Many people have helped me along this journey, most notably my family members (parents and husband) who provided the encouragement and support needed to stay the course, and my professors who provided such wise guidance and inspired me to push at the frontiers o f scholarly endeavor. During my doctoral studies I was blessed by having a baby daughter, who provided much distraction but without whom my life would be much less. Members o f my dissertation committee have been especially helpful in completing this long journey. Professor Robert Stavins was my supervisor from the outset and 1 am very grateful to him for introducing me to environmental economics and helping to chart a course o f studies and research o f such deep interest. I am also very grateful to Professor Stavins for providing career guidance and for taking the time and effort to ensure that I get well started in the next chapter o f my professional endeavors. Professor Dale Jorgenson’s stature in the economics community could have been intimidating but for his kindness and gentle manner in probing to the core o f subjects o f inquiry. My first dissertation paper, on whether air pollution emissions in the United States tend to converge to predictable patterns, originated from a required project in his vi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. class. Like Professor Stavins, Professor Jorgenson encouraged me to reach high standards o f scholarship, however difficult theory, data sets and econometric methods might prove. Professor Cary Coglianese, the third member o f my dissertation committee, helped me bridge academe and the policy world by guiding my research on EPA’s volunteer programs to encourage good corporate practices. With the addition o f Professor Coglianese, my dissertation committee reflected the spirit o f the public policy program, for it included three scholars o f distinction from three different fields. I much appreciated Professor Coglianese’s energetic inquiry and partnership in preparing research papers, assisted so ably by Jennifer Nash, Executive Director, Corporate Social Responsibility Initiative, Kennedy School. The association led to one o f my three dissertation papers. Many other professors at Harvard have contributed importantly to my progression. Among these were Professors Alberto Abadie, Majid Ezzati, and Richard Zeckhauser. Professor Alberto Abadie provided me step-by-step guidance in designing and executing the econometric procedures employed in one o f m y dissertation papers. I have also benefited a great deal from inspiring and fun discussions with Professor Zeckhauser while co-authoring a paper together. Professor Majid Ezzati introduced me to an important health issue in developing countries, adding another dimension to my studies. Fellow students in the Environmental Economics Program o f Harvard University provided me encouragement and constructive commentary, as well as the opportunity for me to learn o f their wide ranging fields o f inquiry. Louisa van Baalen, Director o f the vii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PhD Program in Public Policy, was a constant friend and always an answer to my frequent questions. My research efforts would have been stymied but for the cooperation and assistance o f a number of people and organizations. In particular, I would like to thank Julie K. Spyres, Director, Program Development and Member Services, National Environmental Performance Track, EPA. Also o f particular note, I would like to thank Center for Disease Control o f PR China for sharing with me its extensive data set on indoor air pollution derived during a joint study with the World Bank. Professor Majid Ezzati, Harvard School o f Public Health, and Dr. Enis Baris, World Bank, guided the study and engaged me in the process, providing the foundation for one o f my three dissertation papers. Schumpeter’s description o f capitalism as creative destruction brings to mind my struggles to first define the subject and then structure theory or empirical research to develop the hypotheses and counter hypotheses. Certainly there was much destruction during the process, but hopefully some creativity as well. However, that is for others to judge. It has been a great honor and privilege to attend Harvard University. In closing, I wish to thank the selection committees that approved my application and financial support, including fellowships from the Belfer Center at the Kennedy School o f Government and the Graduate School of Arts and Sciences. viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 1 Revisiting the Environmental Kuznets’ Curve: Do Pollution Levels Tend to Converge to Predictable Patterns? 1.1 Introduction The Environmental Kuznets Curve, a possible inverse U-shaped relationship between economic growth and environmental quality has generated a good deal of interest. At low levels o f development and income, environmental degradation is largely the result o f a subsistence economy and relatively low levels o f biodegradable wastes. W ith commercialization o f the agricultural sector and more extensive resource extraction, and as industrialization intensifies, environmental degradation worsens. Yet higher levels o f development and income, however, provide the resources and incentive for more efficient and environmentally friendly technologies and/or constraints, and environmental degradation declines (Panayotou 2000). This inverse U-shaped relationship between economic growth and environmental quality is referred to as the Environmental Kuznets’ Curve (EKC). If such a relationship can be verified, the policy implications are significant.1 There is the possible prospect of sustainable development, although it depends on factors o f technology and preference. Further, it suggests the need for research into the turning point, the degree o f environmental degradation before this point is reached, including potentially pushing beyond ecological thresholds, and the institutional and policy 1 Although the existence o f the EKC are yet to be proved, the EKC relationship is already playing an important role in policy-making. For example, the postulated relationship plays a substantial role in forecasting greenhouse gas emissions. The International Panel o f Climate Change (IPCC) implicitly assumes an EKC in their forecast o f greenhouse gas emissions. 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. conditions required to ensure environmental recovery as development advances to a high level. Most fundamentally, though, the premise itself requires careful scrutiny. Using cross-country data for the United States, Grossman and Krueger (1993) showed an inverted U-shape in the relationship between per capita GDP and ambient levels o f both sulfur dioxide and suspended particulates. They estimated the GDP per capita inflection point was in the range o f $4000 to $5,000 (in 1985 US dollars). Selden and Song (1994) and Grossman and Krueger (1995) later produced more refined estimates for these and other pollutants, based on better-quality data. Subsequent studies have been conducted using as dependent variables various airborne emissions, ambient concentrations o f various pollutants, water pollutants, deforestation, per capita solid waste, carbon dioxide, lack o f safe water, and other factors. The independent variable common to most models is income per capita, but some studies have used purchasing power parity data. Results o f the empirical analysis vary widely: some show a U-shaped relationship between the selected environmental indicator and per capita income; some show a downward sloping linear relationship; and yet others show an upward sloping linear relationship (Panayotou 2000). The estimated inflection points range from $823 for deforestation (Panayotou 1993) to over $18,000 for carbon dioxide (Moomaw and Unruh 1997). The factors contributing to an EKC relationship have been classified into broad groups. Panayotou (1997) and Islam, Vincent, and Panayotou (1999) identified three distinct structural forces that affect the environment: (a) the scale o f economic activity, (b) the composition or structure o f economic activity and (c) the effect o f income on the demand for and supply o f pollution abatement efforts. The scale effect on pollution, 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. controlling for the other two factors, is expected to be a monotonically increasing function o f income; the larger the scale o f economic activity per unit o f area, the higher the level o f pollution, all else equal. Structural change that accompanies economic growth affects environmental quality by changing the composition o f economic activity toward sectors o f higher or lower pollution intensity. The composition effect is likely to be a non-monotonic (inverted-U) function o f GDP; that is, as the share o f industry first rises and then falls, environmental pollution will first rise and then fall - in relative terms - with income growth, controlling for all other influences transmitted through income. Isolated from the scale and composition effects, the income variable reflects the demand for and supply o f pollution abatement. At low income levels, increases in income are directed primarily towards food and shelter, and have little effect on the demand for environmental quality. At higher income levels, increases in income lead to higher demand for environmental quality (since it is a normal good). Engel’s curve for environmental quality translates into an inverted-J curve between income and environmental degradation (Selden and Song 1995). On the supply side, higher incomes make available the resources needed for increased private and public expenditures on pollution abatement. Further, they induce stricter pollution regulations to help internalize environmental externalities. Grossman (1995) has suggested a fourth structural force or factor contributing to the EKC relationship - the “technological change effect” . This refers to technological progress that accompanies economic growth, since wealthier countries can afford to spend more on research and development. In turn, this contributes to the substitution of 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. obsolete and environmentally-insensitive technologies with cleaner ones, leading to improvements in the quality o f the environment. Policy-induced changes in pollution abatement technology and use o f more energy efficient technologies illustrate the influence o f this fourth factor. There are many theoretical models offering explanations o f the EKC. Some authors focus on shifts in production technology brought about by structural changes that accompany economic growth (Grossman and Krueger 1993, Panaytou 1993). Others have emphasized the characteristics o f abatement technology (Selden and Song 1995, Andreoni and Levinson 1998). And yet others have focused on preferences and especially the income elasticity for environmental quality. A few authors have formulated complete growth models, with plausible assumptions about the properties of both technology and preferences from which they derive the EKC. A number o f critical surveys o f the EKC literature have been published. Arrow et al. (1995) argue that the EKC model as presented in the 1992 World Development Report and elsewhere implicitly assumes there is no feedback between economic production and environmental damage, as income is treated as an exogenous variable. This highlights the importance o f adopting a dynamic optimization model, wherein both income and pollution are endogenously determined. There is a seeming disconnect between theoretical models and empirical models. The income-environment relationship specified and tested in much o f the literature is in a reduced form function that aims to capture the “net effect” of income on the environment. Income is used as an omnibus variable representing a variety o f underlying influences, whose separate effects are obscured. Decomposition models that test the four sets o f 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. structural factors contributing to the EKC relationship, outlined above, also seem to lack rigorous theoretical frameworks. Another problem is that most studies o f EKC are cross-country studies, with developed countries on the high end o f the income axis and developing countries on the low end. In order to infer the environment-income relationship o f a single country over time, cross-country studies implicitly assume that all countries will follow the same growth pattern. Critiques argue that the EKC that emerges with cross-country and crosssection analysis “may simply reflect the juxtaposition o f a positive relationship between pollution and income in developing countries with a fundamentally different, negative one in developed countries, not a single relationship that applies to both categories o f countries” (Vincent 1997). A further source o f concern regarding empirical estimates o f the EKC is the comparability and quality o f available environmental data. Stem et al. (1996) noted that pollution data used in environmental Kuznets curve studies are “notoriously patchy in coverage and/or poor in quality”. For these reasons, a number o f EKC studies involve only single countries, where the data may be better and more consistent. Most o f these studies, except those for the United States, suggest that the EKC relationship is weak and does not hold over time. However, different studies indicate conflicting results as to the effects o f growth on the environment (Borghesi,1999). In what follows, a somewhat different question is probed than the hypothesized inverse U-shaped relationship between economic growth and environmental quality. Rather than trying to trace out the EKC, this paper focuses on testing whether pollution levels in a country (or state/province) tend to converge over time towards the levels 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. defined in the steady state and/or the balanced growth path.2 Brock and Taylor (2004) used OECD data to test convergence predicted by a “green Solow model” and found significant evidence o f convergence. The models employed for this purpose are those developed by Stokey (1998); an endogenous growth model and an exogenous technical change model with pollution as a by-product o f production. The case study is for the United States, employing 1985-99 data concerning state emissions o f five o f the seven elements used to measure air quality: CO, NOX, PM10, SO2 and VOC.3 Two hypotheses are tested: (1) emissions at the state-level follow the convergence process predicted by Stokey using an optimal growth with pollution model; significant evidence is found supporting the proposed convergence phenomenon; (2) states converge to their respective steady states rather than the same steady states; there appears to be significant inter-state differences concerning steady state or balanced growth path emission levels, reflecting a wide variety o f state-specific characteristics; coefficient estimates will be biased if the modeler assumes interstate homogeneity. Section 1.2 following describes in more detail the data and econometric models used to test these hypotheses. Section 1.3 presents the empirical results and Sector 4 provides the main conclusions. 2 Steady state refers to no growth in per capita income while a balanced growth path refers to common growth rates in capital, output and consumption. 3 Emissions o f the other two criteria air pollutants, NH3 and PM2.5 are not available for the years from 1985 to 1999. 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.2 Data and Econometric Models Employed State-level data for annual emissions o f CO, NOx, SO2, PM10 and VOC4 were drawn from the National Emissions Inventory o f the US Environmental Protection Agency (EPA)5. Emission estimates for the years 1985-1999 were constructed using a ‘bottoms-up’ methodology, whereby emissions derived at the facility or county level were aggregated to the state-level. The emission estimates include non-point, point and mobile sources. All emissions data are in tons per year. The 1985-1999 data used in this study were last updated in 2002. Income and demographic data used in the study come from the US Bureau o f Economic Analysis and the US Census Bureau. All income data are expressed in 1983 dollar terms. Table 1-1 provides the summary statistics. 4 These pollutants are: carbon monoxide (CO), sulfur dioxide (S 02), particulate matter (PM 10), volatile organic compounds (VOC) and nitrogen oxides (NOx). 5 EPA's National Emission Inventory (NEI) database contains detailed information about sources that emit criteria air pollutants and their precursors, and hazardous air pollutants. An extract o f the database for AirData includes estimates o f annual air pollutant emissions from point, nonpoint, and mobile sources in the 50 States, the District o f Columbia, Puerto Rico, and the Virgin Islands. EPA conducts a national inventory o f air pollutant emissions at three-year intervals, and adds the new data to the National Emission Inventory database. Between inventories, EPA refines and corrects the emissions data, and updates the database several times. Data are extracted for use in AirData approximately once per year. 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1-1: Summary Statistics of Data Annual State-Level Data for 51 States: Emissions pc (ton per year), population and income pc, 1985-1999 Variable Obs Mean Std. Dev. Min Max 8.595 1.010 CO 765 0.543 0.596 0.151 nox pmlO 765 765 0.089 0.205 0.030 0.009 so2 765 0.123 0.196 0.102 0.106 0.004 0.669 VOC 765 0.106 0.103 0.031 1.491 popu 765 765 5000229 14557.910 5509847 453401 33100000 2389.469 9279.550 23175.330 Min Max 0.105 -2.338 -0.770 2.376 1.295 -0.012 0.200 0.124 -1.253 -0.602 1.358 1.030 pci 1.556 Annual Growth (log difference) for Testing Eq. (2), Mean Variable geo gnox Obs 714 714 gpmlO gso2 714 714 gvoc 714 714 -0.016 0.130 -1.203 1.229 gpopu 0.007 gi 714 0.013 0.010 0.021 -0.038 -0.112 0.069 0.086 -0.005 -0.002 -0.041 Std. Dev. 0.204 85-92 Interval Growth Rates (log difference) and Structure Variables, Testing Eq. (6) Variable Obs Mean Std. Dev. Min Max geo gnox 51 -0.036 0.141 0.001 0.073 -0.844 -0.444 0.263 0.153 gpmlO gso2 51 51 51 -0.019 -0.014 0.089 -0.538 -0.179 0.077 0.461 gvoc gpopu 51 51 -0.032 0.012 0.084 0.078 0.035 gi sso -0.008 -0.192 -12.583 svoc 51 51 51 51 51 0.040 -0.518 -0.015 -0.003 -0.471 -0.379 -12.830 -0.249 spm 51 0.015 -0.445 SCO snox 0.021 -0.190 -0.424 0.111 0.010 0.013 0.119 0.054 0.171 0.053 0.364 0.078 -12.083 0.021 -0.358 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1-1 (continued) 92-99 Interval Growth Rates (log difference) and Structure Variables, Testing Eq. (6) Variable Obs Mean geo gnox 51 0.213 -0.104 gpmlO gso2 gvoc gpopu gi sso2 SCO spm svoc snox Std. Dev. Min Max 0.285 -0.575 1.337 -0.119 -0.109 0.041 0.093 0.305 -0.362 -1.253 0.185 0.608 0.121 0.299 -0.443 -1.137 0.148 0.987 51 51 0.008 0.016 0.008 0.011 -0.007 -0.008 0.037 0.047 51 51 -0.296 0.144 0.079 0.097 -0.574 -0.242 -0.129 0.402 51 51 51 -0.233 -0.092 0.021 -0.335 -0.170 0.090 0.029 -0.283 -0.094 0.227 0.069 51 51 51 51 -0.037 85-99 Interval Growth Rates (log difference) and Structure Variables, Testing Eq. (6) Variable Obs Mean Std. Dev. Min Max geo 51 -0.091 0.287 -0.411 1.300 -0.045 0.200 -0.534 0.497 -0.746 -0.227 -0.290 0.122 0.515 0.436 -1.843 -1.702 0.316 0.130 -0.839 -0.201 0.713 0.665 0.830 0.199 -0.079 -0.662 0.066 0.128 -0.115 -0.555 -0.690 gnox gpmlO gso2 gvoc gpopu gi Sco Spm 51 51 51 51 51 51 51 0.319 0.023 -0.096 -0.588 0.121 Svoc 51 51 0.020 0.046 -0.267 0.119 -0.533 0.153 sso2 51 -0.294 0.127 -0.942 -0.109 Snox 51 0.643 0.277 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The five emissions included in the study contribute to a number o f serious health effects and together they comprise a reasonably comprehensive set o f the main factors that compromise air quality. CO exposure can lead to high levels o f carboxyhaemoglobin in the blood and to angina attacks; nationwide, three-quarters o f carbon monoxide emissions come from motor vehicles (cars and trucks) and non-road engines (such as boats and construction equipment). NOx is one o f the main ingredients in the formation of ground-level ozone, which can trigger serious respiratory problems; the primary sources o f NOx are motor vehicles, electric utilities, and other industrial, commercial, and residential sources that bum fuels. SO 2 contributes to respiratory illness. Over 65 percent o f S 0 2 released to the air, or more than 13 million tons per year, results from electric utilities, especially those that bum coal. Other sources o f S 0 2 are industrial facilities that derive their products from raw materials like metallic ore, coal, and crude oil, or that bum coal or oil to produce heat. Both NOx and S02 contribute to acid rain. PM 10 is associated with serious health effects; again the source is largely cars, trucks, buses, factories, and construction sites. VOC emissions also contribute to particulate formation, and VOCs and NOx are main contributors to tropospheric ozone pollution, which is linked to acute and chronic respiratory problems. Motor vehicle exhaust and industrial emissions, gasoline vapors, and chemical solvents are some o f the major sources of VOC.6 The econometric models employed are based on Stokey’s optimal growth model with pollution (Stokey 1998). Optimal growth models provide one general class o f theoretical foundation for the empirically observed EKC. They also represent an 6 EPA website on six common pollutants: www.epa.gov. 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. independent literature o f models o f development. These models use a Ramsey framework with pollution as a by-product o f production, and there is a trade-off between the disutility from a pollution byproduct and the utility derived from consumption. In Stokey’s model, the social planner specifies an optimal emissions standard that must be achieved. The social planner then chooses paths for consumption and for the technology in use so as to maximize the utility o f the (infinitely lived) representative household. Stokey considers a one-sector endogenous growth model with linear technology (AK model) and a one-sector growth model with exogenous technical change. Under the endogenous growth model, when environmental regulation is binding, the economy reaches a steady state7. As the economy approaches the steady state, total pollution declines if the form of the utility function satisfies certain conditions. Stokey assumes a standard constant inter-temporal elasticity o f substitution utility function. When the elasticity o f marginal utility is greater thanl, total pollution declines. The proportionate decline in marginal utility from increases in consumption is sufficiently significant that the increase in disutility from pollution, as a by-product o f production, dominates the increase in utility from consumption. Total pollution along the optimal path and at the steady state is, in both cases, determined by preference, technology and discount factors. Under the exogenous technical change model, there is a balanced growth path along which capital, output, and consumption all grow at the same rate. In the presence 7 Normally in AK models there are no steady states and no convergence. However, with pollution, output is produced with the capital stock and the aggregate pollution level as inputs. Thus, the real return to capital is not the constant A, but something that declines with capital for any fixed level o f pollution. As utility is concave in consumption and convex in total pollution, it is not optimal to increase pollution proportionally to the increase o f capital. As the capital stock grows the optimal emission standard becomes stricter, reducing the real rate o f return. When the emission standard gets strict enough, capital accumulation ceases (Stokey, 1998). 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. o f pollution, the common growth rate is lower than without pollution. Pollution also grows at a constant rate, which is lower than the common growth rate assumed for capital, output and consumption. Again, if the elasticity o f marginal utility is greater thanl, total pollution declines during transition towards and along the balanced path. Total pollution during transition and along the balanced path depends on preference and technology parameters and the share o f capital. Since preference and technology parameters are hard to measure, it is difficult to conduct tests o f the optimal growth path and the steady state. It is possible, however, to test the relationship between the common growth rate o f capital, output and consumption and the growth rate o f pollution emissions. If the economy is on the balanced growth path, the growth rate o f pollution emissions should be a linear function o f the common growth rate. Pollution emissions decline if the linear coefficient is negative, which depends on preference parameters. Another possible test is o f convergence. Both models predict convergence o f total pollution emissions - to the steady state for the AK model and to the balanced growth path for the exogenous growth model. Convergence can be tested through log linearization around the steady state or the balanced growth path. It should be noted that it is not assumed here that all economies converge to the same steady state or balanced growth path. Instead, it may be the case that economies converge to different steady states or balanced growth paths, reflecting differences in preference, production functions and endowment. The relationship between the common growth rate of capital, output and consumption and the growth rate o f pollution emissions is given by 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (1.1) Ge = fiG k Where Ge is the growth rate o f pollution emissions per capita and Gk is the common growth rate o f capital, output and consumption, again in per capita terms. P is a constant determined by preference factors. In standard econometric notation, the specification o f the empirical model is: (1.2) Yi = p X ,+ rjZi + eI Where Y; is the annual growth rate o f pollution emissions per capita; X; is the annual growth rate of personal income; Z, is a vector o f state characteristics to control for differences in preference8; and s; is a shock term. It is assumed that Sj-s are independently distributed with mean zero. A panel data analysis procedure is used to control for statespecific characteristics, and to correct for heterogeneity in the error term. Prais-Winsten regressions were run for the five pollutants to correct for serial correlation and to produce panel corrected standard errors (PCSE). The convergence test is applicable to both the AK model and the exogenous technical change model. For the AK model, the speed o f convergence when approximating the steady state is given by9 ^ ln^ > at -A [ln (£ * )-ln ( £ (l)) ] (1.3) Where A is the rate o f convergence, which is a function o f preference, technology and the discount rate.10 This equation implies that 8 The differences in preference across states are not readily measurable or observable. Initial per capita income, initial population levels, initial population density, and the population growth rate are included as control variables. 9 The log linearization technique is used for studying income convergence. (Mankiew, Romer and Weil, 1992; Islam, 1995) 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ln (£ (0 ) = (1 - e ~u) In(E*) + e~u ln(£(0)) (1 .4) Where E(0) is per capita emissions at some initial date. E* is per capita emissions at the steady state. Subtracting In(E(0)) from both sides, In(E(t)) - ln(£(0)) = (1 - e "*) ln(£*) - (1 - e * ) ln(£(0)) (1.5) In econometric notation, the specification o f the empirical model is: Yit = a , + f3Xlt_T + r]Zlt_T + s (1.6) Where Yt t is the log difference o f the initial and end per capita emissions; oti equals ( l - e -^) ln(£; *), which is a function o f the per capita emissions at the steady state; X; is the log initial level of per capita emissions; Z^-t is a vector o f control variables; andf,.( is state-specific shift or shock term. Individual US States may have different steady states determined by production technology, resource endowments, institutions etc. and the differences in the steady state could be correlated with both the growth rate o f emissions and the initial levels o f emissions per capita. If these differences are included in the error term, the coefficient for the initial level o f emissions could be biased. In order to avoid such bias, a set o f control variables is included to try and capture state-specific shifts. Based on this correction, it is assumed that ei t is independent o f the explanatory variables.11 The theory predicts that if there is convergence the coefficient for the initial level o f emissions per capita will be negative. The control variables include initial per capita income levels, initial population levels, initial population densities, income growth rates, population growth rates and a 10 The half-life is log(2)/A,. Hence, if X = 0.05 per year, then the half-life is 14 years, which means half o f the initial gap disappears in 14 years. (Barrow and Sala-Martin Xavier, 1999) 11 Such an assumption is common in growth models. 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. structural variable (Si) constructed specially for this analysis. The EKC literature has typically expressed environmental quality as a function o f average per capita income. In the case o f an economy approximating the steady state, it is reasonable to assume that the initial income level is correlated with the steady state emissions. Initial income may be negatively correlated with initial emission levels and emission growth rates as well, which is shown by some EKC empirical studies. Another variable o f concern is population growth. Population growth may affect per capita emissions independently and through population levels. However, it is not clear whether these variables are positively or negatively correlated with growth of emissions and initial emission levels. For example, Ravallion et al. (1997) argue that the demand for public goods, such as infrastructure and defense, may entail independent effects o f population growth on emissions. A higher population would result in higher public goods-related emissions for given income per capita, and population growth would result in growth o f emissions independently o f the growth in per capita incomes. The effect o f population density is ambiguous. Panayotou (1997) argues that there can be no a priori expectation as to the sign o f the population density variable. On the one hand, more people per square kilometer would result in higher S02 emissions, due to more intensive use o f coal and other fuels for cooking, heating and other energy needs. On the other hand, densely populated countries are likely to be more concerned than less populated countries about abating S02 at every level o f income. The structural variable relates to classification state i’s emissions into 13 source categories, as specified by the EPA12. The national growth rates o f emissions per capita 12 The 13 categories are fuel combustion from utilities, industrial fuel combustion, other chemical & allied product manufacturing, metals processing, petroleum & related industries, other industrial processes, 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for each category were first computed. The national growth rates were then weighted by the share o f each category in state i’s emissions per capita. Hence, the formula for Si is: 13 (1.7) Where Wy t_T is the weight o f category j in state i’s personal income at time t-T, and ln(is v / E -t_T) is the national average growth rate o f emissions per capita from source category j during the period t-T and t. Aside from the effect of changing source weights within a state, the variable S,t would equal the total growth rate o f per capita emissions in state i between years t-T and t if each o f the state’s sources grew at the national average rate for that source category. In particular, the variable S;t reflects shocks to sources of pollution in a way that interacts with state i’s concentration in those sources. Sit can be thought o f as a proxy for common effects related to source composition in the error term, The exogenous technical change model predicts convergence to the balanced growth path. Approximating around the balanced growth path, the speed o f convergence is given by: ln (£ (0 ) - ln(£(0)) = a - (1 - e- * ) ln(£(0)) (1.8) where a = g k + ( l - e A/)[ln(£*) + g k( t - T )], and gkis the common grow rate o f capital, output and consumption, in per capita terms. solvent utilization, storage & transport, waste disposal & recycling, highway vehicles, off-highway and miscellaneous 13 A similar structure variable was used in Barro and Sala-I-Martin’s 1991 paper on convergence o f income in the US states. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The difference between the exogenous technical change model and the AK model is the constant. However, this is not distinguishable in the econometric model specified above. Under both models, the coefficient for initial per capita emission is negative if there is convergence to the steady state or the balanced growth path. 1.3 Results 1.3.1 Inconclusive Evidence of Convergence: Growth Rates of Income and Pollution on the Balanced Growth Path Equation (1.2) was estimated for CO, NOx, PM10, S02 and VOC over the 14year period from 1985 to 1999. Lagrange multiplier tests were conducted and significant evidence o f serial correlation was found at the AR(1) level. To correct for serial correlation, Prais-Winsten regressions were run, which also provided panel adjusted standard errors. Table 1-2 shows summary estimation results with and without control variables. Table 1-2: Tests for Conditional Convergence, Balanced Growth Path Panel Data Analysis Corrected for Serial Correlation Dependent variable: log difference air pollutant emissions 1986-1999 With Control Variables Sample (713) NOx PM10 S 02 CO Ln(Yjt)-ln(Yjit_]) -0.7776 -0.5376 0.1673 -0.0690 (0.8654) (1.0204) (0.3533) (0.3860) R2 0.0171 0.0351 0.0199 0.0237 Without Control Variables Sample (713) CO NOx PM10 S 02 Ln(Yit)-ln(Yijt_i) 0.1862 -0.7897 -0.5235 -0.1108 (0.8696) (0.3673) (1.0297) (0.3808) R2 0.0067 0.0123 0.0033 0.0006 VOC -0.4600 (0.4604) 0.0121 VOC -0.4781 (0.4673) 0.0037 The regressions with and without control variables produce similar estimates of the coefficients. Coefficients o f income growth are mostly negative, but not significant at the 5 percent level. This gives a weak indication that emissions per capita grow at a negative rate, inversely to growth in per capita income. The poor fit o f the model, with R 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. squares around 1-2 percent, may be the reason why the coefficients are not significant. If more control variables were to be included, and the fit o f the model improved, the coefficients could become more significant. However, given that the coefficients are not significant at 5 percent nor 10 percent, the evidence supporting the decline o f pollution emissions on the balanced growth path is inconclusive. It should be noted, though, that this could simply be because the economy is in transition and the balanced growth path conditions don’t apply. This is in line with the common belief held by most economists that the US economy is in transition and the steady state has not been reached yet. 1.3.2 Strong Evidence of Conditional Convergence: Log Linearization around the Steady State or Balanced Growth Path Equation (1.6) was estimated with OLS procedure with robust standard errors. Cross-section regressions were run for CO, NOx, PM 10, S02 and VOC emissions over the period 1985-1999. The 14 year period from 1985 to 1999 was divided into two sub­ periods, 1985-1992 and 1992-1999. The model coefficients were estimated for the 14 year period first, and then for the two sub-periods separately so as to allow for possible different rates o f convergence. Panel data analysis was not employed, as the Lagrange multiplier tests showed significant serial correlation. The serial correlation problem cannot be easily solved because a lag term o f the dependent variable is included as one o f the explanatory variables. The first set o f regressions was run without controlling for state characteristics, while the second set includes control variables. Table 1-3 provides the regression estimates for the three periods, indicating in each case the coefficients for initial emission levels, the implied rate o f convergence, A,, 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and the equivalent half-lives o f convergence.14 The first part o f the table includes regression results without control variables for state characteristics. This specification assumes that all 51 states converge to the same steady state or balanced growth path. During the period 1985-1999, only S 02 showed a negative coefficient on initial pollution emissions, with an implied A, o f 0.0122 and a half-life o f 60 years. The coefficients for CO, NOx and VOC were positive but not significant. The coefficient for PM 10 was positive and significant, which suggests that there is divergence rather than convergence for this pollutant. During 1985-1992, CO, PM10 and VOC each had significant and negative coefficients, which supports the hypothesis o f convergence towards the steady state or balanced growth path. During 1992-1999, only VOC showed a significant negative coefficient, supporting the hypothesis o f convergence. The fit o f the model varies, with most of the R squares less than 10 percent, except for two regressions. In summary, the evidence o f convergence is weak when state characteristics are not controlled for. 14 Half-life means the time needed for half o f the initial gap to disappear. For example, if the half-life is 10 year, half o f the initial gap disappears in 10 years, and three quarters o f the initial gap disappears in 20 years. 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1-3: Tests for Conditional Convergence, Transition Dependent variable: log difference air pollutant emissions Restricted Model Without Control Variables 1985-1999 NOx PM10 Sample (51) CO 0.0134 0.1440 0.4908* Ln(E i>M) (0.0566) (0.1143) (0.0988) 0.0314 0.0011 0.3350 R2 -0.0010 -0.0285* Implied X -0.0096 Implied Half-life 1985-1992 Nox Sample (51) CO PM10 Ln(E ,,M) -0.2124* -0.0160 -0.0285* (0.0353) (0.0198) (0.0139) -0.0069 R2 0.4256 0.0787 0.0341* 0.0012 Implied X 0.0021* Implied half-life 20* 601 336* 1992-1999 NOx CO PM10 Sample (51) Ln(E ,iM) -0.1365 0.0119 0.0459 (0.0838) (0.0247) (0.0475) R2 0.0514 0.0047 0.0187 Implied X 0.0105 -0.0008 -0.0032 • Implied half-life 66 With Control Variables for State Characteristics 1985-1999 Sample (51) CO NOx PM10 Ln(E i>t.,) -0.0744 -0.3064* 0.0531 (0.2100) (0.0609) (0.2103) R2 0.4709 0.6160 0.5742 Implied X 0.0055 0.0261* -0.0037 Implied Half-life 126 27* 1985-1992 Sample (51) CO NOx PM10 Ln(E iiM) -0.2070* -0.0352 -0.0643* (0.0424) (0.0247) (0.0207) R2 0.7586 0.2735 0.2788 Implied X 0.0331 0.0051 0.0095* 21* Implied half-life 271 146* 1992-1999 Sample (51) NOx CO PM10 Ln(E j , , . , ) -0.4024* -0.0158 ' -0.0554 (0.0454) (0.1150) (0.0590) R2 0.2774 0.2357 0.4116 Implied X 0.0735* 0.0023 0.0081 Implied half-life 19* 609 170 S 02 -0.1566* (0.0591) 0.1076 0.0122* 60* VOC 0.0738 (0.1323) 0.0063 -0.0051 - S 02 0.0067 (0.0179) -0.0175 -0.0005 VOC -0.1508* (0.0192) 0.5485 0.0117* 59* - S 02 0.0254 (0.0189) 0.0354 ‘ -0.0018 - VOC -0.1805* (0.0864) 0.0818 0.0142* 49* S 02 -0.1551* (0.0490) 0.6111 0.0120* 58* VOC -0.4680* (0.1612) 0.4292 0.0451* 15* S 02 -0.0094 (0.0216) 0.1231 0.0013 1027 VOC -0.1844* (0.0293) 0.7800 0.0291 48* S 02 -0.0163 (0.0317) 0.1509 0.0023 590 VOC -0.3773* (0.1151) 0.2213 0.0677* 20* * significant at 5% level In contrast, when state characteristics are controlled for there is stronger evidence o f convergence in all periods. For the period 1985-1999, NOx, S 0 2 and VOC each had 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. significant negative coefficients - compared to only S 02 in the model without control variables. For the period 1985-1992, CO, PM 10 and VOC each had significant negative coefficients - as in the model without control variables. For the period 1992-1999, CO and VOC each had significant and negative coefficients compared to only VOC in the model without control variables. After control variables were included, the fit o f the models improved significantly with R squares ranging from a low o f 12 percent to a high o f 78 percent. In terms o f the five pollutants included in the study, VOC and CO emissions show relatively strong tendencies of convergence in at least two of the three periods, while S02, NOx and PM 10 show tendencies toward convergence in only one o f the three periods. The implied rates of convergence for the significant coefficients vary somewhat, ranging from around 1 percent to 7 percent, and the equivalent half-lives range from a low o f 15 years to a high o f 146 years. However, most o f the rates o f convergence fall between 1-5 percent. The rates o f convergence and the implied half-lives are comparable to those o f income convergence for the US states, with a rate o f convergence o f 2 percent and corresponding half-life o f 34 years (Barro and Sala-Martin X av ier, 1992). Other explanatory variables were included to control for state-specific characteristics determining the steady state. During the period 1985-1999, population growth and population density both had a significant negative effect on the growth rate o f emissions per capita. The structural variable was significant in some cases but not others. Initial income and population levels were mostly insignificant as control variables except for two regressions. For the 1985-92 period, initial income level was significant for most regressions, while other control variables were mostly insignificant. For the 1992-99 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. period, initial income and population levels were significant and negative in the case o f two regressions. These results support to some extent the earlier hypotheses concerning correlation between control and dependent variables. The model with control variables produces lower coefficients for initial emission levels than when no control variables are employed. In other words, omitting the control variables produces an upward bias o f the coefficients. The above results suggest that if there is convergence, it is likely conditional convergence. That is, each state converges to different steady states or balanced growth paths; there is not absolute convergence, where all states converge to the same steady state or balanced growth path. 1.4 Conclusions Stokey’s optimal growth model with pollution produces the EKC when certain technology and preference parameters are assumed. On the downward sloping part o f the EKC, per capita pollution emissions decline and eventually converge towards the steady state or the balanced growth path. The analysis presented in this paper, based on US data for state-level air pollutant emissions from 1985 to 1999, provides support for Stokey’s hypothesis. The US evidence suggests that individual state economies are not yet in steady states or on balanced growth paths. Rather, the pattern o f growth in emissions shows convergence towards the steady state or balanced growth path. The speed o f convergence appears to be comparable to that o f income convergence. These results compliment the extensive literature on the Environmental K uznef s Curve, which is one o f the first direct empirical tests o f the optimal growth model with pollution as a framework to study the 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. relationship between income and the environment during long term growth. However, unanswered questions by the paper are whether the state economies reach steady states or balanced growth paths and, if the balanced growth paths are reached, what are the growth rates in emissions. The empirical analysis underscores the importance o f controlling for state characteristics that determine the steady state or the balanced growth path, and hence the importance o f data relating to these characteristics. Such research may contribute to determining the government structure best suited for regulating the environment. While internalization o f environmental externalities suggests the need for a strong federal or central authority, wide differences among the 51 states in their steady states or balanced growth paths could suggest the need for a highly decentralized approach to minimize suboptimal provision o f environmental quality as a public good. Stokey’s model is a very useful framework for addressing sustainable growth issues. Since both income and environmental quality are endogenously determined, the feedback relationships are robust and pollutant emissions more contained. The model helps to address sustainable growth issues, offering prospect o f income growth with decline or no growth in pollution, though it hinges on whether increasingly strict environmental regulation is compatible with a constant rate o f return on capital. The findings o f this paper may have broad application, but caution is in order. There exist different steady states or balanced growth paths. The US example may not be generalized to the rest o f the world because different countries may have different steady states or balanced growth paths. In fact, this study shows that if there is convergence o f pollution, it is likely to be conditional convergence. Stokey’s model assumes optimal 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. policy-making on the part o f the social planner, that is, the level o f environmental regulation is optimal. This is unlikely to be the case, especially if pollutants travel over long distances and there is externality associated with emissions. The following factors may contribute to optimal policy-making: pollutant doesn’t travel over long distances, especially not across national borders; harm from pollution is visible and localized; the government is not dominated by interest groups, particularly industries. 24 ' Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References Andreoni, J., and Levinsion, A., The Simple Analytics o f the Environmental Kuznets Curve, NBER Working Papers Series 6739 (1998) Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C., Holling, C.S., Jansson, B.O. Levin, S., Maler, K.G., Perings, C., and Pimental, D., Economic Growth, Carrying Capacity and the Environment, Science 268, 520-521 (1995) Barro, R. and Sala-Martin Xavier, Convergence across States and Regions, Brookings Papers on Economic Activity, 1:1991 Barro, R. and Sala-Martin Xavier, Economic Growth, The MIT Press, 1999 Borghesi, S., The Environmental Kuznets Curve: a Survey o f the Literature, unpublished paper, Nov. 1999 Brock, William A. and Taylor, M. Scott, The Green Solow Model, NBER Working Paper No. 10557, June 2004 Carson, R.T, Jeon Yongil and McCubbin, D.R., The Relationship Between Air Pollution Emissions and Income: US Data, Environment and Development Economics 2 (1997): 433-450 Growssman, G., and Kreuger, A., Environmental Impacts o f a North American Free Trade Agreement, The U.S Mexico Free Trade Agreement (1993) Grossman, G., and Kreuger, A., Economic Growth and the Environment, Quarterly Journal o f Economics 110 (2), 353-377 (1995) Islam,N. Growth Empirics: A Panel Data Approach, The Quarterly Journal o f Economics, Vol. 110, No. 4(No., 1995), 1127-1170 Islam, N., Vincent, J., and Panayotou, T., Unveiling the Income-Environment Relationship: an Exploration into the Determinants o f Environmental Quality, Working Paper, Department o f Economics and Harvard Institute for International Development (1999) List, J.A. and Gallet, C.A., The Environmental Kuznets Curve: Does One Size Fit All?, Ecological Economics 31 (1999) 409-423 Mankiew N.G.; Romer, D; Weil D.N., A Contribution to the Empirics o f Economic Growth, The Quarterly Journal o f Economics, Vol. 107, No.2 (May, 1992), 407-437 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Moomaw, W.R., and Unruh, G.C., Are Environmental Kuznets Curves Misleading Us? The Case o f C 02 Emissions, Environment and Development Economics 2, Part 4, 451463 (1997). Panayotou, T., Demystifying the Environmental Kuznets Curve: Turning a Black Box into a Policy Tool, Environment and Development Economics (1997) Panayotou, T., Sachs, J., Peterson, A., Developing Countries and the Control o f Climate Change: A Theoretical Perspective and Policy Implications, CAERIID iscussion Paper No. 45. August (1999) Panayotou, T., Economic Growth and the Environment, CID Working Paper No. 56, Harvard University, 2000 Ravallion, M.H., and Jalan, J., A Less Poor World, but a Hotter One? Carbon Emissions, Economic Growth and Income Inequality, World Bank October 15 (1997) Selden, T.M., and Song, D., Environmental Quality and Development: Is there a Kuznets Curve for Air Pollution Emissions? Journal o f Environmental Economics and Management 27, 147-162 (1994) Stem D. I., “The Environmental Kuznets Curve”, International Society o f Ecological Economics - Internet Encyclopedia o f Ecological Economics, June 2003 Stokey, N.L., Are there Limits to Growth? International Economic Review 39, 1-31 (1998) Unruh, G.C., and W.R. Moomaw; An Alternative Analysis of Apparent EKC-type Transitions, Ecological Economics 25, 221-229 (1998) Vincent, J., Testing for Environmental Kuznets Curves within a Developing Country, Environment and Development Economics Vol. 2 part 4, 417-433 (1997) World Bank, World Bank Development Report 1992, New York: Oxford University Press, (1992) US Bureau of Census, Population Estimates, http://eire.census.gov/popest/estimates.php US Bureau of Economic Analysis, State and Local Personal Income, http ://eire.census.gov/popest/estimates .php US Environmental Protection Agency (1998) National Air Pollutant Emission Trends, 1900-1998, Environmental Protection Agency, Office o f Air Quality Planning and Standards, Research Triangle Park, NC, October US Environment Protection Agency, Airdata http://www.epa.gov/air/data/reports.html 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 2 Participation of Firms in Voluntary Environmental Protection Programs: An Analysis of Corporate Social Responsibility and Capital Market Performance 2.1 Introduction and Literature Review The role and overall responsibility o f the business community in the USA with regard to environmental protection reflects, at least in part, the evolving policy and regulatory framework set by the federal and state governments. Economic theory suggests that policy-makers set policy and regulatory instruments consistent with socially-optimal levels o f environmental protection and that businesses operate in a manner fully compliant with the framework set by government for achieving these levels. This still is the mainstay o f the environmental protection regime. Over the past ten years or more, the role o f the business community beyond mere compliance has received increasing attention. Although there is no well agreed-upon definition o f corporate social responsibility, examples o f firms acting “green” are abundant. Examples include an internal carbon-trading regime introduced by British Petroleum, the Responsible Care Program initiated by the Chemical Industry and the establishment o f the Socially Responsible Investment Funds. There is, in fact, a very extensive list o f firm or industry environmental initiatives that go well beyond mere compliance with the law. A plethora o f questions have arisen surrounding the issue. Do firms have additional moral and social responsibilities prompting them to devote resources to environmental protection above and beyond what is stipulated by the law? What are the motivations for doing so, and what is the frequency o f such behavior? (Hay, Stavins and Vietor (ed), 2005). 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Partly in response to the corporate social responsibility (CSR) movement, and partly to address the inadequacies o f conventional means o f environmental protection, the Environmental Protection Agency (EPA) has introduced voluntary environmental programs which seek to recognize and reward above average environmental performance. The Environmental Leadership Program, Star-track Program and National Environmental Performance Track Program are o f this form. These programs are characterized by the concept o f tiered regulation, which has been described as the “tailoring o f regulatory requirements to fit the particular circumstances surrounding regulated entities.” 15 Tailoring may include flexibility in compliance schedules, adjusting the frequency o f inspections or monitoring requirements, or differentiating the level and form o f sanctions. The effectiveness o f voluntary programs as a policy instrument depends pivotally on whether firms can participate on a sustainable basis and whether the benefits (including rewards) are sufficient to justify the extra costs incurred by beyondcompliance activities. In this connection, an extensive literature has developed relating some measure o f a firm’s economic performance with its performance regarding one or more dimensions o f social responsibility. There have been some 95 separate studies and 13 reviews o f the literature (Portney, 2005). Margolis and Walsh (2001) surveyed the 95 studies, which measure the economic performance o f firms in a variety o f ways, including their cumulative abnormal returns, return on equity, assets, and sales. The variables used to measure some dimension o f the commitment to CSR is diverse, including measures related to firms’ environmental performance, the products or services, corporate governance practices, and investments in 15 “Tiering: A Practical Guide to the Use o f Tiering as a Regulatory Alternative,” Project on Alternative Regulatory Approaches (Sept. 1981), 1. 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. countries that are considered to have less stringent policy/regulatory frameworks or labor practices. Frequently, the studies control for industry variables in which the firm operates, including the size o f the firm, its debt-to-equity ratio, and the intensity o f its R&D and advertising. According to Margolis and Walsh, slightly more than half (53%) o f the studies found a positive relationship when CSR was used as an independent variable to explain financial performance. For the remaining studies, no relationship or a negative or mixed relationship was found. Most reviewers o f the literature have concluded that the validity o f such studies is often compromised due to problems with measurement, specification o f estimating equations and omitted variable bias. Measurements o f a firm’s financial performance include accounting measures such as return on book equity, return on sales, and marketbased measures such as Tobin’s “q” or cumulative abnormal returns. None o f these measurements are free of problems. Accounting measures involve historic denominators that may bear no relation to market values; return on sales is not very useful from the financial perspective; market-based measurements can also be problematic, particularly following the sharp contraction o f equity markets in the late 1990s. Measuring environmental performance or social performance is even more problematic. Emissions output measures such as Toxic Releases Inventory (TRI) data are common, but they don’t account for technological differences across firms. Ratings of environmental performance prepared for investors are highly subjective. Use o f content analysis of annual reports risks conflating environmental performance with environmental rhetoric (Reinhardt, 2005). 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Other problems include sampling bias and direction of causality. Margolis and Walsh found that “over half of the 95 studies examine exemplary, notorious, or very large firms” (Margolis and Walsh 2001). The direction o f causality is hard to determine if one only finds a correlation between two variables. Another commonly applied approach is to conduct event studies around release o f important environmental news (e.g. TRI, reception o f an award, and environmental disaster). An influential study in this category is Hamilton’s 1995 study (Hamilton, 1995) of the announcement TRI data, which found significant negative cumulative abnormal returns during a 10 day window following the announcement o f TRI. This paper will employ an event study method for firms having been awarded membership in the EPA’s National Environmental Performance Track (NEPT) Program. Cumulative abnormal return is the dependent variable measuring firms’ economic performance, while NEPT membership is the measurement o f the environmental performance o f firms. Event study approach has the advantage o f establishing causality, as it is implausible to think that cumulative abnormal returns around the event time cause the event itself (Hamilton 1995). Further, as will be discussed later, NEPT membership is a relatively good measure o f the overall environmental performance for a firm. EPA devotes considerable resources to screening applicants and monitoring NEPT members. NEPT members are also quite diverse in terms o f size and industries. In this manner, this paper provides insight on the relationship between the environmental and financial performances o f firms. The results should help inform environmental policy concerning the use o f voluntary programs as a policy instrument. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.2 Corporate Social Responsibility and Capital Market Performance As noted above, there is no generally agreed-upon definition for CSR. Portney (Portney, 2005) proposed a definition for the purpose o f analyzing CSR from an economic perspective: CSR is “a consistent pattern, at the very least, o f private firms doing more than they are required to do under applicable laws and regulations governing the environment, worker safety and health, and investments in the communities in which they operate” . This definition, or definitions along these lines, are quite frequently used by economists in analyzing CSR.16 Heal (2005) defined CSR rather differently, reflecting his interest in capital market performance: “CSR involves taking actions which reduce the extent of externalized costs or avoid distributional conflicts.” This is not in conflict with Portney’s definition. In fact, Heal’s definition incorporates both positive and normative judgment on why and whether or not CSR should occur. In the realm o f environmental protection, conflicts between corporations and society almost always derive from differences between private and social costs associated with pollution. This conflict needs to and eventually will be resolved or at least reduced, including through legislation and other government interventions. How this conflict is resolved or reduced has or will have an impact on the economic performance o f firms. Heal views CSR’s role as “anticipating and minimizing conflicts between corporations and society and its representatives, aligning private and social costs if differences are the source o f the conflict, or minimizing distributional conflicts if these are the issue.” This role compensates at least to some extent for market imperfections and is a supplement to government intervention. 16 Some argue for incorporating the requirement that firms sacrifice shareholder value into the definition. Others argue that the best definition depends on the inquiry addressed (Hay, Stavins and Vietor, 2005 pp 146). For the purpose o f this study, Portney’s definition seems more appropriate. 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Citing anecdotal evidence, Heal claims that private/social differences can be reduced or compensated at little or no cost to the firms. The literature outlines the benefits from CSR programs, which can be broadly categorized as: reducing risk; improving relations with regulators; generating brand equity; improved human relations and employee productivity; and lower cost o f capital. In terms o f risk, pollution could lead to costs associated with tort and litigation, and the cost of conflicts with other groups in society, especially environmental NGOs. Tort and litigation can lead to financial loss, and conflicts with NGOs can depress earnings and share prices and give competitors an opportunity to gain market share. A positive relationship with regulators could be very important for heavily regulated industries. In general, a regulatory decision in favor o f a company with a strong reputation for socially responsible behavior will be greeted more positively than one in favor o f a company seen as anti-social in its conduct, which likely influences regulators in their decisions. In terms o f brand equity, there is evidence that consumers’ purchasing decisions are sensitive to companies’ positions on CSR, which has implications for the value o f a company’s brand. CSR helps improve human relations and employee productivity through attracting talented personnel who care about the firm’s image. Lower cost o f capital refers to access to Socially Responsible Investment (SRI) Funds.17 Proponents maintain that CSR policy makes firms more attractive to investors and raises their profit in the long run through the mechanisms outlined above. Hence the connection between a firm’s policies towards CSR and its position in the capital market. One of the first papers to examine the connection between CSR and capital markets was 17 SRI now accounts for o f the order o f 12% o f funds under professional management in the U.S. and a smaller but growing fraction in European countries (Heal, 2005). 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hamilton’s 1995 study o f the announcement of TRI data. The EPA makes public a firm’s self-assessment o f its releases o f toxic chemicals. Hamilton reviewed the press treatment o f these announcements and their impact on share prices. He used event study methodology to identify how the announcement o f toxic releases affected the stock market values of the firms concerned, relative to the market as a whole. He found a significant negative impact o f the information releases on stock prices, with an average impact on a firm’s stock market values o f $4.1 million. The size o f the impact depends on the number o f chemicals released by the firm, increasing by $236,000 for each additional chemical. Several event studies reached similar conclusions for other countries. Dasgupta, Laplante and Memingi (2001) studied the way in which capital markets in Argentina, Chile, Mexico and the Philippines reacted to information about a firm’s environmental performance. Their raw data were public recognition o f firms’ superior or inferior environmental performances, drawn from articles in major business newspapers that addressed corporate environmental performance. In the case o f recognition o f superior performance, the average rise in stock market value was 20%, and in cases o f poor performance the drop in value ranged from 5% to 15%. A later study by Dasgupta, Hong, Laplante and Mamingi (2004) found similar results for the Republic o f Korea. 2.3 Methodology This paper employs the standard event study method in finance to examine the reaction o f investors to announcement o f firms’ participation in the NEPT Program, which is the event o f interest. The event-study methodology is based on the efficient market hypothesis, i.e. capital markets operate efficiently to evaluate the impact o f new 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. information on expected future profits o f the firms. In the case o f the NEPT Program, membership applications are reviewed twice a year and decisions on acceptance are conveyed to facilities individually by regional EPA offices. When facilities are informed o f their acceptance into the NEPT Program, some may choose to release a statement on their own. In any case, a news release event is usually organized at the EPA headquarters in Washington DC to announce the names o f facilities admitted into the Program. In recent few years, the dates of the news releases were typically a few days or up to a month after the acceptance. Event study analysis is widely used and the procedures are quite standard. Campbell et al (1997) and MacKinlay (1997) gave detailed descriptions o f the methodology. The methodology involves five steps: event definition, including definition o f the event and estimation windows; determining selection criteria for including firms in the study18; estimating normal performance within the estimation window and predicting normal returns during the event window in the absence o f the event; calculation o f the abnormal return and cumulative abnormal returns within the event window, and testing whether the abnormal return and cumulative abnormal return for all firms treated as a group are statistically different from zero. In the following analysis, abnormal returns are estimated using the market model which assumes a linear relationship between the return o f any security to the return o f the market portfolio19. The CRSP value-weighted index is used for the market portfolio. 18 The criteria may involve restrictions resulting from data availability such as listing on the NYSE or AMEX, and occurrence o f other confounding events at the same time o f the event o f interest. 19Other models include the constant return model and Capital Asset Pricing Model (CAPM). CAPM is an equilibrium theory where the expected return o f a given asset is a linear function o f its covariance with the return o f the market portfolio. It was commonly used in 1970s but more recently there has been some 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R it - a i + P i R mt + (2 .1 ) e i< W ith E(eit) = 0 and Var (eit) = Sft Where R it — returns on security; Rmt — returns on the market portfolio t— the time index, i— index for security This model is estimated drawing upon stock price data for the period 150 to 30 trading days prior to the events. The news release events usually occurred around 30 trading days after acceptance to the Program. Excluding the 30 trading days prior to the news release event eliminates possible effects from publicizing activities o f individual firms upon notification o f membership approval. Estimates o f oij and p, are used to predict a normal return for each security over the time o f the event window. Abnormal return (AR) is defined as the difference between the normal return and the actual return. For a single security i at a given time t AR is calculated as: (2 .2 ) statistical evidence against the CAPM based on US stock market data for the past 30 years. Therefore, use o f the market model has been preferred. Market model represents a potential improvement over the constant-mean-retum model. By removing the portion o f the return that is related to variation in the market’s return, the variance o f the abnormal return is reduced. This can lead to increased ability to detect event effects. The benefit from using the market model will depend upon the R2 o f the market-model regression. The higher the R2, the greater is the variance reduction o f the abnormal return, and the larger is the gain. (Campbell, 1996) 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Under the null hypothesis that the given event has no impact on the mean or variance o f returns, and the assumption o f joint normality o f the abnormal returns, the distribution for any single abnormal return observation is as follows: ARu ~ N (0, Vi) r, = s \A R ,) = (2.3) si + L 5m The abnormal return observations are then aggregated along two dimensions through time and across firms to draw overall inferences for the event o f interest. During the period (72-7)), the cumulative abnormal return for a given firm (i) is aggregated as follows: C A R ,( T M = ^ A R , (2.4) t=T\ Asymptotically the variance o f the cumulative abnormal return for firm i is S?(Tu,T2) = (T2 - T l +l ) Sl (2.5) To test the null hypothesis o f zero cumulative abnormal return, one can formulate a Z test as: C A R ^T ^-N ^S fiT ^): ( 2 .6 ) C AR Z = (^ (T T ^ ~ N m ( 2 ' 7 ) An aggregation o f interest can also be performed across both time and events. In that scenario, the average cumulative abnormal return for a subset o f N events between two dates 7j and T2 is defined as: CAAR{Tt, r 2) = - j - J ] CAR, (T„T2) N M (2.8) Where N is the number o f events. 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The variance o f CAAR is: Var(CAAR (TX,T2) = ^ T f i 8 2{TX,T2) N2 (2.9) M Under the null hypotheses that the abnormal returns are zero, Z = CAAR(T x,T2) (var{CAAR{Tx,Ts) ) f 2 n {q X) (2.10) MacKinlay pointed out that this distributional result is asymptotic with respect to the number o f securities N and the length o f estimation window L. 2.4 Data For some thirty years the US Government’s policy approach to environmental protection has been characterized by technology-based regulations, ambient standards, environmental impact assessments, and information disclosure mandates. Based on powerful laws and tough regulations to curb pollution and to enforce the “polluter pay” principle, these instruments are largely highly centralized and inflexible (Hirsch 2001). Despite considerable progress in protecting and improving the environment, it is increasingly recognized that new approaches could and should be employed. The introduction o f market-based instruments and other innovations during the later half o f the 1980s and the first half o f the 1990s signaled a more flexible approach to environmental protection. The essence o f these new initiatives was “tiering”, whereby regulatory requirements were tailored “to fit the particular circumstances surrounding regulated entities.”20 More proactively, programs were introduced to recognize and reward above average environmental performance. These included the Environmental 20 “Tiering: A Practical Guide to the Use o f Tiering as a Regulatory Alternative,” Project on Alternative Regulatory Approaches (Sep. 1981), 1. 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Leadership Program (ELP) and StarTrack programs, the forerunners o f the National Environmental Performance Track (NEPT) Program. Under these programs, entities are encouraged by various forms o f incentives and rewards to go beyond compliance; the rewards include reduced regulatory requirements and public recognition o f corporate social responsibility. The NEPT Program is a voluntary program that seeks to recognize, reward, and encourage facilities that exemplify “strong” or “top” environmental performance. Its “mission” has been described by EPA as improving environmental performance, transforming relationships, and encouraging innovation.21 The four main goals o f the Program are: (1) recognizing top environmental performance, (2) rewarding top environmental performance, (3) encouraging continuous environmental improvement, 22 and (4) transforming relationships . EPA goes through a multi-stage admission process to ensure that accepted members o f the Performance Track Program meet standards set for “top environmental performance.” The screening stages include active recruitment, review o f application, and checks for violations. Applicants are required to meet Performance Track criteria in four areas: (1) establishing and maintaining a comprehensive environmental management system (EMS), (2) going beyond legal requirements by making commitments to continuous environmental improvement, (3) informing and seeking input from its local community about the facility’s environmental performance, and (4) maintaining a record o f sustained compliance with environmental requirements (EPA 2006). Yu and Coglianese (2006) examined the effectiveness o f the EPA’s screening process and found 21 Progress Report, 3. 22 http://www.epa.gov/performancetrack 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that it has succeeded in admitting relatively superior performers - as determined by three indicators (permit compliance system, risk screening environmental indicators, and toxic releases inventory) and controlled for industry and firm characteristics. The recruitment process is overseen by an EPA official, assisted by external contractors. Applications for Performance Track are accepted on a semi-annual basis, from February 1st to April 30th and from August 1st to October 31st.23 Potential new members are contacted by program representatives, to appraise them o f the benefits and application process. Follow-up contacts and information may lead to a formal letter of membership invitation from the Program Director. EPA’s outreach activities supplement recruitment activities under the NEPT Program. Decisions on membership acceptance are conveyed to facilities in August and February, respectively. Once a facility is awarded membership, it is required to provide annual performance reports which, together with public outreach programs and selective site visits by EPA officials, ensures accountability. Memberships are for 3 years and subject to renewal afterwards. If the annual performance reports fail to be delivered as required or, together with site inspections, reveal problems in meeting NEPT standards, EPA may ask the facility to withdraw from the Program. Each year, around 20% o f the members decide not to renew or not to submit an annual report, and drop out o f the Program24 . 23 http://www.epa.gov/performancetrack/apps/app.htm 24 Interview 1/6/2006 with Julie K. Spyres, Director, Program Development and Member Services, National Environmental Performance Track, EPA. 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Membership in the NEPT is publicized in a number of ways. While notification o f membership is private, new members may choose to issue a press release.25 Unfortunately, information about approval dates is only available for rounds 8 through 1326. EPA also issues news releases to announce new members soon after the approval dates. Further, EPA’s website highlights new members and elected officials are informed, as are trade journals, helping to generate media coverage. A search through LexisNexis shows that the NEPT Program received very little coverage in major newspapers, but a lot o f coverage in news wires. From 2000 to 2007, the NEPT appeared only once in The Houston Chronicle (March 27, 2004, Saturday), regarding the Houston Port Authority. However, from 2000 to 2007, the NEPT appeared 99 times in newswires. Eight out o f 13 news releases announcing new members were covered by news wires on the same day when the news events occurred. There were also news items that covered individual facilities following the news releases. The lack o f coverage o f the NEPT Program in major newspapers is consistent with the economic theory o f information provision developed by Downs (1957). Under the classical utility-maximizing framework, rational individuals have incentives to free ride on participation in political issues and demand relatively little information related to public policy issues. As a result, only selective media channels provide such information. The abundance o f coverage o f the NEPT Program in news wires suggests that 25 Interview 1/6/2006 with Julie K. Spyres, Director, Program Development and Member Services, National Environmental Performance Track, EPA. 26 Industrial Economics Ltd. is a contractor that helps screen applications on behalf o f the EPA. They maintain a database with all application information. The approval dates for round 8 through 13 are available. However, the dates for round 1 through 7 are not available from Industrial Economics. 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. environmental performance is an important consideration for investors and the business community in general, possibly due to its implications on firms’ future earnings. To date (May 2007), the Program has admitted facilities through 13 rounds o f membership applications and has about 450 active members. Membership has been increasing by about 11 percent annually. The NEPT Program is facility based; parent firm information is obtained from application forms posted online27. Stock price data and firm general information are obtained from the Center for Research in Security Prices (CRSP). The names o f parent firms obtained from the NEPT website were matched manually with firms included in the CRSP. This means that only publicly traded companies were included in the sample. As shown in Table 1, which includes information about new members accepted into NEPT in each round, 54.5% o f facilities were affiliated with publicly-traded parent firms in the NYSE, AMEX and NASDAQ stock markets. Table 2-1 does not include information about renewal, as it is not considered “news” to investors in the same sense that new memberships are. 27 Application forms, and annual reports o f the member facilities are stored online on a EPA website http://www.epa.gov/performancetrack/. 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2-1: Summary of New Membership Information by Application Rounds Round Date o f News Release Date o f Members hi p Approval 1 12/13/2000 2 With Stock data Total New Members3 Firm/ event2 # of Facilities n.a. # Firm (with first event)1 33 33 91 % o f total member facilities 35.5% 8/1/2001 n.a. 5 13 18 75.0% 24 3 2/11/2002 n.a. 1 4 19 59.4% 32 4 8/23/2002 n.a. 3 10 14 60.9% 23 5 3/6/2003 n.a. 6 14 20 54.1% 37 256 6 8/22/2003 n.a. 4 7 9 36.0% 25 7 2/11/2004 n.a. 9 13 33 84.6% 39 8 8/30/2004 8/1/2004 5 9 11 57.9% 19 9 3/4/2005 2/1/2005 6 18 28 51.9% 54 10 8/24/2005 8/1/2005 4 16 34 87.2% 39 11 4/27/2006 2/1/2006 6 11 20 54.1% 37 12 10/25/2006 9/18/2006 6 13 18 56.3% 32 13 3/6/2007 2/13/2007 5 18 30 54.5% 55 93 179 51.3% 672 Total 345 . Note: 1. This column only includes the number o f firms that had a facility joining NEPT for the first time. If other facilities from the same firm join NEPT in later rounds o f applications, the firm is not counted again. 2. This column includes firms that have facilities joining the NEPT in each given round. Repeats in later rounds are counted. 3. This column includes all new members that were ever accepted into the NEPT. Since some o f them dropped out o f the program, the total (672) is larger than the number o f current members (450). Many firms have multiple facilities in NEPT. Three hundred and five facilities, or sixty-eight percent o f all active members, are affiliated with 53 firms (organizations) that have more than one facility in NEPT. Public firms with the largest number of facilities in the NEPT Program include Johnson & Johnson (39), 3M (17) and Lockheed Martin Corporation (11). Most other firms have 2 to 5 facilities in the NEPT. In order to examine the determinants o f abnormal returns, firm-level data on industrial sector, size, and advertising and R& D expenditures were compiled. 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. O f the 93 publicly-traded firms with facilities that joined the NEPT Program, 87 firms had complete stock price data during the estimation and event windows. 28 * Firms with facilities that joined during the fifth round (March 2003) were dropped due to a confounding event —the Iraq War that began on March 19th 2003. This left 81 firms in the data set for the analysis o f abnormal returns and their determinants. A high percentage of firms (90%) were in the manufacturing industry; pollution intensive industries account for 26% o f the total. In the analysis that follows, 7 industries were categorized as pollution intensive: pulp and paper, chemical, petroleum, cement, iron and steel, non-ferrous metals and metal mining29. The relatively high proportion o f firms in pollution intensive industries reflects firms’ concern about reducing risk and securing good relationships with the EPA as a regulator. According to an EPA survey, the primary reason for joining the NEPT is TO gaining or securing a good collaborative relationship with EPA . In terms o f firm size, a majority o f the firms had between 16,000 to 125,000 employees. In terms of advertising expense, the data may not be complete. Many firms do not report advertising expense but, rather, selling/general/administrative expenses in their income statements. Forty-three out o f 81 firms reported no advertising expense. These firms are treated as 28 CRSP only has stock price data up to Dec. 2006, hence firms in Round 13 were dropped. 29 The definition o f pollution intensive industries varies from one study to another. In general there are three approaches: (1) ranking pollution intensity according to abatement costs (Tobey 1977 and Low and Yeats 1992), (2) ranking according to toxic intensity (Lucas et al, 1992) and (3) ranking according to emission intensity (Gallagher and Ackerman 2000). These different definitions seem to yield similar lists o f polluting industries. 30 Interview 1/6/2006 with Julie K. Spyres, Director, Program Development and Member Services, National Environmental Performance Track, EPA. 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. having zero advertising expense in the analysis, which may have biased the results. R&D data is relatively more complete. Only 10 out o f 81 firms reported no R&D expense. Table 2-2: Summary Statistics of Publicly-Traded Member Firms (N=81) Obs Mean Std. Dev. Min Max 81 3.27 5.43 1 40 each round o f application 81 1.81 2.51 1 20 Price o f share (USD) 81 40.89 23.91 1.45 115.10 Variable Total number o f participating facilities Number o f participating facilities in the Shares outstanding (millions) 81 567.56 1066.17 1.03 6295.49 Total assets ( millions USD) 81 18314.26 20107.24 54.92 103946 Total sales (millions USD) 81 16100.34 18437.8 7.44 91685 Number o f employees (thousands) 81 56.98 73.01 0.1 461 Advertising expense (millions USD) 81 263.76 689.90 0 3399 R & D expense (millions USD) 81 760.90 1266.42 0 6215.9 Manufacture industry 81 0.90 0.30 0 1 Metal mining 81 0.014 0.120 0 1 Pulp and paper 81 0.071 0.250 0 1 Chemical 81 0.143 0.352 0 1 Petroleum 81 0.029 0.168 0 1 Cement 81 0.014 0.120 0 1 Iron and steel 81 0.014 0.121 0 1 Non-ferrous metal 81 0.014 0.121 0 1 2.5 Results Some participating firms had multiple facilities that joined the NEPT Program at different dates; they each had more than one event date. Since the application rounds are only 6 months apart from each other, it is possible that investors do not value multiple events for the same firm equally. Therefore, two sets o f analysis were conducted reflecting possible differentiation in the event impacts. The first set o f analysis includes only event dates for firms that had facilities approved for membership in the NEPT Program for the first time. Another set o f analysis included all event dates for participating firms. 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Abnormal returns for the participating firms were calculated for five event windows. One event window only included the day when the news releases occurred (day 0). The underlying assumption is that the market is efficient and information about joining NEPT is reflected almost instantly in the stock prices. The second event window is defined over a period o f 5 trading days from one day before the event (day -1) to the third trading day (day 3) after. Including one trading day immediately prior to the event is common in event studies, reflecting concerns over the possibility o f information being leaked to the market before the occurrence o f the actual events. In the same way as above, 10, 15 and 20 day event windows were defined. Confounding events concerning the participating firms during the event windows were identified through coverage by the Wall Street Journal. Such events included major mergers and acquisitions, important news on earnings, developments on key products, etc., which could significantly affected the firms’ returns. Firms with confounding events were excluded from the analysis —the longer the event window, the more exclusions. Table 2-3 contains the estimations o f abnormal returns for the event windows. Table 2-3: Abnormal Returns for Participating Firms Event Window day 0 5 day event window # Observations 82 79 10 day window 75 15 day window 72 20 day window 70 Firms with First Events -0.0017 (-0.5667) 0.00325 (0.55272) 0.0118 (1.0261) 0.0371** (2.1953) 0.0337 (1.4978) # Observations 153 148 140 136 134 Firms with All Events -0.0018 (-1.0588) 0.00069 (0.20235) 0.0134** (2.00) 0.0176* (1.8333) 0.0096 (0.7619) ** statistically significant at 5% level. * statistically significant at 10% level, t-statistics are in parenthesis. 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. All estimates o f abnormal returns are positive except for day 0. Over the 10 day event window, firms with all events experienced a positive cumulative abnormal return of 1.34% on average, which is significant at the 5% level. Over the 15 day event window, on average firms with first events experienced a positive cumulative abnormal return of 3.7% (significant at 5% level), while firms with all events experienced a positive cumulative abnormal return o f 1.76% (significant at 10% level). These estimates are in a comparable range as those from other studies in North America, which typically vary from 0.3 to 2%31. These results indicate that on the day o f the news releases, the news did not cause significant average abnormal return among member firms. After the day o f the news release, the news about NEPT membership continued to reach investors and caused significant average cumulative abnormal return during the 10 and 15 day event window. When the event window is as long as 20 days, the average cumulative abnormal returns become insignificant again, possibly due to diminishing impact and more potential for other confounding information. This is quite typical o f the way that markets react to unexpected news. To illustrate,, unexpected interest rate changes by central banks are not instantaneously taken account o f in stock prices but, rather, cause sustained movement o f prices over periods from hours to months. Some economists and market practitioners have long disputed the efficient market hypothesis in its strong form. They believe that markets are subject to inefficiencies including the slow diffusion o f information, the relatively great power of 31 Previous studies in North America include Hamilton (1995), Muoghalu (1990), Lanoie and Laplante (1994), Klassen and McLaughlin (1996), Konar and Cohen (1997), and Lanoie et al ( 1998). The Dasgupta et al 2000 and 2004 found abnormal returns as high as 20% in developing countries. 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. some market participants (e.g. financial institutions), and the existence o f apparently sophisticated professional investors. In the case o f the NEPT Program, diffusion o f information is likely an important issue. As mentioned earlier, the Program received little coverage in the general media but extensive coverage in news wires. Institutional investors are likely the ones that received this information first and subsequently passed it on to the rest o f the market. Delays in media coverage may have also slowed down information diffusion. On the day o f the event, 3 out of 11 news releases examined in the analysis were not reported by news wires yet subsequent coverage on individual firms was extensive. In summary, news o f the NEPT events was likely spread among investors in the stock market not instantly but rather over a number o f days. The estimated average cumulative abnormal returns for firms with first events were more than twice as large as those for firms with all events, indicating that investors react to the first event much more strongly than to the later events. This is expected, as the abnormal returns are caused by membership announcements as “news” to the investors. Repeated acceptance into the NEPT Program in later rounds would carry less weight than the first time. Following the same methodology, abnormal returns for Rounds 8 to 12 were reestimated using approval dates as the event dates. None o f the estimates o f abnormal returns were significant. Earlier it was noted that some firms may have taken actions to publicize their membership upon notification o f approval. If they did, these publicizing activities did not cause - at least for members joining in Rounds 8 to 12 - significant abnormal returns over the periods defined by the event windows. 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The estimated cumulative abnormal returns can be translated into monetary terms. The stock market value o f the cumulative abnormal returns was calculated by multiplying the estimated cumulative abnormal returns by the average stock price and shares outstanding during the event window. During the 15 day window, firms with facilities joining for the first time experienced an average gain o f $418 million. On average, having facilities join the NEPT Program (including repeated events in different rounds of application) is associated with an average gain o f $329 million. As also mentioned earlier, joining the NEPT Program may impact firms’ market value through various channels, including reduced risk and waste, improved relations with regulators, generating brand equity, improved human relations and employee productivity, and lower cost o f capital. These mechanisms can be examined using the estimated cumulative abnormal returns and other relevant firm information. Potential of risk and waste reduction is greater for pollution intensive industries compared to nonpollution-intensive industries. If investors interpret the news o f NEPT membership as a signal for reduced risk, the stock values o f pollution intensive industries should experience a greater positive shock than others. To examine the mechanism o f generating brand equity, advertising expense was used as a proxy for brand equity. I f NEPT membership increases brand equity, the more important the brand equity is relative to other assets, the greater is the gain from news o f NEPT membership. Therefore, higher advertising expense should be associated with higher abnormal returns. R & D expense is a measure o f a firm’s investment in human capital. If NEPT membership helps attract talent and improve productivity, firms with higher R&D expense should receive bigger positive shocks to their stock values. Whether NEPT membership leads to lower cost of 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. capital could be examined through firms’ borrowing interest rates, but most firms do not report borrowing interest rates in the CRSP database. In addition to the channels above, the impact o f announcement o f NEPT membership may also vary with firm size and the number o f facilities joining the NEPT at each event date. Controlling for these factors, a linear regression model was estimated using the 15 day cumulative abnormal return in US dollars as the dependent variable and the independent variables o f interest including dummy variables for pollution extensive industries, advertising expense and R&D expense. For estimating the standard errors, robust and clustered standard errors were calculated for the individual firms. O f the 136 firm/events with complete stock price information during the 15-day event window, 133 firm/events (74 firms) had complete general firm information. Table 2-4 presents results o f linear regressions examining the factors which influenced the impact o f the announcement o f NEPT membership on stock values. 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2-4: Determinants of Cumulative Abnormal Returns8 Constant Advertising expense R&D Chemical Paper Steel Non-ferrous metal Cement Petroleum Metal mining 1 333007.8 (1.78) -170.3 (-0.38) 74 (0.24) -100060 (-.12) -94684.5 (-0.51) -66852.3 (-0.33) -320970.1 (-1.71) -279736.2 (-1.49) 1264923 (1.15) -336852.8* (-1.84) # facilities joining on event date # employees Total sales 2 -50491.5 (-0.27) 131.4 (0.36) 325.4 (1.26) -266238 (-0.45) 87842.3 (0.22) -210942.8 (-1.14) -165332.5 (-0.98) -145098.8 (-0.93) -126457.3 (-0.10) 678374.7*** (3.17) 179850.9*** (3.42) -3107.3 (-0.54) gy y*** (4.51) Total assets Dummies for application round Dummies for rank o f eventb # observations # firms R-squared -78.8*** (-4.39) 133 74 0.0037 133 74 0.07 3 152321.2 (0.33) 202.7 (0.40) 693.8* (1.89) -11188.6 (-0.01) 174750.6 (0.20) 1007881 (1.36) 1255011 (1.06) -588234.5 (-0.90) -828201.4 (-0.73) 2260077* (1.82) 95177.5 (1.31) 766.2 (0.10) 101.7*** (-4.00) -106.4*** (-3.04) Yes Yes 133 74 0.39 a. Dependent variable is cumulative abnormal return in thousands USD; t-statistics are in parentheses. b. For firms with multiple event dates, the event dates are ranked in time. The first event date is ranked first and so on. *** significant at 1% level. ** significant at 5% level. * significant at 10% level. In Specification 1, no control variables were included for firm size and application time. The independent variables o f interest are mostly insignificant, except for metal mining which has a negative significant coefficient o f -336853. This indicates 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that firms in metal mining have on average a lower cumulative abnormal return than firms in non-pollution-intensive industries; this is contrary to the hypothesis. In Specification 2, firm size and number o f facilities of each firm gaining NEPT membership at a given event time were controlled for. The coefficient for the metal mining industry become positive and significant, indicating a higher cumulative abnormal return o f US $678 million on average for firms in the metal mining industry compared to those in non-pollution-intensive industries. All control variables except number o f employees have statistically significant coefficients. Having more facilities accepted into NEPT increases cumulative abnormal returns. On average, one more facility joining NEPT at a given event time is associated with an additional US $180 million in cumulative abnormal returns. Firms’ total assets are inversely related to cumulative abnormal returns. A dollar increase in total assets is associated with a decrease o f about 8 cents in cumulative abnormal returns. A possible explanation is that the larger the firm is the more factors there are affecting its stock value, which decreases the relative importance o f news of NEPT membership and hence the impact on the stock value. In contrast, sales revenues are positively associated with cumulative abnormal returns —a dollar increase in sales causes an increase o f about 9 cents in cumulative abnormal returns. Holding total assets constant, a larger total sales revenue indicates a higher total asset turn-over ratio which measures a firm’s efficiency in using its assets to generate sales. This may simply mean that better management increases gains from news about NEPT membership. In Specification 3, time dummies for each application round and the number o f event times for a given firm were controlled for, in addition to those included in 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Specification 2. The coefficients for metal mining industry, total assets and total sales remained unchanged in sign and significance. In addition, R& D expense became positive and significant. An additional dollar on R&D expense is associated with 69 cents more in cumulative abnormal returns. R&D expense is a proxy o f firm’s investment in human capital. This indicates that NEPT membership may be viewed as important in improving human relations and productivity by investors. The coefficients for the time dummies were mixed in sign and significance. The same is true for the number o f event times for a given firm. 2.6 Conclusion The NEPT Program represents an attempt by the EPA to use voluntary programs as a policy instrument to encourage firms to go beyond compliance in protecting the environment. The effectiveness o f voluntary programs as a policy instrument depends on whether firms receive enough rewards to rationalize sustained participation. This paper investigates the stock market reactions to the news o f NEPT membership. Significant positive shocks to the stock value were determined in the 10 and 15 day event windows following the announcement o f NEPT membership. There is strong evidence that acceptance to the NEPT adds to market capitalization o f the accepted firms, thereby benefiting the shareholders. Further, the paper explores determinants o f shocks to the stock value through examining the hypothesis o f CSR. There is strong evidence that R&D, as a proxy for investment in human capital, is a significant determinant of cumulative abnormal returns. As indicated earlier, this may result from a firm being able to attract better quality employees and to raise overall productivity through joining the NEPT Program and improving its corporate image. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. There is not much evidence supporting the hypothesis that reducing risk may be another channel by which NEPT membership causes positive shocks to stock value. NEPT members are required to make commitments on reducing pollution in two major areas. These commitments may not have been as credible an indication o f risk reduction as actual performance would be. Also relative to the scale o f risk o f liability and tort faced by pollution-intensive industries, the reductions committed under the NEPT Program may not be sufficiently important. This may help explain the lack o f significant larger positive cumulative abnormal returns for firms in pollution intensive industries. The lack o f evidence for the role o f advertising expense as a proxy for brand name is puzzling. However, the fact that around 50% o f the firms analyzed do not report advertising expense may have biased the results. Further, advertising expense is a very inadequate measurement of brand name. Firms with well established brand names will have less need to rely on advertising expenses to enhance their brand names; firms that do not have well-established brand names are more likely to spend on advertising. These results suggest that firms have an incentive to join voluntary programs, through the stock price effect and positive returns to investors. However, it is unknown whether the pollution reduction resulting from the additional commitments under voluntary programs is socially optimal. They could be too little or too much. CSR theory offers an alternative approach for reaching socially-optimal levels of environmental protection, in addition to the government acting as a planner maximizing social welfare. Provided that the resulting pollution reduction brings us closer to the socially optimal levels, voluntary environment programs can be an effective complement 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to performance-based instruments, which encourage firms to engage in beyondcompliance pollution reduction. 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References Campbell, John Y.; Lo, Andrew W.; MacKinlay, A. Craig; Lo, Andrew Y.; MacKinlay, Archie Craig, Econometrics o f Financial Market, Princeton University Press, December 9, 1996. Dasgupta, Susmita; Laplante, Benoit; and Mamingi Nlandu, “Pollution and Capital Markets in Developing Countries”, Journal o f Environmental Economics and Management 42, 310-335, 2001. Downs, A. An Economic Theory o f Democracy, Harper, New York (1957) Gallagher K. and Ackerman F., Trade Liberalization and Pollution Intensive Industry in Developing Countries: A Partial Equilibrium Approach, G-DAE Working Paper No. 00-03, Oct. 2000 James T. Hamilton. "Pollution as News: Media and Stock Market Reactions to the Toxics Release Inventory Data." Journal o f Environmental Economics and Management 28 (13/8 1995): 98-113. Hay, Bruce L., Robert N. Stavins, and Richard H.K. Vietor (editors) . Environmental Protection and the Social Responsibility o f Firms —Perspectives from Law, Economics, and Business. Washington DC: Resources for the Future, 2005. Heal, Geoffrey. "Corporate Social Responsibility - An Economic and Financial Framework." Columbia Business School, May 2005. Hirsch, Dennis D., Second Generation Policy and the New Economy, 29 Cap. U. L. Rev. 1 ( 2001 ). Klassen, Robert D; McLaughlin, Curtis P., “The Impact o f Environmental Management on Firm Performance”, Management Science, Vol. 42, No. 8 (Aug., 1996), 1199-1214. Konar S. and Cohen M. A., Information as regulation: The Effect O f Community Right to Know Laws on Toxic Emissions, Journal o f Environmental Economics and Management 32, 109-124, 1997. Lanoie P. and Laplante B., The Market Response To Environmental Incidents in Canada: A Theoretical and Empirical Analysis, Southern Economic Journal 60, 657-672, 1994. Lanoie P., Laplante B., and Roy M., Can Capital Markets Create Incentives For Pollution Control? Ecological Economics 26, 31-41, 1998. Low, Patrick and Alexander Yeats. Do ‘Dirty’ Industries Migrate? International Trade and the Environment. Ed. Patrick Low. World Bank, Washington DC, 1992. 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Lucas, Robert, David Wheeler, and Hemamala Hettige. “Economic Development, Environmental Regulation, and the International Migration o f Toxic Industrial Pollution, 1960-1988,” International Trade and the Environment. Ed. Patrick Low. World Bank, Washington DC, 1992. MacKinlay, A. C., Event Studies in Economics and Finance, Journal o f Economic Literature. 35, 13-39, 1997. Margolis, Joshua, and James, Walsh, People and Profits? Mahwah, NJ: Lawrence Erlbaum Associates, 2001. M. I. Muoghalu, H. Robison, and J. L. Glascock, Hazardous Waste Lawsuits, Stockholder Returns, And Deterrence, Southern Economic Journal, 357-370, 1990. Portney, Paul R., “Corporate Social Responsibility: An Economic and Public Policy Perspective”, in Hay, Bruce L., Robert N. Stavins, and Richard H.K. Vietor (ed) . Environmental Protection and the Social Responsibility o f Firms — Perspectives from Law, Economics, and Business. Washington DC: Resources for the Future, 2005. Reinhardt, Forest, “Environmental Protection and the Social Responsibility o f Firms: Perspectives from the Business Literature”, in Hay, Bruce L., Robert N. Stavins, and Richard H.K. Vietor (ed). Environmental Protection and the Social Responsibility o f Firms — Perspectives from Law, Economics, and Business. Washington DC: Resources for the Future, 2005. Tobey, James. The Effects o f Domestic Environmental Policies on Patterns o f World Trade, Kyklos, 43, 2, 1990, 191-209. US EPA, Progress Report, 3; http://www.epa.gov/performancetrack/about.htm. Yu, Fei and Coglianese, Cary, Recognizing and Encouraging Environmental Leaders: An Assessment o f Performance Track’s Selection Process, Beyond Compliance: Business Decision Making and the U.S. EPA’s Performance Track Program. Coglianese and Nash (ed). Regulatory Policy Program Report R PP-10, Harvard University, 2006. 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix 2-1: List of All Firms in Sample Set Name # facilities _________________ Roundl: 12/13/2000___________ Dana Corp 2 Eaton Corp 1 Ingersoll-Rand Co Ltd 2 Marathon Oil Corp 1 Bristol-Myers Squibb Co 2 Lockheed Martin Corp 9 Meadwestvaco Corp 1 Inti Paper Co 6 Johnson & Johnson 20 3M Co 7 Motorola Inc 3 CMS Energy Corp 1 Rohm And Haas Co 1 Cooper Tire & Rubber Co 1 Ryder System Inc 4 Hewlett-Packard Co 1 Baxter International In 2 Masco Corp 1 Fuji Photo Film -Adr 1 Interface Inc -Cl A 2 Sony Corp -Adr 1 Teradyne Inc 1 PNM Resources Inc 1 Baker Hughes Inc 1 Akzo Nobel N v -Adr 1 Rio Tinto Group (Gbr) 2 Sanmina-Sci Corp 1 Cytec Industries Inc 1 Concur Technologies Inc 1 Infineon Technologies A________________________1_ Round 2: 8/1/2001____________ Eaton Corp 1 Pfizer Inc 1 Johnson & Johnson 3 Louisiana-Pacific Corp 1 Temple-Inland Inc 1 Ibis Technology Corp 1 Concur Technologies Inc 1 B asfA g -Adr_________________________________ 1_ Round 3: 2/11/2002___________ Chevron Corp 1 Inti Paper Co 1 Johnson & Johnson 7 Baxter International In 4 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Snap-On Inc Visteon Corp 1 1 Round 4: 8/23/2002 Honeywell International Lockheed Martin Corp Inti Paper Co Pfizer Inc Johnson & Johnson Motorola Inc TdkCorp -Ads Baker Hughes Inc Rio Tinto Group (Gbr) Visteon Corp 1 1 2 1 2 1 1 1 1 1 Round 6: 3/6/2003 Caterpillar Inc Johnson & Johnson Baxter International In Snap-On Inc Lafarge North America I Spartech Corp 1 1 1 1 2 1 Round 7: 2/11/2004 Timken Co Texas Instruments Inc Gillette Co Dow Chemical Lockheed Martin Corp Johnson & Johnson 3m Co Rohm And Haas Co Unilever N v -Adr Nucor Corp Valspar Corp United States Steel Cor Concur Technologies Inc Rockwell Collins Inc 6 1 1 1 2 3 2 3 1 1 1 1 1 6 Round 8: 8/30/2004 Eastman Kodak Co Inti Paper Co Pfizer Inc Motorola Inc Georgia-Pacific Corp Inti Rectifier Corp Dupont Photomasks Inc Visteon Corp Rockwell Collins Inc 1 1 1 1 1 2 2 1 1 Round 9: 3/4/2005 Inti Paper Co 1 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Pfizer Inc Johnson & Johnson 3M Co Schering-Plough Hewlett-Packard Co Baxter International In Xerox Corp Fuji Photo Film -Adr Weyerhaeuser Co Interface Inc -Cl A Spartech Corp Baker Hughes Inc Rockwell Collins Inc__________________________ 2 1 2 1 2 2 1 1 1 1 3 1 1_ _________________ Round 10: 8/24/2005___________ Coca-Cola Co Eaton Corp Applied Materials Inc Lockheed Martin Corp Inti Paper Co Pfizer Inc 3M Co Motorola Inc Rohm And Haas Co Schering-Plough Fuji Photo Film -Adr Louisiana-Pacific Corp Hitachi Ltd -Adr Monsanto Co_________________________________ 1 1 1 2 4 1 4 1 1 1 1 5 1 1 _________________ Round_11:4/27/2006___________ Coca-Cola Co 1 United Technologies Cor 2 Pfizer Inc 1 Dover Corp 1 Stanley Works 2 Louisiana-Pacific Corp 2 Spartech Corp 1 Cytec Industries Inc____________________________1_ Round 12: 10/25/2006__________ Coca-Cola Co 1 Olin Corp 1 United Technologies Cor 2 Pfizer Inc 1 Johnson & Johnson 1 3M Co 1 Xerox Corp 2 Tyco International Ltd 1 Analog Devices 1 Alliant Techsystems Inc 1 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 Measuring Health Benefits from Interventions to Reduce Indoor Air Pollution in Rural China 3.1 Introduction Indoor air pollution (IAP) in poor rural households in developing countries is a leading health risk factor (Ezzati, et al, 2002). The combination o f heavy reliance on low energy fuels (biomass and coal) and low efficiency stoves, together with poor ventilation, results in dangerous levels o f pollutants in the form o f gases and suspended liquids and particulates (Smith, 1999). Although detailed epidemiological and toxicological research on the health effects o f IAP is still at an early stage, there is growing consensus that it is a causal agent of acute respiratory infections (ARI), chronic obstructive pulmonary disease, lung cancer, tuberculosis, nasopharyngeal and laryngeal cancers, and asthma. It may also cause low birth weight and perinatal mortality (Rehfuess and Rouse, 2005). O f 20 leading health risk factors in very low and low income developing countries, IAP ranks, respectively, as the fourth and eighth most important mortality risk factor (WHO; 2002; Ezzati et al, 2002). It accounts for more than 500,000 deaths annually in China (Ezzati and Baris ed. 2006). Combining the mortality and morbidity effects, IAP ranks as the fourth most important cause o f loss o f a healthy life in developing countries (DALYs).32 Women and children are particularly afflicted by IAP, 32 Disability-adjusted life years (DALYs) lost to mortality is the total discounted value o f years lost to premature death across all causes and age groups. DALYs lost to disability are based on the incidence and duration o f various types o f disability multiplied by a weight that accounts for the severity o f the disability compared to loss o f life. Total DALYs result from the sum o f DALYs lost to mortality and disability, adjusted by a discount rate so that years o f life lost at different ages are given different relative values (Nuria Homedes, 1995). 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. as normally they are in the cooking area or home environment much longer than other family members (Ezzati et al, 1997). During 2002-2005, the World Bank, in collaboration with the Government of China (Center for Disease Control), conducted an extensive IAP project in four provinces: Shaanxi, Guizhou, Gansu and Inner Mongolia. The project introduced affordable household energy technologies and household behavioral changes in selected townships, designed to substantially reduce indoor air pollution and exposure to it and, thereby, to lower the associated health risks. Some 5,500 households were included in the project. This paper analyzes the degree o f reduction in IAP emissions and ARI risk resulting from these interventions, and their net benefits. In particular, the paper examines the relative effectiveness o f new stove technologies in combination with health education/ behavioral changes, versus the latter alone. By comparing the net benefits o f these two options, conclusions are drawn relevant to public policy alternatives for addressing the problem o f indoor air pollution. The paper utilizes on an extensive set of data compiled under the World Bank/Government o f China project and it appears to be one of the first attempts to quantify, at the household level, the net health benefits o f alternative IAP interventions. 3.2 The Literature There is a large literature concerning household energy use, but the early publications tended to report on improved stove projects in developing countries and their contribution to energy efficiency, with relatively little attention to the health benefits o f reduced IAP. Projects in this category include the government sponsored National 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Improved Stove Program in China (Smith et al, 1993) and smaller projects in India, Mongolia, Nepal and Mexico. As pointed out by review studies, these projects sometimes had unintended negative effects for indoor air quality (Sinton et al, 2004). More recent projects have shifted the focus to improving indoor air quality. The World Bank/Government o f China project examined in this paper is one o f the largest with this objective. Studies o f the health implications o f IAP exposure are concerned with causally linking IAP to certain diseases and quantifying - where possible - this relationship. One study reviewed more than 100 papers reporting health effects o f household solid fuel combustion in China (Zhang and Smith, 2005). Another survey reviewed 13 more recently published studies that quantify the relationship between exposure to IAP and ARI in young children (Smith et al, 2000). An ongoing study in this category is the randomized intervention trial being carried out in the Western Highlands o f Guatemala (Smith et al, 2006).33 ARI includes a complex group o f conditions o f various aetiology and severity.34 Non-serious Acute Upper Respiratory Infections (AURI) include the common cold, sinusitis, tonsillitis, otitis media and pharyingitis. Potentially life-threatening Acute Lower Respiratory Infections (ALRI) include pneumonia, bronchitis, bronchiolitis and laryngitis. From the public health perspective, ALRI is o f greater concern as it results in most of the costs associated with ARI, including loss in DALY. 33 Publications associated with this project are listed on the website: http://ehs.sph.berkelev.edu/guat/page. asp? id=07 34 Aetiological agents may include diphtheria, influenza, pertussis and measles. 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Some 20 studies have been cited as indicating strong evidence o f IAP as a source o f ALRI among children under five in rural households in developing countries (Rehfuess and Rouse, 2005). In terms o f relative risk, children under five exposed to indoor smoke are indicated as more than twice as likely to suffer from pneumonia as children not exposed.35 Children under one year o f age are particularly susceptible to ALRI, due to their immature immune systems. In developing countries, they normally suffer at least one episode of ALRI every 2-3 years (Benguigui et al, 2001). Children are quickly responsive to reductions in indoor air pollution, hence ARI evaluations can be undertaken fairly shortly (6 months) after IAP interventions. The best form of ARI evaluation is a physician-based assessment o f pneumonia in children. A questionnaire-based assessment o f respiratory disease is much less reliable, especially when conducted before and after the IAP interventions; the first survey alerts those interviewed to the nature o f the problem, influencing responses during the second survey and thereby potentially adding to those citing ARI symptoms. A critical review o f the quantitative literature and data sources in nine countries found consistent evidence indicating a significant increase in the risk o f ALRI for children exposed to IAP (Smith et al., 2000). Since ALRI is the chief cause o f death o f children in developing countries, the authors conclude that “there is an urgent need to conduct randomized trials to increase confidence in the cause-effect relationship (between IAP and ARI), to quantify the risk more precisely, to determine the degree o f reduction in exposure required to significantly improve health, and to establish the effectiveness o f interventions” . 35 It is also stated that women exposed to indoor smoke are more than three times as likely to suffer from chronic respiratory disease than women not exposed (Smith et al, 2004). 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. While there is a relatively large literature on the health effects o f IAP, estimates o f the health benefits from IAP interventions at the household level are few. Household level analysis is important because it provides the micro foundation for formulating public policy initiatives. Further, to understand the potential market demand for IAP interventions, and the public policy initiatives needed to complement market responses to the IAP problem, analysis is needed o f the household benefits o f these interventions. The ARI implications for children under five are an important component o f this analysis. 3.3 The Project and Associated Data 3.3.1 The World Bank/Government of China IAP Project During 2002-2005, the World Bank and Government o f China tested affordable household energy technologies and behavioral interventions designed to substantially reduce IAP and exposure to it and, thereby, to lower the associated health risks (Ezzati and Baris ed, 2006). Eleven townships were selected to test the interventions, three in each o f Shaanxi, Guizhou and Gansu provinces, and two in Inner Mongolia. Criteria for site selection included similar economic and housing conditions, a preponderance o f rural households, geographical separation o f the sites, low family incomes, women and children family members, and the need for space heating and reliance on solid fuels (coal and biomass) for heating and cooking purposes. One o f three townships for each province (with the exception o f Inner Mongolia involving only two townships) was subject to the full range of interventions, combining stove and ventilation technologies with health education and behavioral changes). New alternative stoves were provided at approximately one-third the market cost36. The 36 The households were mainly responsible for labor and some material cost. 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. second township for each province was subject to only health education and other activities to induce behavioral changes; stove alternatives were not introduced. The third township for each province served as a control group, where no interventions were undertaken. In this way, ex ante and ex post comparisons for each o f the stove and behavioral intervention (S+B) and partial (B) intervention townships could be matched against the ex ante and ex post comparisons for the control (C) townships, providing the basis for analysis on the effects o f the interventions. Five hundred households from each township were selected for participation, for a total o f 5,500. Almost 2,500 households in the study areas were assisted in acquiring new stoves and in improving the chimney/ventilation systems. Another 700 households undertook stove and/or chimney improvements at their own expense. The new stove technologies were designed and tested to meet local conditions and needs in each o f the study provinces. Health education and behavioral activities were extensive, directly involving 3,500 study households and 5,000 students. The health education materials were prepared by local education experts and professional designers, emphasizing the health risks o f IAP exposure and the behavioral options for minimizing exposure37. They were presented in the local media and included in school and social mobilization activities, such as workshops, forums, field visits and demonstrations. 3.3.2 The Surveys, On-Site Measurement and Health Evaluations The several phases o f the IAP study generated an extensive set o f data. Household, health and other surveys, together with tests concerning energy use, on-site 37 The World Bank project report (Ezzati and Baris eds, 2007) details the activities that took place and the content o f the educational material. All original health education/behavioral material can be found on the CDC website: http://www.54rz.com/iao/eg/index.asp 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. measurement o f household IAP levels, and health evaluations, were conducted before and after the interventions. The baseline household, health and other surveys were conducted in March and April, 2003. The household energy technology interventions were undertaken during August to December 2004. The behavioral interventions were carried out over a sixmonth period in the second half o f 2004 and the post-evaluation surveys were conducted in April and May, 2005. Surveys were conducted for all households involved in the project, including questions about general household information and stove and energy use characteristics. Most respondents were female household members. Health questionnaires applied to all women and children in the project households, including information on general health conditions and IAP-related health symptoms. With respect to IAP knowledge and behavior, approximately 150 households from each township were surveyed. The surveys were conducted by the staff o f provincial and county Health Bureaus and Centers for Disease Control and Prevention. Additional data on energy use behavior were collected through field observations by village health workers and village committees and leaders. With respect to IAP, about 25 households from each o f the 11 townships were tested for three indoor pollutants: inhalable particles (PM4 and PM 10, o f median aerodynamic diameter o f less than 4 mm and 10 mm, respectively); carbon monoxide (CO); and sulfur dioxide (S02). For the baseline data, the households were tested in both December/January 2003 and March/April 2003, over a 24 hour period. The tests were conducted at multiple points within the households, so as to capture data on the levels and 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. pattern o f IAP exposure. A second method o f testing involved smaller samples of households (approximately 6 for each province), who were subject to continuous testing over a four-day period in both December and March. The two sets o f tests yielded broadly similar results. The post evaluations were conducted in Dec. 2004-Jan. 2005 and March-April 2005. For all the project areas, the heating season lasts from approximately November to late March, with the peak heating season in December and January. Doctors trained according to the WHO Integrated Management o f Childhood Illness (IMCI) procedures conducted ARI evaluations for all children aged five and under in the project households. The doctors examined the children (some 300-500 in each province) bi-weekly for six times before the project interventions and eight times after.38 The evaluations included questions posed to mothers about their children regarding ARI symptoms, such as cough, phlegm, difficult breathing, runny nose, sore throat, ear pain, etc. In addition, the children were examined for breathing frequency, chest in-drawing and other danger signs o f ALRI. 3.3.3 Baseline Pollution Levels, KAP and Health Information and Summary Statistics IAP levels are directly related to stove and fuel type and intensity o f use, as well as ventilation and interior home design. These factors vary considerably among the four provinces. While households in Gansu and Inner Mongolia rely almost entirely on biomass (fuel wood and straw) for cooking purposes, households in Guizhou rely predominately on coal; households in Shaanxi use both biomass and coal in more equal proportions. Housing structure also varies by province, in particular in terms o f house 38 As noted earlier, children are quickly responsive to reductions in indoor air smoke, hence ARI evaluations can be undertaken fairly shortly (6 months) after IAP interventions. 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. material and number of rooms and cooking space separation. An earlier paper on the project summarized the housing structure in detail (Jin et al, 2005). 39 These and other variations in the co-determinants o f IAP and ARI required a province-by-province analysis o f the IAP concentrations and health benefits o f the stove and behavioral interventions. Upon closer examination o f the data for the four provinces, it was found that children five years o f age or younger in the project households o f Shaanxi province were reported as having virtually no incidence o f ALRI before or after the interventions.40 Since ALRI is the main health indicator o f interest, Shaanxi province was excluded from the analysis. As described earlier, PM, CO and S02 concentrations were measured in selected households before and after the interventions. Table 3-1 presents summary statistics for the PM dataset. This dataset was obtained through merging data from households subject to PM measurements with data from completed survey questionnaires;41 households 39 Most houses in Gansu have a kitchen separated by a wall from the sleeping/living room, with separate entrances. The most common housing design in Guizhou consists o f 2-3 rooms— cooking/living, sleeping, and entrance/storage— connected with doors. Although most houses have a separate cooking area, this is only used on special occasions and cooking is normally done in one o f the main rooms (cooking/living room), especially during the heating season. Older homes in Inner Mongolia are constructed inside a cave­ like structure with a single room used for cooking, living, and sleeping, with the cooking stove connected to the bed for heating. The newest homes in the study area have a wall with windows and door between the cooking and sleeping/living areas. Most houses in Shaanxi have a cooking area connected to the main house by a door, a living room with a ground stove (fire-pit) used for heating and boiling water, and one bedroom which sometimes also has a ground stove. Most houses have a small attic used for storage but not for sleeping. 40 In total, 464 children under five years o f age were each examined six times before the interventions in Shaanxi Province; only two children were determined to have had ALRI, translating into an incidence rate o f 0.0004%. After the interventions, 290 children were examined eight times, and no cases o f ALRI were found. This means virtually zero incidence o f ALRI, while in other provinces the incidence o f ALRI ranged from 1.1% to 3.7%. The very low incidence o f ALRI in Shaanxi may have been due to quality issues o f diagnosis, or simply random factors. 41 Some households had measurements before the interventions but not after, and vise versa. This dataset does not include information about children under five; matching for children under five would decrease the number o f observations by more than half. 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lacking completed questionnaires were not included.42 Analysis based on this reduced sample may be subject to biases or at least loss o f precision. However, the analysis o f PM concentrations was conducted using both the full sample and the reduced sample, as presented in later sections. CO and S 02 concentrations were not included in the analysis because the sample sizes were too small and the measurements were undertaken in only two provinces (Ezzati and Baris ed., 2006). In terms o f CO, 24-hour mean CO concentrations were consistently within health-based standards and guidelines except for a small number of observations in Gansu, Inner Mongolia and Shaanxi43 Cost considerations limited S02 measurements to the two provinces where the primary fuel is coal (Guizhou and Shaanxi). S 02 concentrations were higher than the WHO guideline value o f 0.04 ppm at all locations in both provinces. 42 In Gansu province, 9 out o f 78 households were discarded because they did not have household survey information. In Guizhou province, all households were included. In Inner Mongolia, 7 out o f 53 households were discarded. 43 The WHO ambient air guideline for CO standard is lOppm; the ACGIH guideline is 25ppm for 8-hr exposures. 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3-1 Summary Statistics Basic IAP Dataset Province Group Household (n) Family size Per capita cash income (yuan) House material Mud, wood and tile % Brick, wood and tile % Brick and concrete % Coal consumption (jin) Non-heating season Heating season Biomass fuel consumption (jin) Non-heating season Heating season Cooking stove % Unimproved biqmass Improved biomass Unimproved coal Improved coal Heating stove % Unimproved biomass Improved biomass Traditional coal Improved coal Daily Stove use (hr) Cooking time Heating (Mar.-Apr.) Heating (Dec.-Jan.) Smoking in the house S+B 21 4.4 Gansu B C 25 23 5.1 5.1 514.6 389.7 0.74 0.26 0.00 All S+B B Guizhou C 23 4.3 All Inner Mongolia B C All 22 23 45 3.6 3.6 3.6 1566. 1356. 1160. 2 7 6 21 4.7 68 5.4 290.9 315.4 258.7 0.46 0.54 0.00 0.45 0.43 0.12 0.37 0.63 0.00 0.44 0.53 0.04 0.14 0.43 0.43 0.00 0.54 0.46 0.05 0.50 0.45 4.40 114.0 215.76 630.1 258.33 536.8 617.33 989.3 310.35 682.4 0.00 414.3 0.00 450.0 0.00 431.5 474.5 603.0 493.5 657.9 721.3 209.6 577.8 201.2 671.4 98.0 691.0 189.3 400.6 350.0 389.2 569.2 394.2 473.9 0.35 0.65 0.55 0.43 0.47 0.49 0.47 0.20 0.60 0.02 0.02 0.07 0.79 0.06 0.16 0.03 0.87 0.08 0.27 0.13 0.71 0.05 1.00 0.14 0.43 0.00 1.00 0.08 0.77 0.00 1.00 0.11 0.59 0.00 0.24 0.15 0.23 0.03 0.34 0.04 0.28 0.06 0.07 0.00 0.23 0.03 0.00 0.00 0.52 0.01 0.00 0.00 0.37 0.13 0.04 0.00 0.35 0.05 0.00 0.00 0.14 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.07 0.00 1.1 1.5 4.7 0.56 1.4 2.0 4.3 0.57 1.2 0.3 2.5 0.44 1.2 1.2 3.7 0.52 4.3 6.9 7.5 0.78 4.4 6.5 7.8 0.81 4.6 7.4 7.5 0.79 4.4 6.9 7.6 0.79 3.3 3.1 3.2 5.5 0.93 7.3 0.85 6.3 0.89 69 4.9 24 6.4 420.7 432.5 216.6 0.75 0.25 0.00 0.70 0.30 0.00 0.73 0.27 0.00 0.00 245.8 0.00 56.8 10.81 85.2 406.3 576.4 572.3 775.2 0.48 0.36 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3-1 continued. Mother fn) Age Elementary school plus Cooking time (min) Occupation % Housewife Farming Respiratory disease % PM obs before (n) Avg. concentration (mg/m3) PM obs after (n) Avg. concentration 21.0 31.7 0.68 119.2 30.0 37.8 0.82 152.9 32.0 40.3 0.72 124.4 83.0 37.2 0.75 133.4 30.0 41.9 0.30 148.6 26.0 39.1 0.54 107.9 24.0 42.8 0.55 185.1 80.0 41.2 0.42 142.7 11.0 34.5 0.64 135.0 11.0 35.0 0.85 123.6 22.0 34.8 0.74 130.0 0.41 0.56 0.82 70 0.44 0.56 0.57 108 0.59 0.39 0.66 106 0.49 0.49 0.67 284 0.07 0.93 0.67 188 0.09 0.91 0.18 108 0.06 0.94 0.48 74 0.07 0.93 0.51 370 0.07 0.93 0.64 30 0.38 0.62 0.77 26 0.22 0.78 0.70 56 0.60 62 0.44 84 0.29 116 0.42 262 0.21 180 0.20 108 0.49 76 0.26 364 0.87 26 0.47 26 0.68 52 0.176 0.228 0.241 0.221 0.222 0.287 0.221 0.241 0.222 3 0.217 6 0.219 9 Note: S+B Group were households subject to both stove improvements and health education/behavioral changes. B Group were households subject only to health education/behavioral changes. C Group were households serving as the control group, where no interventions were conducted. 71 Compared to the national averages, the project areas have low income levels and high illiteracy rates (especially for females). In particular, the sample households have extremely low cash income levels, especially those in Guizhou and Gansu provinces. Further, they have substantial ethnic minorities (with different languages) and largely rural populations heavily reliant on farming for their livelihoods. Income levels, illiteracy rates and ethnicity have important implications for stove and fuel type, and stove-use practices (Ezzati and Baris ed, 2006). Project households in Gansu and Inner Mongolia rely entirely on biomass fuels during the non-heating season, and use a combination o f biomass and coal during the heating season. Guizhou relies more on biomass than coal during the non-heating season but reverses this pattern during the heating season. In a substantial number o f cases, especially in Gansu, improved cooking stoves (from the earlier National Improved Stoves Program) were in use prior to the World Bank project interventions. Mean hours o f stove use varied by province. Guizou households reported the highest number o f hours (4.4 hrs) spent on cooking, and Gansu reported the lowest number of hours (1.2 hrs). During the heating season, Guizhou and Inner Mongolia reported use o f heating stoves for 6-7 hours daily, while in Gansu the number o f hours of use was about half that. Guizhou households have the longest hours o f stove use and the highest consumption o f fuel.44 The baseline PM data indicated high levels o f PM in all provinces, but especially in Gansu and Inner Mongolia where biomass use is more pronounced. There, PM levels exceeded the US EPA health guideline (65 pg/m3) by as much as 13 times. Both 44 This is not consistent with mean temperature in the three project areas. Inner Mongolia has the lowest mean temperature, Gansu follows, and Guizhou has the highest mean temperature in December and January. 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. provinces also experienced dramatic reductions in PM levels following the W orld Bank project interventions. Guizhou, in contrast, had the lowest baseline levels o f PM concentration and experienced only small changes in PM levels. In Gansu and Inner Mongolia, women with children were diagnosed as having high rates o f respiratory disease, while those in Guizhou were reported to have much lower rates. The KAP (knowledge, attitude, practices) survey information could not be merged with the PM dataset without a major reduction in the observation numbers. The survey information is summarized in the World Bank report (Ezzati and Baris ed., 2006). The baseline survey indicated that, in all provinces and for all socio-demographic groups, the majority o f respondents were aware that smoke from cooking and heating stoves is a health hazard. However, only a small percentage could identify cooking, heating and smoking as the sources of the indoor air pollution. Respondents below the age o f 40, with better education and with higher incomes had the greatest knowledge o f the health hazards and risk. Knowledge about effective means to reduce IAP exposure was limited except for simple measures such as opening the window for ventilation. Unfortunately, the PM dataset could not be merged with the child ARI dataset without reducing the observation numbers dramatically. In addition, household survey did not document how much time each o f the household members (children in particular) spent in the kitchen and other areas o f the home45. Lacking personal exposure levels for children five years or younger, it is difficult to link ex post changes in PM levels with ex post changes in ARI.46 45 For the children’s health survey, the questionnaire, only inquired about how long the children were present when their mothers were cooking. 46 Total personal exposure to an air pollutant can be estimated as the weighted average o f the pollutant concentrations in the environment where a household member spends time; the weights are proportional to 73 Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission. A separate dataset was constructed, through merging the ARI dataset for children, the women’s and children’s health survey dataset, and the household survey dataset. Table 3.2a-3.2c presents summary statistics for the basic ARI sample. the time spent in each o f these environments having distinct pollutant concentrations. References include Sexton K, Ryan PB, Watson AY, et al, 1988 and Ezzati and Kammen, 2001. 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3-2a. Summary Statistics Basic ARI Sample — Gansu Province Group Household Ini Family size PC annual cash income (yuan) House material Mud, wood and tile % Brick, wood and tile % Brick and concrete % Coal consumption (jin) Non-heating season Heating season Biomass fuel consp. (jin) Non-heating season Heating season Cooking stove % Unimproved biomass Improved biomass Unimproved coal Improved coal Biogas Charcoal Heating stove % Unimproved biomass Improved biomass Traditional coal Improved coal Charcoal Open fire Coal stove no chimney Smoking in the house M other Ini Age Elementary school plus Cooking time (min) Occupation % Housewife Farming Respiratory disease % Child Ini Age Male % Daily exposure time (min) [t-stat] C Mean All mean SD [1.22] 50 4.6 107 4.8 1.25 447.9 [-.024] 477.5 409.9 441.01 [-1,42] [-0.81] [-0.81] 0.88 0.11 0 [2.15] [-2.00] [-0.70] 0.66 0.32 0.02 0.75 0.24 0.01 0.43 0.43 0.09 8.6 85.1 [-2.49] [-0.24] 16.5 103.8 [-1.68] [0.35] 43.1 93.2 26.4 93.1 63.01 254.57 541.4 784 .2 [1.43] [2.10] 513.1 792.3 [0.99] [1.961 455.9 607.1 495.2 703.7 156.46 394:16 0.6 0.34 0 0 0.14 0.14 [1.71] [-2.64] [0.87] 0.38 0.62 0 0 0.12 0.04 [-0.26] [-0.06] [-0.84] 0.42 0.62 0.05 0 0.09 0.02 0.46 0.54 0.50 0.50 0.14 0.31 0.06 0.14 0 0.03 0.14 0.06 0.57 31 29.1 0.8 148.5 [0.51] [-0.62] [-0.96] 0.38 0 0.08 0 0.08 0.06 0.08 0.77 26 30.2 0.96 121.5 [1.08] [-1.62] [-1.64] 0.26 0.09 0.23 0 0.06 [1.43] 0.31 0.06 0.17 0 0.05 0.11 0.7 50 28.1 0.93 115.1 0.68 28.9 0.89 127.6 0.34 0.64 0.66 53 2.5 0.49 28.9 0.35 0.63 0.64 114 2.5 0.5 21.6 S+B mean [t-stat] 31 5 [1.41] B Mean 26 4.96 279.3 [-1.95] 0.8 0.2 0 0.32 0.60 0.66 35 2.4 0.46 17.1 [0.69] [2.30] [-0.61] [1.38] [-1.57] [-1.21] [0.83] [-0.53] [1.52] [-0.24] [-0.39] [-0.03] [-0.25] [-0.30] [-1.26] 0.38 0.62 0.58 26 2.7 0.58 12.7 [0.28] [0.51] [0.34] 0.15 [-1.12] [0.65] [1.49] [0.67] [0.38] [0.39] [-0.22] [-0.72] [0.78] [0.72] [-1.50] 0 0.11 0.06 0.32 0.24 0.46 0.24 0.37 0.22 0.31 0.32 0.47 5.96 0.31 71.73 0.48 0.49 0.48 1.46 0.50 40.64 Notes: The means for each group and the whole sample are reported. T-statistic is reported for the null hypothesis that the average for the treatment group (S+B or B) is the same as the control group. Standard deviation is reported for the whole sample. The t-statistics are shown in bold if the differences are significant at 5% level. 75 Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission. Table 3-2b. Summary Statistics Basic ARI Sample — Guizhou Province Group Household fnl Family size PC annual cash income (yuan) House material Mud, wood and tile % Brick, wood and tile % Brick and concrete % Coal consumption (jin) Non-heating season Heating season Biomass fuel consp. (jin) Non-heating season Heating season Cooking stove % Unimproved biomass Improved biomass Unimproved coal Improved coal Heating stove % Unimproved biomass Improved biomass Traditional coal Improved coal Coal stove no chimney Smoking in the house Mother fnl Age Elementary school plus Cooking time (min) Occupation % Housewife Farming Respiratory disease % Child fnl Age Male % Daily exposure time (min) S+B mean 23 5.1 [t-stat] B Mean [t-stat] C mean All mean 117 4.5 1 .3 8 SD [2.91] 38 4.5 [0.95] 56 4.3 371.7 [0.80] 539.5 [3.10] 310.5 390.4 3 9 6 .3 0 0.5 0.44 0.04 [1.72] [-1.22] [-0.57] 0.26 0.49 0.26 [-0.48] [-0.97] [2.94] 0.31 0.58 0.07 0.33 0.55 0.12 0 .4 7 0 .5 0 0 .2 0 382.6 733.3 [-1.04] [-0.17] 470.9 695.3 [0.25] [-0.45] 444.1 747.6 440.6 729.3 5 2 2 .1 0 5 5 0 .6 6 335.6 134.1 [1.49] [0.80] 11.1 10.7 [-2.65] [-1.59] 183.1 83.1 160.3 71.1 3 9 6 .5 0 2 4 1 .5 7 0.41 0.07 0.22 0.07 [3.33] [1.59] [-0.35] [0.11] 0.07 0.12 0.40 0.20 [-0.89] [2.47] [1.57] [2.30] 0.12 0.01 0.25 0.07 0.16 0.06 0.79 0.11 0 .3 7 0 .2 3 0 .4 1 0 .3 2 0.05 0.05 0.4 0.12 0.7 0.77 38 29.1 0.6 118.7 [1.88] [1.88] [1.56] [1.21] [-1.68] [-0.23] 0 0 0.26 0.05 0.81 0.78 56 28.9 0.63 164.7 0.01 0.01 0.29 0.08 0.79 0.77 0 .1 2 0 .1 2 0 .4 6 0 .2 8 0 .4 2 0 .4 2 29.1 0.7 146.2 8 .9 8 0 .4 8 7 2 .8 7 0.03 0.95 0.96 43 2.4 0.61 29.4 [-1.22] [1.31] [-4.26] 0.04 0.92 0.53 74 '2.2 0.56 85.9 0.03 0.95 0.6 144 2.3 0 58.3 0 .2 8 0 .1 4 0 0 0.22 0.11 0.89 0.74 23 29.8 0.76 141 0.05 0.95 0.4 27 2.3 0.53 28.7 [-0.35] [0.95] [0.56] [0.45] [0.53] [0.01] [-1.42] [0.99] [2.40] [-1.01] [0.51] [-1.13] [-2.961 [0.12] [0.01] [-3.73] [0.84] [1.29] [-3.45] Notes: variables reported same as in table 2a. 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 .2 7 0 .7 0 8 5 .4 4 Table 3-2.c Summary Statistics Basic ARI Sample — Inner Mongolia Group Household fnl Family size PC annual cash income (yuan) House material Mud, wood and tile % Brick, wood and tile % Brick and concrete % Coal consumption (jin) Non-heating season Heating season Biomass fuel consp. (jin) Non-heating season Heating season Cooking stove % Unimproved biomass Improved biomass Unimproved coal Improved coal Heating stove % Unimproved biomass Improved biomass Traditional coal Improved coal Coal stove no chimney Smoking in the house Mother fnl Age Elementary school plus Cooking time (min) Occupation % Housewife Farming Respiratory disease % Child fnl Age Male % Daily exposure time (min) B mean 64 4.1 [1.01] C Mean 60 3.4 All mean 124 3.8 3.59 1301.8 [-1.48] 2871.7 2086.7 5964.43 0.23 0.58 0.19 [-3.14] [-0.70] [3.00] 0.42 0.58 0 0.34 0.58 0.09 0.47 0.50 0.24 37.5 604.7 [-0.69] [1.89] 59.8 442.7 48.7 523.7 183.72 439.43 742.5 790.6 [5.31] [3.10] 367.7 479.2 555.1 634.9 490.05 586.87 0.88 0.14 0.31 0.02 [3.77[ [-2.55] [3.30] [1.00] 0.59 0.33 0.59 0 0.73 0.23 0.45 0.01 0.44 0.43 0.50 0.09 0.05 0 0.22 0.03 0 0.78 64 28.8 0.75 147.5 [1.76] [-1.43] [-0.41] [-1.46] 0 0.03 0.25 0.09 0 0.86 60 30.1 0.78 120.3 0.02 0.02 0.23 0.06 0 0.82 124 29.5 0 134.3 0.15 0.12 0.43 0.24 0.57 0.42 0.16 68 3.2 0.47 40.2 0.31 0.69 0.26 135 2.7 0.49 54.3 0.05 0.96 0.36 67 2.2 0.55 68.5 [t-stat] [-1.15] [-1.05] [-0.41] [3.90] SD 0.39 6.79 0.43 54.62 0.46 0.44 1.84 0.64 69.89 Notes: variables reported same as in table 2a. In accordance with the need to distinguish between the two types o f ARI, respiratory data for children five and under were categorized as either AURI or ALRI, based on the Integrated Management o f Childhood Illnesses (IMCI) procedure. AURI includes coughing, some discomfort in breathing, and some combination o f a sore throat, 77 Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission. ear discharge and runny nose. ALRI (clinical pneumonia) includes severe coughing with fast and difficult breathing and chest indrawing47. For Gansu and Inner Mongolia, ex ante ARI bi-weekly evaluations for children five and under were conducted during April and May 2003, and ex post evaluations during April and May 2005. In Guizhou province, ARI evaluations for children five and under were done in June and July 2003, and then again in April and May 2005. As noted earlier, the stove and behavioral interventions were undertaken from August to November 2004 in Gansu, from September to October 2004 in Guizhou, and from August to October in Inner Mongolia. The ARI dataset was merged with the household, children’s and women’s datasets. In this process, there were significant reductions in sample sizes. Again, this may introduce biases and loss o f accuracy, as in the case o f IAP datasets. However, analysis was conducted based on both the full sample and the reduced sample, as presented in later sections. Table 3-3 following summarizes the ARI evaluation and incidence data. The ARI incidents vary widely. Nonetheless, the range is consistent with previous studies on China’s rural and semi-rural population. For example, a study by Zhang (1996) in a semi-rural area near Beijing city reported ALRI incidence between 6% and 8% among children under five. 47 Non-serious Acute Upper Respiratory Infections (AURI) include the common cold, sinusitis, tonsillitis, otitis media and pharyingitis. Potentially life-threatening Acute Lower Respiratory Infections (ALRI) include pneumonia, bronchitis, bronchiolitis and laryngitis (Lanata et al, 2004). 78 Reproduced with permission o f the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3-3: Summary Statistics for ALRI Evaluation and Incidence Province Group S+B B Gansu C ARI Sample* for Individuals evaluated # Children evaluated 66 96 AURI cases 137 400 ex ante 177 104 ex post 223 33 ALRI cases 89 18 ex ante 28 13 ex post 61 5 Child-times evaluation 1343 943 ex ante 397 577 ex post 546 766 times/child 13.99 14.29 ex ante 6.02 6.01 ex post 7.98 8.27 Crude AURI rate 29.8% 14.5% ex ante 30.7% 26.2% ex post 29.1% 6.0% Crude ALRI rate 6.6% 1.9% ex ante 4.9% 3.3% ex post 8.0% 0.9% All S+B B Guizhou C both before and after the interventions 69 88 250 103 85 128 677 128 465 617 74 301 582 285 199 54 376 632 180 418 120 46 24 13 0 4 45 43 0 1 75 3 0 23 9 all B Inner Mongolia all C 257 1230 558 672 70 44 26 116 604 248 356 15 11 4 92 324 135 189 1 0 1 208 928 383 545 16 11 5 3340 1548 1792 13 6.02 6.97 36.8% 36.0% 37.5% 2.1% 2.8% 1.5% 2186 806 1380 18.84 6.95 11.9 27.6% 30.8% 25.8% 0.7% 1.4% 0.3% 1679 637 1042 18.25 6.92 11.33 19.3% 21.2% 18.1% 0.1% 0.0% 0.1% 3865 1443 2422 18.58 6.94 11.64 24.0% 26.5% 22.5% 0.4% 0.8% 0.2% Matched ARI Sample* , merged with household, women and children’s survey datasets # Children evaluated 45 25 115 26 40 60 126 45 232 56 48 434 669 AURI cases 339 627 187 ex ante 46 27 277 99 168 313 111 139 ex post 21 392 133 10 314 295 171 76 ALRI cases 50 5 63 19 19 38 8 0 ex ante 16 5 2 23 18 0 1 19 64 342 139 203 12 10 64 227 88 139 1 0 128 569 227 342 13 10 1370 666 704 15.57 7.57 8 49.4% 45.2% 53.4% 1.0% 0.6% 1.3% 3654 1640 2014 14.62 6.56 8.06 3.5% 35.5% 31.4% 3.3% 2.7% 3.7% 914 414 500 13.25 6 7.25 14.0% 17.9% 10.8% 5.0% 10.4% 0.6% 79 1287 624 663 12.5 6.06 6.44 36.1% 45.7% 27.1% 0.0% 0.0% 0.0% 1139 510 629 13.4 6 7.4 54.2% 39.0% 66.5% 2.1% 0.2% 3.7% Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3-3 continued. ex post Child-times evaluation ex ante ex post Times/child ex ante ex post Crude AURI rate ex ante ex post Crude ALRI rate ex ante ex post 34 602 259 343 13.4 5.8 7.6 38.5% 16.4% 38.8% 8.3% 6 .2 % 9.9% 0 336 144 192 13.4 5.8 7.7 16.7% 13.7% 5.2% 1.5% 3.5% 0 .0 % 6 40 1 0 18 19 2 686 1624 745 879 14.1 6.5 7.6 38.6% 19.3% 35.7% 3.9% 3.1% 4.6% 349 156 193 13.4 511 246 265 1674 762 912 13.3 1201 1 2 .8 814 360 454 13.6 6 .0 6 .2 342 344 15.2 7.6 7.6 49.4% 24.5% 49.7% 1 .2 % 0 .6 % 1.7% 7.4 13.8% 7.7% 10.9% 5.4% 11.5% 0.5% 80 6 .6 36.6% 21.7% 28.7% 0 .0 % 0 .0 % 0 .0 % 6 .0 7.6 53.3% 17.1% 65.0% 2.3% 0.3% 4.0% 6 .1 7.2 40.0% 16.5% 43.0% 2.3% 2.5% 2 .1% 442 759 18.8 6.9 11.9 28.5% 1 1 .6 % 26.7% 1 .0 % 2.3% 0.3% 1 1189 451 738 18.6 7.1 11.5 19.1% 7.4% 18.8% 0 .1% 0 .0 % 0 .1% 3 2390 893 1497 18.7 7.0 11.7 23.8% 9.5% 2 2 .8 % 0.5% 1 .1% 0 .2 % The household and energy use characteristics for the ARI basic sample are similar to those for the IAP basic sample. There are, however, several important differences. In Guizhou, the ARI sample households used, on average, much less biomass fuel than the IAP sample households. In Inner Mongolia, considerably fewer households reported using unimproved biomass stoves in their homes; further, lower rates o f diagnosed respiratory diseases among women were reported in the ARI sample households than in the IAP sample. Before the interventions, children five and under experienced AURI some 20%50% of the time; they experienced ALRI between 1%-10% of the time. ALRl rates were much higher for the stove and behavioral intervention groups than for the control groups. After the interventions, mean AURI rates decreased in Guizhou and Inner Mongolia; the situation in Gansu was mixed. Change in mean ALRI rates showed no consistent pattern, although the ALRI rates were still much higher in the intervention groups than in the control groups. 3.4 Methodology The ARI indicators are directly relevant to estimating the public health benefits from the project interventions. PM concentrations were used to corroborate the ARI conclusions, with the caution that the PM sample sizes were smaller and they cannot be translated into exposure to IAP due to lack o f information about children’s time-activity allocation.48 ARI indicators include AURI and ALRI. AURI carries little cost, as illness PM, CO and SO2 concentration data were analyzed in detail for all the provinces in Zhou et al (2006). Conclusions were drawn separately by province, season, and location inside the house. The analysis in this paper is different from the Zhou et al paper in the following aspects. First, overall changes in average PM concentration were analyzed for each province, controlling for season and location. Second, PM datasets only included households that have observations both before and after the interventions to control for household fixed effects. Third, additional climate, household and women’s characteristics were controlled for in this paper, but not in Zhou et al (2006). 48 81 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. is limited mostly to coughing and the common cold. However, it can be used as an indication o f the effectiveness o f the interventions in reducing emissions. ALRI costs include increased mortality and the loss o f healthy lives. ARI indicators are used as dependent variables, including both the one-time evaluation results for each individual, and the incidence rates for the same individual over time. The analysis is based on a difference-in-difference (DID) approach combined with both parametric and non-parametric models. The standard DID approach compares the after intervention outcome between the treatment group and the control group, but adjusting for differences in the initial, pre-intervention period between the treatment group and the control group. To obtain a simple DID estimate o f the average treatment effect, a standard fixed effects linear regression model49 and a standard random effects linear regression model were used. Further, the balanced subsets o f the original unbalanced panel datasets were used. A robust variance matrix estimator was employed to adjust for possible heteroskedasticity or serial correlation50. Following the fixed effects and random effects linear probability model, a conditional fixed effect logit model was estimated.51 The fixed effects model is chosen over the random effects model because the latter assumes that individual effects are not correlated with treatment. This condition is likely violated when there are significant differences in household, mothers’ and children’s characteristics between the treatment and control groups. The fixed effects model, on the other hand, allows the treatment status to be correlated with individual effect, and only requires that treatment be uncorrelated with deviations from the individual effect. 49 50 The robust variance matrix estimator is valid in the presence o f any heteroskedasticity or serial correlation, provided that T is small relative to the number o f individuals (Wooldridge, 2001). 51 An individual fixed effect logit model leads to an incidental parameters problem. To solve this problem, a conditional fixed effect logit model employs a sufficient statistic, conditioning on the whole set of individuals rather than each individual. 82 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. i:i = /8. + A - G, +| f f A+ C , + » „ (3.1) Where Yit is (1) For the ARI analysis, a zero or one variable for each evaluation o f the child. (2) For the PM analysis, the PM concentration measurement for each location within the household. M s a time dummy variable, indicating time o f ARI evaluation and PM measurement. X it is a set o f time varying individual characteristics. G,. is a dummy variable for treatment status. C( is a set o f time invariant individual characteristics. w(. is the idiosyncratic error. /?, is the average treatment effect. In the ARI and PM analysis, changes in fuel consumption were controlled for as covariates. Other household, mothers’ and children’s characteristics were assumed to remain unchanged. The average treatment effects in terms of ARI were estimated for the different age groups. In the following set o f analysis, changes in ARI incidence were used as outcome variables, rather than evaluation results (which were recorded as simply zero or one). The mean ARI rates per child were calculated for before and after the interventions, and the difference between the two for each child was taken as the new outcome variable. Similarly, the mean PM concentration per household-location was calculated for before and after the interventions and the difference between the two periods was taken as the 83 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. new outcome variable. In this manner, the model employed also eliminates the unobserved individual effect. Ay* = A + f t ■G, + + All* (3.2) Where AYit (1) In the case o f the ARI analysis, is the difference in ARI incidence rates for the same individual before and after the interventions. (2) In the case o f the PM analysis, is the difference in mean concentration for the same location (kitchen, living room or bedroom) in the same household before and after the interventions. A X it is the difference in time varying individual characteristics before and after the treatment. Auit is the change in ui . t , G, and are as previously defined. In the ARI analysis, changes in the quantity o f fuel use before and after the treatment were controlled for. Differential treatment effects were allowed for reflecting children’s gender and age. In the PM analysis, changes in the quantity o f fuel use before and after the interventions were controlled for, as well as changes in temperature, humidity, pressure, wind and weather. Season, fuel and stove type were controlled for as time invariant variables. The average treatment effects (ATE) as per the ARI and PM analyses were estimated for each o f the three provinces. Using the same outcome variable, the analysis also employed a non-parametric matching method to estimate the treatment effects. Matching is widely used in the 84 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. program evaluation literature, so as to eliminate any relationship between assignment of treatment and individual effects through selecting, duplicating and dropping observations from the original dataset. A procedure developed by Abadie and Imbens (Abadie and Imbens, 2001) was employed, which implements nearest-neighbor matching on the Mahalanobis distance with bias adjustment. The Mahalanobis distance adjusts for the difference in scales o f the matching covariates. The bias-corrected matching estimator adjusts the difference within the matches for the differences in their covariate values. To estimate the average treatment effect, the analysis imputed the unobserved outcomes - outcome if there were no treatment for the treatment group, and outcome if there were treatment for the control group. The basic idea behind matching estimators is to impute the missing outcome by finding other individuals in the data whose covariates are similar but who were exposed to the other treatment. Abadie and Imbens (2002) show that the matching estimator o f the treatment effect has a term corresponding to the difference in covariates between pairs o f matched units. This term can be estimated based on two regression functions (Rubin 1973). Abadie and Imbens (2002) approximate these regression functions by linear functions and estimate them using least squares on the matched observations. Abadie and Imbens (2001) show that the adjustment term can be estimated as follows: A A / * » ( * ) = For P A coo + P a \ (3.3) x > co = 0,1 indicating the treatment received, where A ( P A o , o ) = ar§min Z K i:W=a> m (0 •(Xt ~ P »o “ P v i X i ? (3.4) 85 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Where 7 is the observed outcome, X t is a set o f covariates, K u (/) is the number o f times the unit is used as a match. Based on the estimated regression functions, the missing potential outcomes are predicted as: Yi 7 ,(0 ) = — if 1 — 2 (1 + M * ,) - M X ,» if Wt = 0 V,=\, ( 3 ' 5 ) 0 ) leJMU) And if 7,(1) = # J t f ( 0 l e J u (i) r, if w, = o , w ,= i W ith corresponding estimator for the average treatment effect: C =^72(F(i)-F(°)) N tt J M(i) is the set o f indices for the matches for unit i that are at least as close as the Mth match. Wi are dummy variables indicating treatment status, 0 for control and 1 for treatment. Yt (0) is the outcome without treatment; and Yt (1) is the outcome with treatment. The Abadie and Imbens procedure has two advantages: (1) it does nearest neighbor matching based directly on the matching covariates rather than indirectly on the covariates via the propensity score, hence it calculates the correct standard errors without having to adjust for the variance due to the estimation o f the scores and the matching itself; (2) the procedure implements bias adjustment, which is particularly important for this study because the sample sizes are small and the analysis is unlikely to achieve exact matches on important attributes. 86 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. The primary assumption that underlies matching is the conditional independence assumption (CIA), which posits that the treatment status is random conditional on some set of observed X variables or covariates. The selection o f the matching covariates was based on observed differences between the treatment and control groups, regression results on the effect o f control variables on incidence o f ARI, and on theory and earlier work in the field o f IAP and respiratory infections. The validity o f the matching covariates was examined through placebo tests using pre-treatment data. Based on the results of these tests, the observations were matched on a set o f household, stove, fuel and mothers’ and children’s characteristics, and initial levels o f ARI. Without initial levels of ARI, it is implicitly assumed that time trends in ARI are the same for both treatment and control groups. With initial levels o f ARI, this assumption is relaxed. The placebo tests show that the model is not valid without matching on the initial levels o f ARI. Specifically, the children were first matched exactly on the basis o f age; the literature indicates large differences in ALRI incidence among children o f different age groups (children under one year o f age have a significantly higher incidence o f ALRI). The children were then matched according to per capita household consumption and storage of agricultural products, tobacco smoking in the household, fuel consumption, stove type, chimney type, mother’s education, mother’s cooking time, mother’s history o f respiratory infections and children’s general health, gender and children’s exposure time to smoke from cooking. The matching covariates affect ARI incidence either through affecting indoor air quality or through exposure to it. It is obvious that household fuel and stove characteristics affect indoor air quality due to differences in emissions and fuel 87 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. efficiency. The characteristics o f mothers mainly affect stove use practice, and exposure. Consumption and storage o f agricultural products is a proxy for income as well as nutrition, which affects indoor air pollution in both observable and unobservable ways, such as cooking time, heating time, quality and quantity o f fuel and stove, ventilation and housing structure. Children’s health and exposure characteristics affect their vulnerability to infections. Given the relatively small sample sizes and hence the lack o f comparison observations, the analysis conducted single nearest neighbor matching with replacements to keep the matching discrepancy small. The analysis also corrected for heteroskedasticity in the treatment effect due to differences in covariates. This was achieved through matching individuals within the treatment groups and, similarly, matching individuals within the control groups; again, the analysis undertook one match. Finally, the analysis employed a bias-corrected matching estimator, based on the same set o f covariates as used for matching. In the PM analysis, the observations were matched on a similar set o f household and women’s characteristics and initial levels o f PM. These characteristics include cooking room separation characteristics, stove and fuel characteristics, daily stove use time, family size, per capita household annual cash income, per capita consumption and storage o f agricultural products, women’s education, and smoking in the house. 3.5 Results Table 3-4 presents the average treatment effects (ATE) for the interventions. The average treatment effect was estimated with ALRI, AURI and PM indicators as outcome variables for two types o f interventions. In Specification I, a fixed effects linear 88 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. probability model with robust and clustered standard errors was employed. The ARI evaluation outcomes were used as right-hand-side variables, and fixed effects were estimated for each individual. No control variables were included. Balanced panel datasets were utilized, excluding individuals that only had observations before or after the interventions, but not in both periods. For all intervention groups in Guizou, there was significant reduction o f both ALRI and AURI risks. In Gansu, the treatment groups experienced significant reductions o f ALRI and AURI risks with one exception. There was no significant reduction o f risk in Inner Mongolia. Under Specification II, the datasets were used that include individuals with full information from household, women’s and children’s surveys. A fixed effects model was employed controlling for time o f evaluation and change in fuel use. The results remain largely unchanged, with few differences in significance levels. Under Specification III, the same datasets as for II were used, but a random effects model was estimated controlling for a set o f household, women’s and children’s characteristics. The results remain largely unchanged. The coefficients o f the control variables mostly have sensible signs and some are significant. The magnitudes o f the coefficients vary a great deal. Some control variables were consistently significant across provinces. Per capita consumption and storage o f agricultural products lowers ARI incidence. As mentioned before, this is likely a combined measure o f income and nutrition. Increases in both income and nutrition results in lower levels o f ARI. Coal consumption in the heating season increases ARI. Improved biomass stoves reduce ARI incidence. W omen’s 89 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. cooking time and child exposure time increase ARI incidence. Separation o f the cooking room reduces ARI. Old biomass stoves increase ARI. Under Specification IV, the datasets in Specification II and III were utilized to calculate the before-after difference between the mean ARI incidence rates for each individual in the sample. A linear probability model was estimated with control variables for change in fuel use, age and gender. All estimates remained largely the same as in previous specifications. Under Specification V, again using the datasets employed in III, matching was conducted regarding individuals’ initial levels o f ARI rates and the set o f covariates listed earlier. Heteroskedasticity in ATE in variance estimation was allowed, which yielded more significant results compared to assuming a constant treatment effect and homoskedsticity. Sensitivity analysis was conducted through varying the numbers of matches and the matching covariates. For Guizhou and Inner Mongolia, the ATE estimates are robust under various model specifications. In general, increasing the number of matches makes the ATE estimates less significant. This suggests that with a small dataset and a relatively large number o f covariates, increasing the number of matches increases the matching discrepancies and makes the estimates less precise. Specification V reports, therefore, results with a single match, bias correction and heteroskedasticity in variance estimation. In terms o f stove and behavioral interventions combined, the ATE estimates for Gansu for ALRI and AURI become both negative. All other estimates were similar to those under previous specifications. 90 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Table 3-4 Estimates of ATE for Stove and Behavioral Interventions: ARI and PM as Right-Hand-Side Variables Outcome Specification I II III IV V 1288 (90) 1288 (90) 90 90 .025 (.023) -.086** (.034) 270(176) .030 (.029) -.003 (.060) 270(176) .032 (.031) -.008 (.062) 52 .036 (.034) -.0 1 2 -.055* (.031) _ 145** (.070) 52 (.069) 52 -.349 (.172) 2834(274) “ 142*** (.037) - 3 3 4 *** -.493* (.254) 1163 (100) -.145*** (.0 2 1 ) _ 405*** - 546*** (.995) 1163 (100) _149*** -.404* (.2 1 0 ) -286*** (.105) (.048) 308(154) .126 (.161) (.044) 194(102) -.185 (.331) Stove and Behavioral Intervention Group Gansu sample size 1 2988 (232) ALRI overall AURI PM sample size # o f measurements (#locations) 2 coef. Guizhou sample size ALRI AURI PM sample size Coef. Behavioral Intervention Group Gansu sample size 2806 (242) ALRI overall -.032*** (.0 1 1 ) - 2 7 7 *** AURI PM sample size coef. Guizhou sample size ALRI AURI PM sample size Coef. Inner Mongolia sample size ALRI AURI PM sample size PM Coef. (.046) 294 (192) -.096 (.139) 2986(278) -.035*** (.008) - 4 5 4 *** (.050) 272(160) .225 (.2 0 1 ) 4458(467) -.0 1 2 (.009) -.031 (.026) 210(126) -.426 (.273) 1034(84) -.045** (.017) -.284*** (.059) 294(192) -.126 (.156) 1600 (144) -.026*** (.006) -.499*** (.046) 160(104) .167 (.186) 1916(198) -.0 2 1 *** (.006) -.038 (.036) 210(126) -.473 (.333) 100 128*** 100 (.025) -.370*** (.050) 50 -.168** (.069) (.035) _ 3 4 3 *** -.034** (.014) _ 7 1 4 *** (.073) 50 .987* (.504) (.056) 50 -.023 (.240) 1032 (70) -.047** (.018) -.268*** (.061) 56 .006 (.071) 1325 (116) -030*** 70 -.025 (.016) - 330*** (.068) 56 -.062 (.141) 116 -030*** 70 -.049*** (.009) -.282*** (.069) 56 -.190** (.080) 116 -024*** (.007) - 4 3 5 *** (.0 1 0 ) - 451*** (.065) 46 .935* (.503) 126 (.008) -.422*** (.053) 45 -.278 (.236) 126 -.006 (.014) -.041 (.077) 48 -.160 (.116) (.057) 49 -.098 (.067) 2390(126) . 0 1 9 *** (.006) -.040 (.037) 52 -.054 (.104) _ -.0 2 1 (.016) -.074 (.051) 52 -.168 (.260) R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. notes: 1.Each individual was examined 6 times before the interventions and 8 times after, hence there were some 14 observations for each individual. 2.The total number o f measurements for all measurement points in all sample households. Each household had 2-3 measurement points (in the bedroom, living room and kitchen or cooking room). Each household was measured twice before the interventions and twice after. The total number o f measurements locations in all sample households are reported in parenthesis. *** statistically significant at 1 % level. ** statistically significant at 5% level. * statistically significant at 1 0 % level. ARI specifications: I. Fixed effects model, all original data, no controls. II. Random effects model, datasets include households with full survey information, controlling for a set o f household, women’s and children’s characteristics. III. Fixed effects model, balanced panel data with full survey information, controls including time o f evaluation and change in fuel consumption. IV. Linear model, differencing o f mean ARI rates before and after the interventions, controls for change in fuel consumption, differential effects for gender and age groups. V. Matching model, differencing o f mean ARI rates, match on initial levels o f ARI rates, per capita consumption and storage o f agricultural products, baseline level fuel consumption (biomass and coal), stove type, cooking location, smoking in the house, mother's age, education, occupation, cooking time and history o f being diagnosed with respiratory disease, child's gender and exposure time to cooking smoke, and general health status, exact matching on child’s age. PM specifications: I. Fixed effects model, controls for season and sampling point all original data. II. Fixed effects model, controls for pressure, weather, wind, humidity, season and sampling point. III. Random effects model, controls for cooking location, type o f stove, daily stove use time, size o f living space, type o f house material, indoor temperature, women’s education, cooking time and tobacco smoking, plus those in II. IV Linear model, differencing o f mean PM concentration, controls for change in fuel consumption before and after the interventions V. Matching model, differencing o f mean PM concentration, match on cook room separation, type of stove, daily stove use time, fuel consumption, per capita consumption o f agricultural products, type of house material, age o f stoves, women's education, cook time and tobacco smoking. For estimations using PM indicators as outcome variables, Specification I used all original data and estimated a fixed effects model with robust and clustered standard errors for household-measurement locations in heating and non-heating season. In Specification II, some covariates that varied over time were included in the same framework as in I, such as pressure, weather, wind and humidity. In Specification III, a random effects model was estimated, controlling for a set of household, stove, fuel and mother’s characteristics in addition to those included in Specification II. In Specification IV, the before and after difference in mean PM concentration was calculated for each sampling point within the households, both during the heating and non-heating seasons. 92 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. A linear model was estimated with robust and clustered standard errors for the sampling points within each household, controlling for change in fuel consumption before and after the interventions. In Specification V, changes in mean PM concentrations were used as dependent variables and households were matched according to a set o f relevant factors (e.g., house material, cooking location, per capita consumption and storage o f agricultural products, stove and fuel type, age o f stoves, daily stove use time, tobacco smoking in the house and women’s education and cook time). Again, the analysis undertook single match and bias adjustment and allowed for heteroskedasticity in treatment effects. The results for PM concentrations were generally less significant than those for ARI, possibly due to smaller sample sizes. While they were largely consistent with the ARI results for Gansu and Inner Mongolia, they were less so for Guizhou. From the full set o f estimates, the ATE estimates, with few exceptions, were quite robust under different model specifications. The simple DID estimator without matching eliminates individual effects that are constant over time. However, it may not adequately control for variables correlated with both the outcome variables and the treatment assignment. In particular, pre-existing time trends are not controlled for under the simple DID. While the matching estimator is not subject to the bias caused by pre-existing time trends, it is subject to the assumption that the assignment o f treatment is random conditional on the observable matching covariates. The validity o f this assumption could not be tested. 93 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. The results with ARI indicators as outcome variables were generally consistent across the five specifications. In terms o f stove and behavioral interventions combined, the risk reduction for ALRI ranged from 5.5 percentage points to 14.9 percentage points. The risk reduction for AURI ranged from 14.5 percentage points to 71.4 percentage points. In terms o f behavioral intervention alone, the ALRI risk reduction ranged from 1.9 percentage points to 4.9 percentage points. The AURI risk reduction ranged from 26.8 percentage points to 49.9 percentage points. The results for the behavioral intervention groups were more significant and consistent than for the stove and behavioral intervention groups. Table 3-5 summarizes the differences in ATE estimates between the stove and behavioral intervention groups and the behavioral intervention groups. The differences indicate the marginal benefit gained from adding stove improvements to behavioral intervention programs. In Guizhou, there is consistent and strong evidence that adding stove improvements to behavioral interventions resulted in further significant reductions in ALRI, ranging from 3.1 percentage points to 11.7 percentage points. The evidence regarding AURI is mixed. In Gansu, the fixed and random effects models yield very different results from the matching model. Since the evidence for Gansu is conflicting, it cannot be concluded with confidence that the marginal benefit o f stove interventions in that province was different from zero. Apart from estimation uncertainty, there are several possibilities that may explain the lack o f marginal health benefits from adding stove interventions in Gansu. First, high baseline levels o f IAP may have been a factor. In Gansu, both the ex ante and ex post PM concentrations were much higher than in Guizhou province. Unless pollution levels 94 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. fall below some band, the incidence o f ARI may remain little changed. Second, while the indoor air quality may have significantly improved, the ambient air quality would continue to be associated with ARI symptoms.52 Third, moral hazard may have played a role. Household members may have falsely concluded that the improved stoves solved the IAP problems and therefore took less precaution than they would have done in the absence of an improved stove. For example, they may have been close to the stove more often and opened windows for ventilation less often. Table 3-5: Estimates of Difference in ATE between the Stove and Behavioral Intervention Groups and Behavioral Intervention Groups _____ Outcome Gansu sample size 1 ALRI overall Specification I II III IV 2600(220) 751(61) 751(61) 61 .058*** .073** .075** .057 V 61 179*** (.016) .190 (.031) (.032) .282*** (.060) (.034) .269*** (.063) (.051) .280*** (.089) (.038) .516 (.609) 288 (184) -.386 (.247) 278 (176) -462*** 80 (.106) 80 -.387 (.233) 923 ( 6 6 ) 923 ( 6 6 ) 66 AURI 288(184) -.352** (.175) 2800 (277) -.107*** (.013) JY Q * * * (.035) 352 (154) -.115*** (.024) -.062 (.067) PM sample size - 117*** (.0 2 0 ) .103* (.055) 226 (106) 226 (106) -.074* (.039) .1 2 1 * (.064) 77 -.055 -.155 .150** -.042 -.031* (.017) -.047** (.071) 77 .127 (.059) (.181) (.063) (.076) (.083) AURI PM sample size # o f measurements (#locations) 2 coef. Guizhou sample size ALRI Coef. -.0 0 1 (.152) 66 52 Several studies in North American cities and one study in Bangkok, Thailand, have reported associations between ambient air pollution and respiratory symptoms. The studies in North America include Dockery DW, 1994, Katsouyanni, 1997, US EPA 1996, and Vichit-Vadakan, 2001. 95 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. 3.6 Benefit/Cost Analysis The economic value of reducing risk o f ALRI and AURI can be calculated through applying willingness to pay (WTP) and value o f statistical life (VSL) estimates for China. There is an extensive literature on the value o f reducing risks o f mortality and morbidity in the US and other industrial countries, but there have been few estimates using survey data from developing countries. Most benefit valuation studies for developing countries use a cost-of-illness approach, or have adapted VSL estimates from developed countries adjusting for differences in income. The cost-of-illness approach is a valuable alternative measure when there is little information on VSL. However, the approach has little basis in economic theory, as it takes the view that people are producers and assumes that improvements in health equate to the sum o f reductions in labor market earnings and savings in health care expenditures (Berger, et al, 1994). In this paper benefit valuation is based on a study by Hammit and Zhou (2005) in China, as it seems to provide the most recent WTP and VSL estimates for the populations concerned. The Hammit and Zhou (2005) study is one o f the first that directly estimates the economic value o f reducing health risk in developing countries. Hammit and Zhou estimated the economic values for preventing cold, chronic bronchitis and fatality through a contingent valuation (CV) study. The CV surveys were conducted in Beijing, Anqing and the rural areas near Anqing, to represent populations o f a large city, a small city and a rural area o f China. For rural areas in China, the mean VSL is estimated to be about $100,000 to $180,000, although estimates o f the VSL are sensitive to modeling choices. This is much smaller than for the US and other industrialized countries.53 The 53 Viscusi and Aldy conclude that the most reasonable values are $4 million to $9 million for the US. 96 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. estimated sample-average median WTP to prevent a cold ranges from $3 to $5, and the mean WTP ranges from $4 to $1054. Again, these estimates are much smaller than estimates for the US and other countries55. The benefits from the project interventions are calculated in two parts: (1) the reduction in mortality due to ALRI; and (2) the benefit from avoiding AURI. The reduction in ALRI risk is converted to reduction in mortality. According to statistics for China, the five and under mortality rate due to ALRI is 4.15 per 1000.56 Based on this, mortality due to ALRI for the treatment groups was calculated adjusting for the number o f children in each group. The percentage reduction o f ALRI risk was then applied to obtain reduction in mortality due to the interventions. The reduction in mortality was then valued according to the VSL estimates in the Hammit et al (2005) studies to give.the economic value o f ALRI risk reduction. This benefit evaluation procedure is subject to the implicit assumption that the risk reduction rate due to interventions is constant over a year. Previous studies have shown seasonal variations in ALRI. In Northern China, ARI incidence peaks in December and January (Zhang et al, 1986). Whether the risk reduction rate is also subject to seasonal variations is unknown. Another limitation is that the mortality and ALRI rates for China include both urban and rural areas. Given the fact that the intervention groups were at low income levels in rural or semi-rural setting, the mortality rate and the contribution o f ALRI could be higher than assumed in the analysis, with the result that the benefits were underestimated. The average out-of-pocket expenses to treat the respondent’s last cold are $2. Other studies have valued one-day avoidance o f cold as between $10 and $150 for the US, $4 and $24 for Bangkok o f Thailand, and $40 to prevent a cold in Taiwan (see Hammit, et al 2005). 56 The five and under mortality rate is 31 per 1,000 live births and the percentage o f mortality due to ALRI for this age group is 13.4% (WHO 2006). 54 55 97 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. In terms of AURI, each evaluation outcome is assumed to be one episode o f common cold. Baseline data was used to calculate the annual number o f cold episodes among children in the treatment groups. As mentioned earlier, cold incidence peaks in December and January. While the evaluations were undertaken in April, May and June, it is unknown whether the evaluations were representative o f the annual average. The effect o f AURI reduction for children in the treatment group was calculated as episodes o f cold avoided per year. An episode o f cold was then valued according to the Hammit et al (2005) study. The estimated risk reductions in terms o f ALRI and AURI vary across model specifications. In calculating the health benefits, the lower bound was estimated based on the lowest estimated risk reductions for ALRI and AURI. Table 3-5a, 3-5b and 3-5c present benefit valuation for the stove and behavioral intervention groups, and the behavioral intervention groups. In terms o f the stove and behavioral intervention groups, the annual benefit estimates for both Gansu and Guizhou were similar, in the range of US $300-700 per household. In terms o f the behavioral intervention groups, the benefit estimates for all provinces fall in the range o f US $300900. The estimated percentage reductions o f ALRI and AURI rates range from 30% to unrealistically high rates. The latter are probably anomalies caused by problems o f diagnosis and unexplained increases in ALRI and AURI rates for the control groups. In terms o f marginal benefit from adding stove intervention to behavioral intervention, the analysis was only conducted for Guizhou, as there was no consistent 98 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. evidence showing that the marginal benefit is different from zero in Gansu. The estimated marginal benefit for Guizhou was between $30 to $40. Table 3-5a: Benefit Analysis for Stove and Behavioral Intervention Groups Benefit Analysis for ALRI Baseline Rate o f ALRI Reduction o f ALRI incidence in Percentage Points Percentage Reduction from Baseline Rate Total # children Under Five Mortality Attributable to ALRI (per 1 , 0 0 0 births) 1 Mortality Reduction for Treatment Group Valuation o f benefit using VSL $1 0 0 ,0 0 0 $180,000 Benefit Analysis for AURI Baseline Incidence rate o f AURI Reduction o f AURI incidence in percentage points Percentage reduction from baseline rate Baseline Annual AURI cases2 Annual AURI cases reduction Valuation o f benefit using WTP $3 $10 Total Benefit (USD) Total # households with children over five Total Benefit per Household (USD) Gansu Guizhou 6 .2 101 11.5% 3.4% 30% 136 10.55 0.37 10.55 0.17 37,219 66,993 16,703 30,065 35.5% 14.5% 41% 740 107.3 36% 34% 95% 576 213.12 907 3,023 3 8 ,1 2 6 -7 0 ,0 1 6 98 3 8 9 -7 1 4 1,646 5,488 18 ,349-81,808 128 344 - 639 % 5.5% 89% Notes: 1. Source: The World Health Statistics 2006, WHO 2. We assume that the AURI incidence obtained from the baseline evaluation (April, May and June) period is representative o f the AURI incidence over a year. 99 Reproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Table 3-5b: Benefit Analysis for Behavioral Intervention Groups Benefit Analysis for ALRI Baseline Rate o f ALRIa Reduction o f ALRI incidence in Percentage Points Percentage Reduction from Baseline Rate Total # children Under Five Mortality Attributable to ALRI (per 1 , 0 0 0 births) 1 Mortality Reduction for Treatment Group Valuation o f benefit using VSL $1 0 0 ,0 0 0 $180,000 Benefit Analysis for AURI Baseline Incidence rate o f AURI Reduction o f AURI incidence in percentage points Percentage reduction from baseline rate Baseline Annual AURI cases2 Annual AURI cases reduction Valuation o f benefit using WTP $3 $10 Total Benefit (USD) Total # households with children over five Total Benefit per Household (USD) Gansu Guizhou Inner M ongolia 0.03 0.045 1 0 0 %b 0 .0 2 10.55 0.037 0.024 65% 136 10.55 0.42 0.65 0.47 41,955 75,520 64,565 116,217 47,356 85,240 0.36 0.27 75% 540 408 0.36 0.343 95% 576.00 548.80 1,223 4,077 43,178 79,597 107 4 0 4 -7 4 4 1,646 5,488 66,211 121,705 132 502 - 922 101 0.019 83% 138 10.55 47,356 85,240 134 353 - 636 Notes: a Due to the small sample sizes the ALRI rates were found to be zero in some cases. For purposes o f the analysis, average ALRI rates for the whole province were used, b The percentage reduction was constrained to less than 100. 100 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Table 3-5c: Marginal Benefit for Adding Stove in Intervention Guizhou Benefit Analysis for ALRI Reduction in ALRI incidence Total # children Disease burden in DALY per 1000 population attributable to ALRI (China) 2 Reduction in Loss o f DALY Valuation o f benefit using VSL $1 0 0 ,0 0 0 $180,000 Benefit Analysis for AURI Reduction in AURI incidence Baseline Annual AURI cases 1 Annual AURI cases reduction Valuation o f benefit using WTP $3 $6 $10 Total Benefit (USD) Total # households with children over five Total Benefit per Household (USD) 0.031 131 7.2 0.03 $2,924 $5,263 0.047 144 27 81 162 271 4,031 - 5,534 128 31-43 Table 3-6 presents cost data obtained from the IAP project document. The total costs for each component were amortized over ten years assuming that the project life is ten years. The total costs are first amortized at 5% reflecting the real cost o f capital, and then at 10% reflecting the social opportunity cost o f capital57. Costs are calculated on a per household basis, to make it comparable with the benefit analysis. There were 500 households in each treatment/control group. However, only around 100 households had children five or under. These costs are a crude measure o f the real cost o f the intervention programs. For the stove and behavioral intervention component, the costs include the costs for designing, purchasing and distributing the improved stoves. The development o f stoves and marketing mechanisms would have spillover effects for populations other than the 57 Social opportunity cost accounts for the opportunities forgone by society, including externalities, in using a resource in some way. 101 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. treatment groups. One can assume that if more households were involved in the project, the average cost per household would have been lower. The cost measure for this component may overestimate the real cost. In terms o f behavioral intervention groups, the costs include development o f training and publicity material, school curriculum on IAP education, expenses for conducting training workshops and awareness raising activities. However, this measure o f cost does not include the opportunity cost in time spent on IAP awareness education. This is particularly important for students and teachers who are involved in the IAP educational programs, and health workers who conducted activities in villages to raise IAP awareness. The cost measure o f this component likely underestimates the real costs. For both intervention groups, the costs per household are much smaller than the lower bound o f the benefit estimates. In terms o f marginal benefit o f adding stove intervention, the cost is greater than the upper bound o f benefit estimate. While there is net benefit from all interventions, behavioral interventions appear to be more costeffective than stove combined with behavioral interventions. Again, this conclusion has to be qualified by the fact that the cost o f stove interventions are likely overestimated and the cost of behavioral interventions are likely underestimated. 102 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Table 3-6. Summary of Project Costs for Each Province Cost (USD) Stove plus behavioral interventions Total Project Cost Annual cost in financial terms over 10 years Cost per household' Annual cost in social terms over 10 years2 Cost per household2 Behavioral interventions Total Cost Annual cost in financial terms over 10 years' Cost per household 1 Annual cost in social terms over 10 years2 Cost per household2 M arginal cost of stove intervention Total Cost Annual marginal cost in financial terms over 10 years' Cost per household 1 Annual cost in social terms over 10 years2 Cost per household2 215,833 27,468 55 34,224 68 37,500 4,764 10 5,952 12 178,333 22,704 45 28,284 57 Notes: 1. Total cost is annualized at 5% reflecting the weighted average real cost o f capital. 2. Total cost is annualized at 10% reflecting the social opportunity cost o f capital. 3.7 Conclusions and Policy Implications This paper has extensively analyzed the data generated from a recent World Bank IAP project in China, with the objective o f determining the health benefits from stove and behavioral interventions to reduce household exposure to indoor smoke. It was found that the IAP interventions resulted in measurable reductions o f risk in ALRI and AURI among children five years o f age and younger. The reductions in AURI were larger than for ALRI. The effects did not differ significantly across gender or age groups. The interventions also resulted in measurable reductions in PM concentrations in the sample households. These conclusions are broadly consistent across the three project provinces included in the analysis, and are robust to model specifications. Our analysis indicates that both stove and behavioral interventions were effective in reducing ALRI and AURL risk from IAP exposure, although the marginal benefit from 103 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. adding the stove intervention does not seem to outweigh the costs. In light o f this finding, an important question from a public policy perspective is the appropriate role of government in promoting these interventions. The economics o f household energy use, particularly the degree o f market failure in developing clean fuel alternatives and new stove technologies in low income countries, indicate room for government intervention in this regard. However, considering the net benefits o f reducing IAP through public interventions to develop and distribute new household energy technologies versus health education, it would appear appropriate to concentrate on health education/behavioral interventions. The lack o f significant marginal health benefit from the stove intervention may also suggest some inadequacies o f the stove intervention programs. Most importantly, the improved stoves may not lead to improvement in ambient air quality. Research is needed to determine the relative role o f indoor and ambient air pollution in causing ARI. Research is also required concerning the concept o f IAP/ARI thresholds, and the relative benefits o f alternative policy interventions under differing baseline circumstances. Further, the issue o f “moral hazard” o f stove users needs to be examined, i.e. installation o f improved stoves may cause household members to take less precautions to reduce their exposure to IAP. There are several caveats to bear in mind when applying the conclusion's in this study to other locations and communities. First, the estimates for average treatment effects (ATE) are relatively robust to model specification in terms o f direction, but not magnitude. There is considerable uncertainty in estimating the magnitude o f the benefits. Second, the sample sizes are small and further reduced when various matching 104 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. information is incorporated. This could result in bias and loss o f precision. Third, the sample households were characterized by extreme poverty and low education levels, compared to the provincial averages. Conclusions based on these sample households may not apply to other populations, or need to be adjusted in other applications. Knowing that the IAP interventions in this study generated measurable health benefits is only a step in determining whether they are appropriate in addressing the IAP problem more generally. 105 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. References Abadie A, Drukker D, Herr JL and Imbens GW, Implementing Matching Estimators for Average Treatment Effects in Stata, the Stata Journal Number 1, pp. 1-18, 2001. Benguigui Y, Lopez FJ, Schmunis G, Yunes, Respiratory Infections in Children, Pan American Health Organization, WHO, 1999. Benguigui Y et al, Acute Respiratory Infections, Topics in International Health CD-Rom, The Wellcome Trust, CABI Publishing, 2001. Dockery DW, Pope CA III, Acute Respiratory Effects o f Particulate Air Pollution, Annual Review o f Public Health 15: 107-132, 1994. Ezzati M and Kammen DK, Indoor Air Pollution from Biomass Combustion and Acute Respiratory Infections in Kenya: An Exposure-Response Study, The Lancet, Vol.358, August 2001. Ezzati M, Bailis R, Kammen DM, Holloway T, Price L, Cifuentes LA, Barnes B, Chaurey A, and Dhanapala KN, Energy Management and Global Health, Annual Review o f Environment and Resources. 29.2004. Ezzati M. and Baris E (ed), Household Energy and Health in Rural China: Lessons and Potentials fo r Program Design, World Bank, 2006. Ezzati M, Kammen DM. Household Energy, Indoor Air Pollution and Health in Developing Countries: Knowledge Base for Effective Intervention, Annual Review o f Energy and the Environment, 2002. Ezzati M, Kammen DM, the Health Impacts o f Exposure to Indoor Air Pollution from Solid Fuels in Developing Countries: Knowledge, Gaps, and Data Needs, Environmental Health Perspectives 110:1057-1068, 2002. Ezzati M, Lopez AD, Rodgers A, Hoorn SV, Murray CJL, and the Comparative Risk Assessment Collaborating Group, Selected Major Risk Factors and Global and Regional Burden o f Disease, The Lancet, Vol. 360, November 2002. He G, Ying B, Lui J, Gao S, Shen S, Balakrisknan K, Jin Y, Liu F, Tang N, Shi K, Baris E and Ezzati M, Patterns o f Household Concentrations o f Multiple Indoor Air Pollutants in China, Environmental Science & Technology 39, 2005. Homedes N, the Disability-Adjusted Life Year (DALY) Definition, Measurement and Potential Use”, Presentation to the European Bioethics Conference, October 1995. 106 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Jin Y, Ma X, Chen X, Cheng Y, Baris E, Ezzati M, Exposure To Indoor Air Pollution from Household Energy Use in Rural China: the Interactions o f Technology, Behavior, and Knowledge in Health Risk Management, Social Science and Medicine, 11.029, 2005 Katsouyanni K, Touloumi G, Spix C, Schwartz J. Balducci F. Medina S, Rossi G, Wojtyniak B, Sunyer J, Bacharova L, et al. Short Term Effects o f Ambient Sulphur Dioxide and Particulate Matter on Mortality in 12 European Cities: Results from Time Series Data from the APHEA Project. B r M e d J 314: 1658-1663,1997. Lanata CF, Rudan I, Boschi-Pinto C, Tomaskovic L, Cherian T, Weber M and Campbell H, Methodological and Quality Issues in Epidemiological Studies o f Acute Lower Respiratory Infections in Children in Developing Countries, International Journal o f Epidemiology 33: pp 1362-1372, 2004. McCracken JP, Smith KR, An Annotated Bibliography on Acute Respiratory Infections And Indoor Air Pollution, with Emphasis on Children under Five in Developing Countries. Washington, EDC: Environmental Health Project, US Agency for International Development 1997. Nuntavam Vichit-Vadakan, Bart D. Ostro, Lauraine G. Chestnut, David M. Mills, Wichai Aekplakom, Supat Wangwongwatana, and Noppanpom Panich, Air Pollution and Respiratory Symptoms: Results from Three Panel Studies in Bangkok, Thailand, Environmental Health Perspectives, vol. 109, June 2001. Rehfuess E and Rouse J, Household Energy and Health - Options for Health Monitoring in Intervention Projects, African Regional Training Workshop on Indoor Air Pollution and Household Energy Monitoring, Kampala, Uganda, World Health Organization, June 2005. Sexton K, Ryan PB, Watson AY, et al. Assessment o f Human Exposure to Air Pollution: Methods, Measurements, and Models, Air pollution, the automobile and public health, Washington, DC: National Academy Press, 1988. J. Sinton, K. Smith, J. Peabody, L. Yaping, Z. Xialiang, R. Edwards et. al., An Assessment o f Programs to Promote Improved Household Stoves in China, Energy fo r Sustainable Development, VIII, 33-52, 2004. Smith KR, Bruce N, Arana B, Diaz A, Jenny A, Khalakdina A, McCracken J, Schei M, Gove S. RESPIRE - the Guatemala Randomized Intervention Trial. International Society for Environmental Epidemiology, 2006, Paris, France. Kirk R. Smith, Gu Shuhua, Huang Kun, Qiu Daxiong, One Hundred Million Improved Cookstoves in China: How Was It Done? World Development, Vol. 21, No. 6, 1993. Smith K, Mehta S, Maeusezahl-Feuz M. Indoor Air Pollution from Household Use of Solid Fuels. In: Ezzati M, Lopez A, Roders A et al. Comparative Quantification o f 107 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. Health Risks, Global and Regional Burden o f Disease Attributable to Selected Major Risk Factors, Geneva, World Health Organization, 2004. Smith K, Rogers J, and Cowlin S. Household Fuels and Ill-Health in Developing Countries: What improvements can be brought by LP Gas? Environmental Health Sciences, School o f Public Health, University o f California, Berkeley 2005. Smith K, Samer JM, Romieu I, Bruce N, Indoor Air Pollution in Developing Countries and Acute Lower Respiratory Infections in Children, Thorax 2000. Smith KR, Ulma R, Kishore VVN, et al., Greenhouse Gases from Small-Scale Combustion in Developing Countries: Household Stoves in India. Research Triangle Park, NC: US Environmental Protection Agency, 1999. U.S. EPA. Air Quality Criteria for Particulate Matter. EPA/600/P-95/001cF. Washington, DC: U.S. Environmental Protection Agency, 1996. Wooldridge, JM, Econometric Analysis o f Cross Section and Panel Data, the MIT Press, 2001. World Health Organization, World Health Statistics 2006, WHO website www.who.int 108 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. [...]... through income Isolated from the scale and composition effects, the income variable reflects the demand for and supply o f pollution abatement At low income levels, increases in income are directed primarily towards food and shelter, and have little effect on the demand for environmental quality At higher income levels, increases in income lead to higher demand for environmental quality (since it is... total pollution declines if the form of the utility function satisfies certain conditions Stokey assumes a standard constant inter-temporal elasticity o f substitution utility function When the elasticity o f marginal utility is greater thanl, total pollution declines The proportionate decline in marginal utility from increases in consumption is sufficiently significant that the increase in disutility from... approach to minimize suboptimal provision o f environmental quality as a public good Stokey’s model is a very useful framework for addressing sustainable growth issues Since both income and environmental quality are endogenously determined, the feedback relationships are robust and pollutant emissions more contained The model helps to address sustainable growth issues, offering prospect o f income growth... industrialization intensifies, environmental degradation worsens Yet higher levels o f development and income, however, provide the resources and incentive for more efficient and environmentally friendly technologies and/or constraints, and environmental degradation declines (Panayotou 2000) This inverse U-shaped relationship between economic growth and environmental quality is referred to as the Environmental. .. 1127-1170 Islam, N., Vincent, J., and Panayotou, T., Unveiling the Income-Environment Relationship: an Exploration into the Determinants o f Environmental Quality, Working Paper, Department o f Economics and Harvard Institute for International Development (1999) List, J.A and Gallet, C.A., The Environmental Kuznets Curve: Does One Size Fit All?, Ecological Economics 31 (1999) 409-423 Mankiew N.G.; Romer, D;... China for sharing with me its extensive data set on indoor air pollution derived during a joint study with the World Bank Professor Majid Ezzati, Harvard School o f Public Health, and Dr Enis Baris, World Bank, guided the study and engaged me in the process, providing the foundation for one o f my three dissertation papers Schumpeter’s description o f capitalism as creative destruction brings to mind... normal good) Engel’s curve for environmental quality translates into an inverted-J curve between income and environmental degradation (Selden and Song 1995) On the supply side, higher incomes make available the resources needed for increased private and public expenditures on pollution abatement Further, they induce stricter pollution regulations to help internalize environmental externalities Grossman... pollution and income in developing countries with a fundamentally different, negative one in developed countries, not a single relationship that applies to both categories o f countries” (Vincent 1997) A further source o f concern regarding empirical estimates o f the EKC is the comparability and quality o f available environmental data Stem et al (1996) noted that pollution data used in environmental. .. are significant.1 There is the possible prospect of sustainable development, although it depends on factors o f technology and preference Further, it suggests the need for research into the turning point, the degree o f environmental degradation before this point is reached, including potentially pushing beyond ecological thresholds, and the institutional and policy 1 Although the existence o f the... control variables include initial per capita income levels, initial population levels, initial population densities, income growth rates, population growth rates and a 10 The half-life is log(2)/A, Hence, if X = 0.05 per year, then the half-life is 14 years, which means half o f the initial gap disappears in 14 years (Barrow and Sala-Martin Xavier, 1999) 11 Such an assumption is common in growth models