Indoor Air Quality for Poor Families: New Evidence from Bangladesh Susmita Dasgupta* Mainul Huq M. Khaliquzzaman Kiran Pandey David Wheeler Development Research Group World Bank World Bank Policy Research Working Paper 3393, September 2004 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. Authors' names in alphabetical order. We would like to express our appreciation to the field survey team of the Development Policy Group, for their excellent air-quality monitoring work under difficult conditions. We are also grateful to Kseniya Lvovsky, Maureen Cropper, Douglas Barnes, Bart Ostro and Paul Martin for useful comments and suggestions. Financial support for this study has been provided by Trust Funds through the Knowledge for Change Program of the World Bank's Development Economics Vice- Presidency, and by the Development Research Group. Executive Summary Poor households in Bangladesh depend heavily on wood, dung and other biomass fuels for cooking. This paper provides a detailed analysis of the implications for indoor air pollution, drawing on new monitoring data for respirable airborne particulates (PM10) in a large number of Bangladeshi households. Concentrations of 300 ug/m3 or greater are common in our sample, implying widespread exposure to a serious health hazard. For comparison, Galassi, Ostro, et al. (2000) find substantial health benefits for PM10 reduction in eight Italian cities whose annual concentrations are far lower: 45-55 ug/m3. As expected, our econometric results indicate that fuel choice significantly affects indoor pollution levels: Natural gas and kerosene are significantly cleaner than biomass fuels. However, household-specific factors apparently matter more than fuel choice in determining PM10 concentrations. In some biomass-burning households, concentrations are scarcely higher than in households that use natural gas. Our results suggest that cross-household variation is strongly affected by structural arrangements: cooking locations, construction materials, and ventilation practices. To assess the broader implications for poor Bangladeshi households, we extrapolate our regression results to representative household samples from rural, peri-urban and urban areas in six regions: Rangpur in the Northwest, Sylhet in the Northeast, Rajshahi and Jessore in the West, Faridpur in the Center, and Cox's Bazar in the Southeast. Our results indicate great geographic variation, even for households in the same per capita income group. This variation reflects local differences in fuel use and, more significantly, construction practices that affect ventilation. For households with per capita incomes less than $1.00/day, rural PM10 concentrations vary from 410 ug/m3 in Cox's Bazar to 202 ug/m3 in Faridpur. In urban areas, concentrations for such households differ by almost 100 ug/m3 between the highest areas, Jessore and Rajshahi, and the lowest, Sylhet. The poorest households in Faridpur face a lower mean indoor concentration (202 ug/m3) than the highest-income households in Jessore and Rajshahi (215 ug/m3). Great variation also characterizes the 24-hour cycle within households. For example, within the "dirtiest" firewood-using household in our sample, readings over the 24-hour cycle vary from 68 to 4,864 ug/m3. Such variation occurs because houses can recycle air very quickly in Bangladesh. After the midday meal, when ventilation is common, air quality in many houses goes from very dirty to reasonably clean within an hour. Rapid change also occurs within households: Diffusion of pollution from kitchens to living areas is nearly instantaneous in many cases, regardless of internal space configuration, and living-area concentrations are almost always in the same range as kitchen concentrations. By implication, exposure to dangerous indoor pollution levels is not confined to cooking areas. We find that distinguishing between indoor and outdoor pollution may not be useful in biomass-using areas. In Dhaka, the 24-hour pattern of indoor PM10 concentration for the 1 cleanest fuel, piped natural gas, is nearly identical to the pattern for ambient (outdoor) pollution. Baseline indoor pollution is set by ambient pollution, which varies nearly twelve-fold (from 30 to 350 ug/m3; mean 113) over the 24-hour cycle. In areas with heavy biofuel use, our results therefore suggest that health benefits may be less than expected for households that switch to clean fuels or improved stoves. Our survey also suggests that limited information may be a significant impediment to adoption of cleaner, more efficient stoves. Only 15% of our sample households regard improved stoves as a viable option, either because they have not heard of them or because they do not think they are locally available. Even among families that have considered the option, however, improved-stove use appears quite limited because of concerns about convenience or initial investment cost. The intermediate-term prospects for clean-fuel use appear more hopeful in urban and peri-urban areas, if economic growth continues. Our sample evidence suggests very high adoption rates among families whose daily per capita incomes exceed $2.00. However, only 30% of extreme-poverty households (less than $1.00/day per capita) use clean fuels, even in urban areas where their prices are relatively low. In rural areas, our evidence offers little hope for adoption of clean fuels in the near future because their prices relative to biofuels are too high. However, our analysis also suggests that poor families may not have to wait for clean fuels or clean stoves to enjoy significantly cleaner air. Within our sample household population, some arrangements are already producing relatively clean conditions, even when "dirty" biomass fuels are used. Since these arrangements are already within the means of poor families, the scope for cost-effective improvements may be larger than is commonly believed. 2 1. Introduction Indoor air pollution from burning wood, animal dung and other biofuels is a major cause of acute respiratory infections (ARI), which constitute the most important cause of death for young children in developing countries (Murray and Lopez, 1996). Acute lower respiratory infection (ALRI), the most serious type of ARI, is often associated with pneumonia (Kirkwood et al., 1995). ALRI accounts for 20% of the estimated 12 million annual deaths of children under five, and about 10% of perinatal deaths (WHO, 2001; Bruce, 1999). Nearly all of these deaths occur in developing countries, with the heaviest losses in Asia (42% of total deaths) and Africa (28%) (Murray and Lopez, 1996). Through its effect on respiratory infections, indoor air pollution (IAP) is estimated to cause between 1.6 and 2 million deaths per year in developing countries (Smith, 2000). Most of the dead are in poor households and approximately 1 million are children. The size of IAP's estimated impact has prompted the World Bank (2001) and other international development institutions to identify reduction of indoor air pollution as a critical objective for the coming decade. The current scientific consensus is that most respiratory health damage comes from inhalation of respirable particles whose diameter is less than 10 microns (PM10), and recent attention has focused particularly on fine particles (PM2.5). However, the design of cost-effective IAP reduction strategies has been hindered by lack of information about actual PM concentrations in poor households. Data have been scarce because monitoring in village environments has been difficult and costly. Relative small-scale studies of indoor PM10 exposure from woodfuel combustion have been conducted in Kenya (Boleij, et al (1989, 36 households), Guatemala (Smith, et al., 1993, 60 households), 3 Mexico (Santos-Burgoa, et al., 1998, 52 households), and Gambia (Campbell, 1997, 12 households. Recently, a larger sample of houses has been studied in rural India (Balakrishnan, et al., 2002; Parikh, et al., 2001). In Section 9, we will compare the India results to those obtained by this study. Because monitoring studies are costly, IAP exposure analyses frequently use biofuel consumption data to proxy the degree of exposure to fine particulates, and extrapolate to estimates of ARI prevalence and mortality (Smith, 2000). Although fuel- use data are widely available, this approach implicitly assumes a constant relationship between fuel combustion and indoor air pollution across households. However, the previously-mentioned studies indicate that IAP levels in households with identical fuel use are affected by factors such as the location of cooking (inside/outside), ventilation through windows and doors, and air flow through building materials. Additional information could have a large social payoff in this context, since simple alterations in structures, ventilation practices, building materials and cooking locations may be much less costly than switching to cleaner fuels or investing in clean stoves. This paper provides evidence on PM10 and PM2.5 concentrations in poor households, using new air monitoring data from Bangladesh. Recent technical advances have significantly increased the power, portability and durability of equipment for monitoring particulate pollution. Our study has used two types of equipment: air samplers that measure 24-hour average PM10 concentrations, and real-time monitors that record PM10 and PM2.5 at 2-minute intervals for 24 hours. Each device has advantages that we will describe in the paper. Together, their readings provide a detailed record of 4 IAP exposure in poor households, in a stratified sample that captures variations in fuel use, cooking locations, structural materials, ventilation practices, and other factors. Our research has been designed to answer several questions about particulate exposure in poor households. First, is exposure largely confined to areas where combustion occurs? If so, particulate pollution will mostly affect the women who cook and the children whom they supervise. Second, how different are indoor and outdoor air pollution in high-poverty areas where most households burn biomass fuels? If atmospheric persistence of biomass emissions is high, then the difference between indoor and outdoor pollution may be small because air exchange between indoor and outdoor spaces is relatively rapid. Third, what are the actual differences in PM concentrations in houses that use different fuels? Are these concentrations significantly affected by typical variations in cooking practices, cooking locations, structural characteristics and ventilation practice (opening doors and windows)? If such effects are large, then simple alterations in household arrangements may provide a cost-effective alternative to fuel- switching or investment in clean stoves. Fourth, how much do concentrations vary by geographic region and income group? Finally, what are the prospects for increased use of improved stoves and clean fuels in Bangladesh? The remainder of the paper is organized as follows. Section 2 introduces the indoor air quality problem in Bangladesh, discusses our stratified sampling strategy, and describes the data that have been collected for this exercise. Section 3 provides comparative results for PM10 and PM2.5, while Sections 4-7 address the questions that have been posed in this introduction. Section 8 compares our findings to recent results 5 for India. Sections 9 and 10 discuss the prospects for adoption of improved stoves and cleaner fuels, and Section 11 provides a summary and conclusions. 2. Pollution Factors, Sampling Strategy and Data Description Previous studies have identified several potential determinants of exposure to indoor air pollution: fuel type, time spent in cooking, structural characteristics of houses, and household ventilation practices (opening of windows and doors, etc.) (World Bank, 2002, Brauer and Saxena, 2002, Moschandreas et al, 2002, Freeman and Sanez de Tajeda, 2002). All of these factors may be important in Bangladeshi households, which exhibit significant diversity in cooking fuels, stove types, cooking locations, and quality of ventilation. In Bangladesh, middle- and upper-income households in urban areas typically use electricity or relatively clean cooking fuels such as natural gas. However, households in peri-urban and rural areas rely primarily on biomass fuels. These include wood, twigs and leaves, animal dung, and agriculture residues such as straw, rice husks, bagasse, and jute sticks. Seasonal and economic factors may dictate the use of different biomass fuels over the annual cycle. Given the level of emissions from fuel use, the particulate concentration in a space depends on the length of time the emitted particles remain, as well as the ambient (outdoor) concentration. The extent and duration of smoke in the kitchen, and the amount of smoke leaking from the kitchen into the outdoors or other living spaces, may depend on several structural factors: the location of the kitchen, the extent of ventilation, and the porous nature of materials used to construct the roof and walls of the kitchen. 6 Bangladeshi rural and peri-urban households have a number of cooking arrangements. In many cases, kitchens are not enclosed by four walls and a ceiling. Some poor homes do not have separate kitchens; cooking takes place inside the single dwelling room during the rainy season and outside during the dry season. In others, kitchens have three walls (i.e., the entrance is entirely open), with or without a roof. Others have four walls and a gap of a few inches between the walls and the roof. Figure 1 provides descriptions of six typical kitchen arrangements that may have a significant effect on the duration of particles from combustion. Particle duration may also depend on other characteristics of a house that affect ventilation, such as the number of rooms, the number, size and placement of doors and windows, and materials used in the construction of walls and roofs. In Bangladesh, houses incorporate many combinations of these characteristics. We have used stratified sampling in urban and peri-urban areas of Dhaka to incorporate representative variations in fuel use, cooking arrangements and structural characteristics that affect ventilation.1 We separated the households into groups defined by cooking fuel, kitchen type and location, and construction material. Then we selected samples independently from each group. Tables 1a and 1b present the characteristics of the samples for our two air monitoring devices. The 24-hour real-time monitors are much more costly, so their deployment was more limited and the sample consequently smaller. In almost all cases, we generated comparable results by deploying our air samplers alongside the real-time monitors. 1Although we use the term "peri-urban" to describe areas proximate to Dhaka, our sample includes many rural farm-households. 7 Figure 1: Cooking Locations in Bangladeshi Households (Stove denoted by ) 1 2 3 4A 4B 5 6 8 In each household, we monitored PM10 concentrations in the kitchen and living room during the period December, 2003 ­ February, 2004. For a subsample of households, we also monitored PM2.5 concentrations. We monitored most houses for one day, and a few for two days. Our real-time monitoring instrument is the Thermo Electric Personal DataRAM (pDR-1000) (Thermo Electron, 2004). The pDR-1000 uses a light scattering photometer (nephelometer) to measure airborne particle concentrations.2 At each of 67 locations, the instrument operated continuously, without intervention, for a 24- hour period to record PM10 concentrations at 2-minute intervals. Our other instrument is the Airmetrics MiniVol Portable Air Sampler (Airmetrics, 2004), a more conventional device that samples ambient air for 24 hours. While the MiniVol is not a reference method sampler, it gives results that closely approximate data from U.S. Federal Reference Method samplers. Our MiniVols were programmed to draw Table 1a: Sample Composition (Kitchens): Thermo Electric Personal DataRAM Fuel Type Kitchen Type Construction Material Gas, Electricity, Single room dwelling, Kerosene 6 no separate kitchen 9 Wall: Thatch 24 Kitchen with a Firewood, Twigs, 31 partition (4 walls and 11 Wall: Tin 22 Leaves a roof) Separate, attached Cow Dung 15 kitchen (4 walls and a 14 Wall: Mud 14 roof) Rice Husks, Straw, Separate, detached Wall: Brick/ Mud, Jute Sticks, 15 kitchen (4 walls and a 31 Roof: Other than 4 Bagasse, Sawdust roof) Concrete Outside/ open kitchen Wall: Brick/ Mud, (0 walls, no roof) 2 Roof: Concrete 1 Total Number of 67 Total Number of 67 Total Number of 65 Households Households Households 2 The operative principle is real-time measurement of light scattered by aerosols, integrated over as wide a range of angles as possible. 9 Table 1b: Sample Composition (Kitchens): Airmetrics MiniVol Air Sampler Fuel Type Kitchen Type Construction Material Gas, Electricity, 35 Single room dwelling, 36 Wall: Thatch 100 Kerosene no separate kitchen Firewood, Twigs, 89 Kitchen with a 37 Wall: Tin 37 Leaves partition (4 walls and a roof) Cow Dung 42 Separate, attached 43 Wall: Mud 39 kitchen (4 walls and a roof) Rice Husks, Straw 16 Separate, detached 89 Wall: Brick/ Mud, 9 kitchen (4 walls and a Roof: Other than roof) Concrete Jute Sticks, 54 Outside/ open kitchen 31 Wall: Brick/ Mud, 20 Bagasse, Sawdust (0 walls, no roof) Roof: Concrete Total number of 236 Total number of 236 Total number of 205 households households households air at 5 liters/minute through PM10 and PM2.5 particle size separators (impactors) and then through filters. The particles were caught on the filters, and the filters were weighed pre- and post exposure with a microbalance. We operated the air samplers at 236 locations. 3. Sample Evidence: PM10 vs. PM2.5 We focus on PM10 in this paper because our monitoring sample is much larger. However, we have also monitored PM2.5 at over 80 sampling points, to assess the stability of the relationship between the two measures. Overall, we find an extremely stable relationship: The mean ratio (PM2.5/PM10) is .51, with a standard error of .02. This implies a 95% confidence interval of .47 - .55. As Table 2 shows, the ratio does not vary significantly across biomass fuels for which we have information. Our results suggest 10 that the PM10 results in this paper can reasonably be recast as PM2.5 results, simply by dividing them by 2. Table 2: PM2.5/PM10 Ratios for Biomass Fuels Fuel Ratio Houses Dung 0.52 22 Firewood 0.51 18 Straw 0.56 4 Branches, Twigs 0.56 16 Total 0.51 85 4. Is Exposure Largely Confined to Cooking Areas? Analyses of indoor air pollution in poor households often stress the health risks for women who cook and children under their supervision in cooking areas (Rosemarin, 2002; Smith, 2000). Higher health risks are attributed to two factors: higher pollutant concentrations in cooking spaces, and longer times spent indoors. This section examines the first factor, using our 24-hour real time (PDRam) monitoring results for PM10 in kitchens and living rooms. Section 6 reports equivalent results from the MiniVol samplers for a large sample of households. In a forthcoming paper, we will compare times spent indoors by different age / sex groups. Figure 2a displays the time paths of PM10 concentrations in four representative households. We express concentrations in log form to prevent visual scaling problems in the graphs. Although the 24-hour patterns in the four graphs are quite different, all four exhibit very close tracking of PM10 concentrations in the kitchen and living room. The 2- minute observations for the kitchens exhibit many short-interval "spikes", and the living room observations closely resemble smoothed versions of the kitchen series. 11 Remarkably, Figure 2b shows that the close relationship can hold even when cooking is done outdoors. Table 3 provides log-log regression results for various 2- minute lag structures. The implied adjustment lag is very short, and the overall fit is obviously very strong. Summary evidence for all households indicates that the observations in Figure 2 are very common in our sample. The correlation coefficient for the PDRam households is .93 for median PM10 concentrations in kitchens and living rooms; Figure 3 illustrates the strength of the relationship. To summarize, the PDRam evidence strongly indicates that air pollution from cooking diffuses into living spaces very rapidly, and at similar intensity. Figure 2a: Comparative PM10 Concentrations in Four Bangladeshi Houses: Kitchens and Living Rooms (1) (2) niotart ontia en 9 nc 7 Co oncentrC mo 8 Ro gnivLi 6 omoR gniviL 7 goL/nio 5 atrt goL/notia 6 encn 4 entrcn 5 Co nehctiK Co 3 goL 0 5 10 15 20 25 enhctiK 4 24-Hour Time 0 5 10 15 20 25 Log 24-Hour Time Log Kitchen Concentration Log Living Room Concentration Log Kitchen Concentration Log Living Room Concentration (3) (4) noitatrne 7 noiatrt 8 ncoC ncen Co mooR 5 6. mo 7 Ro ngiviL 6 goL n/io 55. attr gniviLgoLn/oitart 6 en nc 5 encn 5 Co Co n n hetciK 5 4. hectiK 4 g 0 5 10 15 20 25 g 0 5 10 15 20 25 Lo 24-Hour Time Lo 24-Hour Time Log Kitchen Concentration Log Living Room Concentration Log Kitchen Concentration Log Living Room Concentration 12 Figure 2b: PM10 Concentrations: Outdoor Kitchens vs. Living Rooms (5) (6) on atitrne noitart 8 8 en ncoC ncoC 7 oomR omoR 7 gniviL gniv 6 Li g 6 n/Logoitatr 5 /Lonoita entr 5 encnoC 4 ncoC n enh tciK 3 hectiK 4 g g 0 5 10 15 20 25 0 5 10 15 20 25 Lo Lo 24-Hour Time 24-Hour Time Log Kitchen Concentration Log Living Room Concentration Log Kitchen Concentration Log Living Room Concentration (7) noi at entr 6.5 ncoC omoR 6 ngiv Li 5.5 Log n/oi atr 5 ent ncoC hentciK 4.5 0 5 10 15 20 25 Log 24-Hour Time Log Kitchen Concentration Log Living Room Concentration 13 Figure 3: Household PM10 Concentrations: Kitchens vs. Living Rooms Median PM-10 (24-Hour ug/m**3) 400 mooR 300 200 ng vi Li 100 0 0 100 200 300 400 Kitchen Table 3: Living Room vs. Kitchen PM10 Concentrations: Lag Relationship Dependent Variable: Log Living Room PM10 (1) (2) (3) (4) (5) (6) (7) Log 0.770 0.383 0.366 0.345 0.336 0.328 0.326 Kitchen (253.14)** (37.79)** (36.59)** (34.44)** (33.68)** (32.87)** (32.67)** PM10 L1 0.405 0.121 0.129 0.120 0.121 0.116 (39.93)** (8.97)** (9.64)** (8.99)** (9.07)** (8.72)** L2 0.314 0.123 0.129 0.122 0.122 (31.34)** (9.16)** (9.63)** (9.10)** (9.16)** L3 0.212 0.070 0.075 0.069 (21.22)** (5.23)** (5.61)** (5.17)** L4 0.159 0.051 0.056 (15.93)** (3.84)** (4.16)** L5 0.121 0.030 (12.11)** (2.22)* L6 0.103 (10.28)** Const 1.094 1.001 0.933 0.892 0.862 0.842 0.825 (67.98)** (62.97)** (58.94)** (56.23)** (54.19)** (52.65)** (51.33)** Obs 33,096 33,061 33,015 32,969 32,923 32,877 32,831 R2 0.66 0.68 0.68 0.69 0.69 0.69 0.69 Adj. R2 0.66 0.68 0.68 0.69 0.69 0.69 0.69 Absolute value of t statistics in parentheses * significant at 5%; ** significant at 1% __________________________________________________ 14 5. Outdoor vs. Indoor Air Pollution Policy researchers often view indoor and outdoor air pollution as separate problems in developing countries. Although a rationale is seldom provided, this distinction reflects two beliefs: that indoor pollution is a more serious problem for poor households, particularly in rural areas, and that outdoor pollution is distinct because it comes from multiple sources that may be distant from the household. Our 24-hour monitoring data for a Dhaka household with an extremely clean fuel ­ piped natural gas ­ provide us with some insight in this context, because we have also monitored the ambient concentration at five locations in urban Dhaka. Figure 4 displays the 24-hour concentrations separately and overlaid. The kitchen readings for the natural-gas household closely resemble the ambient readings. The ambient readings, in turn, appear related to the daily cooking cycle for the great majority of households that use biomass fuels.3 Because outdoor air rapidly replaces indoor air, the indoor concentration in a clean-fuel household closely reflects the ambient concentration. In urban Dhaka, our MiniVol readings at five outdoor locations yield a mean 24-hour concentration of 89 ug/m3.4 The mean daily concentration for the natural gas household monitored with a PDRam unit is 101 ug/m3. For a broader group of 10 urban households monitored by our MiniVol sampler, the mean daily outdoor 24-hour concentration is also 101 ug/m3. By implication, burning piped natural gas adds negligible PM10 pollution to cooking and living spaces. 3 The 24-hour cycle of ambient PM10 concentrations is very similar to the pattern of average hourly residuals from a panel regression that controls for differences in average PM10 concentrations for households that use biomass fuels. 4 For comparison, we cite PM10 concentration measured by the Bangladesh Air Quality Management Program monitor situation at the Parliament building in Dhaka. From March, 2002 to February, 2003, the mean daily concentration was 137 ug/m3. Our thanks to our colleague Paul Martin for this contribution. 15 Our comparative ambient monitoring in peri-urban Dhaka has been undertaken in areas with rural characteristics, well-removed from major transport arteries and industrial sites. MiniVol monitoring at three locations yields a mean 24-hour concentration of 48 ug/m3, which compares favorably with the Indian safe standard of 100 for rural areas (World Bank, 2002). Nevertheless, our results suggest that ambient pollution from biomass burning is substantial in rural villages. Figure 4: PM10 Concentrations: Outdoor vs. Indoor for Piped Natural Gas Ambient Overlay 400 noi 400 atrt en 0 no 30 ncoCt 0 30 atitr en bi en mA 0 ncoCt 0 20 enib mA 0 n/oitartnecnoC 20 0 10 10 n hectiK 0 0 0 5 10 15 20 25 24-Hour Time 0 5 10 15 20 25 24-Hour Time Kitchen Concentration Ambient Concentration Piped Nat. Gas 0 25 0 noiatrtnecnoCn 20 0 15 hetciK 0 10 50 0 5 10 15 20 25 24-Hour Time 16 6. Sources of Variation in Household PM10 Concentration In this section, we provide comparative evidence from both monitoring devices. The two instruments use completely different monitoring techniques (weighted filters for the MiniVol, laser optics for the PDRam), so we use regression analysis to compare their readings for a common sample of kitchen and living spaces. For each PDRam reading, we use the 24-hour mean of 2-minute observations as the closest approximation to the corresponding MiniVol reading. Table 4: Comparative PM10 Concentrations: MiniVols vs. PDRams Dependent Variable: MiniVol Coefficient t-Statistic PM10 Concentration PDRam 24-Hour Mean PM10 Concentration 0.84 9.27 Constant 73.88 2.95 Observations 85 Adjusted R2 0.50 Our results (Table 4) suggest a reasonably close correspondence between readings from the two devices. Across the common sample of 85 kitchens and living rooms, mean PM10 concentrations are 261 for the PDRams and 272 for the MiniVols. The regression of MiniVol readings on PDRam readings yields highly-significant parameter estimates and an adjusted R2 of .50. The regression constant is 73.88 and the marginal coefficient is 0.84 (i.e., the predicted MiniVol reading increases .84 ug/m3 for each increase of 1 ug/m3 in the PDRam reading). Although the overall means for the two devices are almost identical, the regression result suggests that MiniVol readings tend to be higher than PDRam readings in houses with below-average readings, and lower in houses with above-average readings. Since the MiniVol results reflect the current scientific 17 convention for measuring indoor air pollution, we rely on them for our cross-sectional analysis. However, the PDRam readings in Figure 2 provide a very useful perspective on the timing of pollution. They show that the largest source of variation in PM10 concentrations is the 24-hour cycle in individual households. Peaks occur during morning and evening cooking periods, and houses with three cooking periods have an additional peak. Typically, concentration levels plunge in the afternoon in houses where kitchens are aired out, and they fall in the evening as well (although often not as far, since windows and doors may be closed at night). Most houses have relatively low concentrations for significant parts of the day. Inter-household differences in pollution exposure are largely attributable to two factors: the level of peak concentrations during cooking, and the rate at which concentrations decline after cooking. As Figure 2 suggests, these factors differ substantially from house to house. Their significance emerges strikingly when we control for fuel use. Table 5 presents MiniVol PM10 statistics for sample households in the three most common fuel-use categories: firewood, dung and jute. None of these households reports using an improved stove or a chimney. Across households, PM10 concentrations vary from 84 to 1165 ug/m3 for firewood, 60 to 755 for dung, and 72 to 727 for jute. As Figure 2 indicates, many houses are relatively "clean" during parts of the night and afternoon, when indoor readings resemble ambient readings. However, differences in cooking practices, structural arrangements and ventilation behavior generate very large differences in overall concentrations. As a result, some households 18 Table 5: Household PM10 Concentrations (ug/m3) by Fuel Sample 1st 3rd Fuel Households Mean Minimum Quartile Median Quartile Maximum Dung 95 291 60 172 231 380 755 Firewood 159 263 84 161 201 323 1165 Jute 68 190 72 136 165 219 727 using "dirty fuels" such as firewood, dung and jute have PM10 exposures resembling those for natural gas, while others face concentrations at extremely high levels. Of course, part of the variation across households is also determined by fuel use. Table 6 summarizes the MiniVol data for cooking areas across all households, by fuel type. While these descriptive statistics provide some insight, they can be misleading because fuel choices may be correlated with other variables in the sample: cooking practices, structural characteristics and ventilation practices. In Table 6, the cleanest fuels are natural gas (101 ug/m3) and kerosene (134). As we have previously noted, these concentrations are not far above urban ambient pollution levels. Among biomass fuels, dung seems to be the dirtiest fuel (291 ug/m3), followed by firewood (263), sawdust (237), straw (197), jute (190), and twigs and branches (173). In our sample, the relatively few households using LPG/LNG have an average PM10 count (206) above the level for several biomass fuels. We attribute the elevated PM10 concentrations in these households to several factors: Ambient pollution from biomass cooking in the same locality, ambient pollution from other sources (e.g., motor vehicles, industrial sources), underreporting of complementary biomass fuel use (particularly for LPG/LNG), and particulate pollution from the fuels themselves. Recent monitoring research has suggested that kerosene is a significant PM10 source when burned indoors (Leaderer, et al., 1999), and even natural gas produces some PM10 (Beer, 2000). 19 Table 6: Mean PM10 Concentration by Fuel Used (MiniVols) Mean Fuel (ug/m3) Households Dung 291 95 Firewood 263 159 Sawdust 237 7 LPG/LNG 206 8 Straw 197 29 Jute 190 68 Twigs, Branches 173 46 Kerosene 134 18 Piped Nat Gas 101 20 In order to assess the role of other factors (which may be correlated with fuel use), we have used regression analysis to explore the relationships between PM10 concentrations and a large set of variables that describe household cooking and ventilation practices, structure characteristics and building materials. The variables tested are described in the Appendix. Besides fuels employed during the monitored day, these include cooking time, duration of fire after cooking, numbers of people cooked for, stove location (see Figure 1), the use of iron, mud, thatch and concrete for construction, the placement and size of windows, doors and ventilation spaces between walls and roofs, ventilation practices such as opening doors and windows after cooking, smoking practices, and the use of lanterns and mosquito coils. Among these variables, we find a small set that significantly affect household PM10 concentrations through their impact on ventilation: Stove locations, building materials, and opening doors and windows after cooking. Regression results are summarized in Table 7. The first column provides joint estimates for kitchens and living areas; the others provide estimates for the two spaces separately. Columns 3 and 4 drop the control for the relatively weak effect of opening doors and windows after midday cooking. 20 Table 7: Regression Results: Determinants of PM10 Concentrations In Kitchens and Living Areasa Kitchen & Living Kitchen Kitchen Living Living Room Dummy -40.057 (3.52)** Mud Walls 252.921 261.472 253.896 229.729 (9.84)** (6.67)** (6.53)** (6.39)** Mud Walls, -158.160 -121.130 -124.058 -163.725 Detached Kitchen (3.99)** (1.72) (1.76) (3.83)** Thatch Roof -100.357 -70.898 (Living Room) (5.17)** (6.01)** Kitchen Windows, Doors -32.016 -39.906 Open After Midday Meal (2.25)* (1.79) Detached Kitchen -46.711 -40.672 -37.599 -57.381 (4.25)** (2.48)* (2.44)* (4.86)** Open-Air Kitchen -64.134 -88.337 -79.887 -80.504 (4.31)** (4.11)** (3.77)** (5.90)** Jute -41.136 -40.645 -45.233 -41.225 (3.45)** (1.97) (2.20)* (3.27)** Kerosene -89.758 -103.172 -106.729 -76.197 (8.15)** (6.59)** (7.46)** (6.68)** Lpg/Lng -102.597 -113.334 -112.523 -89.441 (4.16)** (3.16)** (3.40)** (3.04)** Piped Natural Gas -136.411 -144.226 -155.285 -135.870 (12.09)** (9.27)** (10.22)** (10.07)** Constant 289.830 287.410 258.563 235.342 (16.43)** (11.47)** (17.47)** (19.94)** Observations 424 207 234 246 R-squared 0.46 0.41 0.40 0.54 aHuber-White robust t statistics in parentheses Robust t statistics in parentheses significant at 5%; ** significant at 1% 21 The results provide several insights into the sources of variation in indoor air pollution in Bangladeshi households. First, PM10 concentrations in living areas are lower to a relatively small but significant degree. The living-area adjustment in regression 1 is -40.1, or about 16% of mean PM10 for kitchens. Comparison of the kitchen and living- area regressions suggests similar responses to fuel and ventilation factors. In the following discussion, we focus on the combined-area results (regression 1). Controlling for factors correlated with fuel use provides a significantly different view of fuel-based pollution factors. Our combined-area results suggest that among biomass fuels, jute is a negative outlier, and all others (dung, firewood, twigs and branches, rice husks, straw) cannot be distinguished from one another. With other biomass fuels as the baseline, use of jute subtracts about 41 ug/m3 from the indoor PM10 concentration. Relative to the biomass baseline, kerosene subtracts about 90 ug/m3, lpg/lng 103, and piped natural gas 136. These results contrast significantly with the unadjusted means in Table 6, which are misleading because they do not account for correlation with ventilation factors. There are particularly divergent results for dung and lpg/lng, whose estimated relative contributions to PM10 are higher in Table 6 than in Table 7. Our results highlight the importance of ventilation factors in the determination of PM10 concentrations. We find that two construction factors ­ mud walls and thatch roofs ­ have highly significant effects on ventilation. Mud walls are particularly important in this context. In most localities in Bangladesh, the soil has low sand content and mud walls and floors are frequently re-coated with fresh mud to prevent cracking. This 22 creates an effective seal that permits almost no ventilation by comparison with thatch and corrugated iron, the other two common building materials. The effect of mud-wall construction depends on the location of cooking. If it is inside the house, the sealing effect of mud walls increases the PM10 concentration by 253 ug/m3 in the baseline case. If cooking occurs in a detached or open-air location, mud walls in the kitchen have the same sealing effect. However mud walls in the living room have an insulating effect when the kitchen is outside, and the overall PM10 concentration is reduced by 158 ug/m3. For other construction materials, the PM10 concentration is reduced 47 ug/m3 by having a detached kitchen (stove location 5 in Figure 1), and 64 ug/m3 by having an open-air kitchen. It is further reduced 32 ug/m3 by opening kitchen doors and windows after the midday meal. We also find a significant ventilation role for thatched roofs in living spaces, which lower the PM10 concentration by 100 ug/m3. In Table 8, we tabulate the interactive effect of critical pollution factors by computing mean PM10 concentrations for groups that distinguish between "clean" (kerosene, natural gas) and biomass fuels, inside and outside (detached or open-air) cooking, and mud-wall and other construction. Table 8(a) tabulates results for all fuels in the sample households, while 8(b) provides the same information for houses using firewood only. We present the information in 8(b) to show how other critical factors affect variations in indoor pollution associated with a single biomass fuel. 23 Table 8: Pollution Factors and PM10 Concentrations (a) All Fuels Difference in Cooking Building Mean Median Mean t- Fuel Location Material Space Abbrev. PM10 PM10 Houses (From BIOL) statistic Biomass Inside Mud Kitchen BIMK 515 528 23 292 11.15 Biomass Inside Mud Living BIML 467 453 20 244 8.86 Biomass Outside Mud Kitchen BOMK 351 258 22 128 4.81 Biomass Inside Other Kitchen BIOK 250 220 74 27 1.46 Biomass Outside Mud Living BOML 244 218 29 21 0.87 Biomass Inside Other Living BIOL 223 213 62 Biomass Outside Other Kitchen BOOK 203 191 101 -20 -1.16 Biomass Outside Other Living BOOL 166 162 116 -57 -3.41 Clean CLN 133 117 46 -90 -4.34 Overall 231 187 493 (b) Firewood Only Difference in Cooking Building Mean Median Mean t- Location Material Space Abbrev. PM10 PM10 Houses (From BIOL) statistic Inside Mud Kitchen BIMK 498 520 10 248 5.24 Inside Mud Living BIML 475 443 9 226 4.59 Outside Mud Kitchen BOMK 638 567 5 389 6.33 Inside Other Kitchen BIOK 267 220 31 17 0.49 Outside Mud Living BOML 293 301 4 44 0.65 Inside Other Living BIOL 250 210 21 Outside Other Kitchen BOOK 210 193 35 -39 -1.15 Outside Other Living BOOL 165 163 44 -84 -2.58 Overall 263 201 159 The tables reveal a nearly-identical pattern of results for all fuels and firewood only, so we focus on the results for all fuels. For statistical comparison of means, our benchmark is the mean living-space PM10 concentration for households with biomass fuels, inside cooking and non-mudwall construction (acronym BIOL in the tables). The mean concentrations for these households are 223 ug/m3 for all fuels, and 250 ug/m3 for firewood. By comparison, cooking and living spaces have far greater pollution for inside 24 cooking and mud-wall construction (BIMK, BIML). In these cases, mean PM10 concentrations for kitchens and living areas are 515 and 467 ug/m3, and the differences from the benchmark (BIOL) mean are highly significant. Outside kitchens with mud- wall construction also have much higher pollution (351 ug/m3 ­ a highly-significant difference from BIOL). However, the mean concentration for mud-wall living rooms with outside cooking (BOML) is no higher than the BIOL benchmark. This reflects the symmetric effect of mud walls, which act as a sealant against airflow in or out of the living area. For other building materials, outside cooking generates a kitchen concentration (BOOK) that is not significantly different from the benchmark concentration. However, the living room concentration in the same case (BOOK) is significantly lower than the benchmark case. As expected, clean-fuel households have the lowest mean concentration (133), which is significantly (90 ug/m3) lower than the benchmark case. Our clean fuel results are quite similar to those obtained by earlier studies in Chile (Pino, et al., 1998; Caceres, et al., 2001) and India (Balakrishnan, et al., 2002; Parikh, et al., 2001). Replication of this pattern in the firewood-only case highlights an important implication of the results: Two ventilation factors ­ kitchen location and mud-wall construction ­ account for very large differences in PM10 concentrations across households. This is as true for a single biomass fuel (firewood) as it is for all fuels combined. 7. PM10 Concentrations by Geographic Area and Income Group We have analyzed the determinants of indoor air pollution using a stratified sample of urban and peri-urban households in the Dhaka region. Our stratification has been 25 designed for cell values large enough to test fuel and ventilation effects, and is not intended to represent all Bangladeshi households. However, extrapolation of our results requires more representative household samples. Accordingly, we have surveyed households in six areas of Bangladesh whose major cities are identified in Figure 5: Rangpur (Northwest), Rajshahi (West Central), Jessore (Southwest), Sylhet (Northeast), Faridpur (Central) and Cox's Bazar (Southeast). In each region, we have attempted to randomly survey 50 rural households, 25 peri-urban households, and 25 urban households. Implementation difficulties intruded in Rangpur, but otherwise, the data in Table 9 reflect our original intent. Table 9: Household Sample Sizes by Area Peri- Area Rural Urban Urban Total Cox's Bazar 51 24 25 100 Faridpur 50 25 25 100 Jessore 50 25 25 100 Rajshahi 54 21 24 99 Rangpur 25 25 50 100 Sylhet 49 25 25 99 Total 279 145 174 598 We have sampled by locality to assess the impact of local differences in ventilation characteristics and fuel use. To indicate the potential significance of these differences, Table 10 presents the rural incidence of four important determinants of indoor air pollution: mud walls, thatch roofs, detached kitchens and open-air kitchens. 26 27 Table 10: Determinants of Indoor Air Pollution Six Rural Areas of Bangladesh Sample Incidence of Pollution Factors Mud Thatch Detached Open-Air Rural Area Walls Roof Kitchen Kitchen of: % % % % Cox's Bazar 71 51 14 0 Faridpur 0 8 60 36 Jessore 64 16 66 12 Rajshahi 89 0 70 7 Rangpur 0 20 72 20 Sylhet 20 18 12 35 Our survey results indicate the potential importance of regional customs and differential availability of building materials and fuels. Mud walls, the most important ventilation characteristic, are extremely common in Cox's Bazar, Jessore and Rajshahi, but extremely scarce in Faridpur and Rangpur. Significant variation is also apparent for thatch roofs. Detached or open-air kitchens are present in 14% of households in Cox's Bazar, 47% in Syhlet, and 96% in Faridpur. To capture the simultaneous effect of ventilation factors and fuel use in each surveyed household, we use the PM10 model in Table 7 to estimate PM10 concentrations in kitchens and living areas. We drop ventilation after midday meals because this was not recorded in the country-wide survey, and use of jute fuel because its incidence in our sample is negligible (3 users in 599 households). We tabulate the results by geographic area in Table 11, which highlights the effect of local variations. For kitchens in rural areas, mean estimated PM10 concentrations range from 410 in Cox's Bazar to 202 in Faridpur. Similar variation characterizes living spaces, although the range is somewhat more limited. For Cox's Bazar and Sylhet, there is a clear pattern of declining PM10 concentration in the transition from rural to urban areas. As Table 12 shows, a significant part of this pattern is attributable to increased use of clean fuels. 28 Table 11: Mean PM10 Concentrations (ug/m3) Kitchens Peri- Region Rural Urban Urban Total Cox's Bazar 410 249 181 314 Faridpur 202 203 185 198 Jessore 295 207 199 249 Rajshahi 248 252 204 238 Rangpur 208 205 178 192 Sylhet 246 214 108 203 Total 274 221 176 233 Living Areas Peri- Region Rural Urban Urban Total Cox's Bazar 333 215 164 262 Faridpur 162 165 172 165 Jessore 219 175 168 196 Rajshahi 276 197 181 236 Rangpur 159 163 151 156 Sylhet 212 185 100 177 Total 234 183 155 199 Table 12: % Clean Fuel Use, by Geographic Area Peri- Region Rural Urban Urban Total Cox's Bazar 0 8 64 18 Faridpur 0 12 44 14 Jessore 0 20 28 12 Rajshahi 0 0 42 10 Rangpur 4 8 28 17 Sylhet 0 44 92 34 Total 1 16 47 18 29 We have also tabulated incomes by source for the surveyed households. Table 13 shows that our sample reflects the generally-high level of poverty in Bangladesh. Overall, 78% of sample households have per capita incomes less than $1.00/day. Across regions, the percentage of households below this extreme-poverty threshold varies from 67% in Sylhet to 94% in Rajshahi. Table 13: Household Distribution by Income Per Capita (%) $.51- $1.01- $2.01- Region 0-$.50 $1.00 $2.00 $5.00 $5.01+ Cox's Bazar 41 36 18 3 2 Faridpur 46 28 20 5 1 Jessore 40 36 19 4 1 Rajshahi 62 32 5 1 0 Rangpur 47 33 17 3 0 Sylhet 41 26 20 9 3 Total 46 32 17 4 1 Table 14 displays the distribution of estimated mean PM10 concentrations by income group for our sample households. In general, higher-income groups have lower concentrations. For the whole sample, the average concentration is 253 ug/m3 for the poorest households (0-$.50/day) and declines steadily to 141 for households with $5.00/day or more. An important part of this trend can be attributed to a greater incidence of clean fuel use by higher-income households. However, the importance of regional building and fuel-use patterns again emerges here. For example, average PM10 for the highest income group in Cox's Bazar (195 ug/m3) is almost identical to the concentration for the lowest-income group in Rangpur (198). 30 Table 14: Kitchen PM10 Concentration by Income Group $.51- $1.01- $2.01- Region 0-$.50 $1.00 $2.00 $5.00 $5.01+ Cox's Bazar 355 330 212 266 195 Faridpur 204 205 181 171 144 Jessore 291 230 218 169 144 Rajshahi 245 237 185 144 . Rangpur 198 201 170 132 . Sylhet 244 191 172 156 103 Total 253 236 190 171 141 Table 15: Kitchen PM10 Concentrations (ug/m3): Households With Less Than $1.00/Day Per Capita Peri- Region Rural Urban Urban Total Cox's Bazar 410 267 196 343 Faridpur 202 209 208 205 Jessore 297 209 215 262 Rajshahi 248 252 215 242 Rangpur 208 206 188 199 Sylhet 248 223 122 223 Total 275 226 193 246 Table 15 focuses on the geographic pattern of indoor air quality for households below the extreme poverty threshold (less than $1.00/day). Even among the poorest households, we see large variations that are attributable to within-region differences in construction practices and fuel use. In Cox's Bazar, the rural poor face far higher concentrations than the urban poor, but there is no difference in Faridpur. Overall, households living in extreme poverty face mean concentrations of 275 ug/m3 in rural areas, 226 in peri-urban areas, and 193 in urban areas. 31 8. Comparison with Monitoring Results for India A recent monitoring study for Indian households (World Bank, 2002; Balakrishnan, et al., 2002; Parikh, et al., 2001) has provided useful comparative information about indoor pollution levels and their determinants. Table 16 displays average pollutant concentrations by room, stove location and fuel type for the two countries. The Indian data are reported for respirable suspended particulate matter (RSPM), defined as the fraction of inhaled aerosols capable of penetrating the alveolar (gas-exchange) regions of the adult lung. The authors report that ratios of RSPM to PM10 varied from 0.57 to 0.73 in their samples, with a mean of .61. Comparison of Tables 16a (India) and 16b (Bangladesh) indicates both similarities and differences. The India results for gas fuels are almost identical to the Bangladesh results when the latter are multiplied by .61 (the RSPM/PM10 ratio). The Indian ambient reading is also within the same range as the Bangladesh urban reading. However, the results for solid (biomass) fuels are quite different. For kitchen areas, the India concentrations are all much higher than their RSPM-adjusted counterparts for Bangladesh. The India concentrations for living areas are also much higher than the Bangladesh concentrations, although the differences are not as extreme. Both studies have used regression analysis to explore the determinants of indoor air pollution, and both have come to generally-similar conclusions: Fuel choice, cooking location and other ventilation factors all play significant roles. However, the India study finds that fuel choice is the dominant factor, while our results for Bangladesh suggest that ventilation factors are collectively more significant. In the India study, pollution intensity is highest for dung, followed by woodfuels, kerosene and gas. Our estimated means in 32 Table 6 follow the same pattern, but our multivariate regressions find no statistically- significant difference between dung and woodfuels. Our results suggest that kerosene is somewhat more pollution-intensive than gas, but the difference is much less than the difference in the India study. Table 16: Comparative Air Pollutant Concentrations (ug/m3) 16a: India (RSPM .61 PM10) Inside Kitchen Inside Open With Without Detached Air Area Fuel Partition Partition Kitchen Kitchen Kitchen Solid 666 652 575 297 Gas 70 70 86 Living Solid 357 559 280 215 Gas 70 76 96 Ambient 91 Source: World Bank (2002), p. 32 16b: Bangladesh (PM10) Open Inside Detached Air Area Fuel Kitchen Kitchen Kitchen Kitchen Solid 313 248 182 Gas 134 Living Solid 286 189 155 Gas 129 Ambient Urban Peri-Urban 89 48 9. Use of Improved Stoves New biofuel stove designs offer the prospect of reduced indoor air pollution, along with more efficient combustion. Air-monitoring research in Guatemala has indicated that improved stoves can lower indoor PM10 concentrations by 50% or more (Smith, et al., 33 1993; McCracken, et al., 1999; Naeher, et al., 2001). In Bangladesh and elsewhere, programs to promote improved stoves have stressed their long-run financial advantages for poor households, as well as their environmental benefits. To date, however, our survey suggests that progress has been quite limited. Of 686 biofuel-using households in our 7-region survey (including Dhaka), only 9 (1.3%) report using an improved stove: 4 in Jessore and 5 in Sylhet. Another 2 households have tried improved stoves, but have stopped using them. Of the 9 current users, 8 are in rural areas. Our results suggest that limited information may be the greatest deterrent to consideration of improved stoves. Of the 659 biofuel-using households that don't use improved stoves and offer an explanation, 45% claim to be unaware of them. Another 40% state that improved stoves are not available locally. After accounting for these, our sample includes only 105 biofuel-using households that have considered improved stoves. Of these, 9 (8.5%) decided to use one, 49% didn't adopt because of the large initial investment, 39% viewed improved stoves as inconvenient, and the remaining 4% had other reasons for non-adoption. Jessore and Sylhet may have witnessed greater promotional efforts than other areas. Of the 21 Sylhet households (in our sample of 99) that consider improved stoves a feasible option, 5 (24%) actually use one. In Jessore, 4 of 37 households (11%) with a viable option use improved stoves. In all other regions, we have found no adoption among biofuel-using households. Since 85% of all sample households regard improved stoves as unknown or unavailable, our results suggest that an information strategy may offer the best near-term prospect for promoting clean stove use in Bangladesh. 34 10. Use of Clean Energy Sources Among 598 households in our six-region random sample, 108 report using one of five clean energy sources for cooking: Piped natural gas (32), lpg/lng (67), kerosene (5), electricity (3) and biogas (1). Although we have small sample sizes for higher-income groups (Table 17b), our results are consistent with a strong preference for clean energy, even at low incomes (Table 17a). They also suggest wide geographic variation in response to local supply conditions and prices. In Table 17a, urban and peri-urban use of clean energy rises rapidly with income, reaching 80% among households reporting daily incomes of more than $2.00 per capita. Even among families living in extreme poverty ($1.00 per capita per day or less), clean energy use is used by 29% of the households in urban areas and 10% in peri-urban areas. A striking contrast is provided by rural areas, where clean-energy prices are increased by distance from urban distribution sources. In our sample, there is effectively no use of clean energy in rural areas, regardless of income. Table 17: Household Use of Clean Energy Sources (a) Percent of Households Using Clean Energy Locality Income Per Capita Rural Peri-Urban Urban Total ($US Per Day % % % % Less Than $1.00 0.4 9.9 28.9 8.6 $1.00 - $2.00 0.0 37.9 63.6 46.5 More Than $2.00 0.0 80.0 81.8 68.8 Total 0.4 17.9 46.6 18.1 (b) Total Households (Clean & Dirty Energy Sources) Income Per Capita Rural Peri-Urban Urban Total Less Than $1.00 259 111 97 467 $1.00 - $2.00 15 29 55 99 More Than $2.00 5 5 22 32 Total 279 145 174 598 35 11. Summary and Conclusions In this paper we have investigated the determinants of indoor air pollution in Bangladesh, using monitoring data for a stratified sample of 236 households in the region of Dhaka. Extrapolating from our results, we have estimated indoor air pollution levels for a random sample of 600 rural, peri-urban and urban households in six regions: Rangpur, Sylhet, Rajshahi, Faridpur, Jessore and Cox's Bazar. We have used the results of our analysis to address several basic questions about air quality for poor households in Bangladesh: (1) Does air pollution from cooking primarily affect the women who cook and children who are with them in the kitchen? An appropriate answer to this question must consider both potential exposure (from pollution levels in kitchens and living areas) and actual exposure (from age/sex differences in time spent indoors, in kitchens and living areas. In a future paper, we will analyze the sources and consequences of age/sex differences in exposure time. In this paper, we can only summarize the implications of our results for potential exposure. Overall, our results suggest that potential exposure is similar in kitchens and living areas. Pollution from cooking diffuses into living spaces rapidly and fairly completely in many cases, so that exposure is similar for all household members who are indoors during the same periods. Table 18 indicates that living-area PM10 is about 17% lower than kitchen PM10 on average, although regional differences in ventilation factors cause this percent difference to vary considerably. In absolute magnitude, all kitchen PM10 concentrations and most living-area levels are above 200 ug/m3 in all rural areas. As we have noted in the paper, the sealing effect of mud-wall construction can pose a major 36 exception to this general pattern. In a firewood-using, mud-wall household with detached kitchen, for example, cooking women face PM10 concentrations over 600 ug/m3, in contrast to concentrations near 300 in living areas. Neither concentration is healthy, but the latter is obviously more desirable. Second, differences in ventilation factors produce great variations in potential exposure for all household members, even for poor families that use biomass fuels for cooking. For example, Table 8 shows that ventilation factors can vary typical PM10 concentrations in firewood-using households from 210 to 638 ug/m3. Table 18: Percent Difference: Kitchen vs. Living Area PM10 Peri- Region Rural Urban Urban Total Cox's Bazar 23 16 11 20 Faridpur 25 23 8 20 Jessore 34 18 18 27 Rajshahi -10 28 13 1 Rangpur 31 26 18 23 Sylhet 16 16 8 15 Total 17 21 13 17 (2) How are indoor and outdoor pollution related? Our analysis suggests that ambient pollution contributes significantly to indoor pollution, with pronounced effects for households that use clean fuels. During our sample period, ambient pollution created a daily indoor pollution "baseline" near 100 ug/m3 in urban Dhaka, and a baseline near 50 ug/m3 in peri-urban areas. Since many of the latter are actually rural (no proximate motor roads or industries), our results suggest that diffusion of smoke from biomass cooking is sufficient to produce ambient pollution of 50 ug/m3. (3) How much difference does fuel choice make for indoor air pollution? After allowing for the effect of household ventilation characteristics, we find very significant 37 differences between biomass and "clean" fuels. As Table 8 shows, mean PM10 in clean- fuel households is 133 ug/m3 ­ little higher than ambient pollution in the urban area where most clean fuels are used. For biomass-using households, on the other hand, the average concentration is 242 ug/m3. Among biomass fuels, we fine a statistically significant difference only for jute (about 40 ug/m3 lower). However, as we noted in Section 7, very few households in our sample use jute as cooking fuel. For biomass fuels, a cautionary note is introduced by large differences in median and mean exposures recorded by our 24-hour monitors. Very large pollution "spikes" for short periods (e.g., 10,000 ug/m3 or higher) under some conditions can have large effects on estimated mean concentrations. Median concentrations are often much lower. This difference highlights the importance of better information about the time-structure of the relationship between pollution and respiratory disease. To illustrate the problem, for two houses with identical mean concentrations, is it better to experience two daily PM10 spikes of 5,000 ug/m3, with very low levels for the rest of the day, or constant exposure at 150ug/m3 over the 24-hour cycle? Limited evidence from time series studies has not yet provided a robust answer.5 Given the intensity of indoor pollution spikes under some conditions, further research on this issue seems warranted. (4) How important is fuel choice for pollution when we account for other household characteristics? Although fuel choice certainly affects indoor air pollution, our results suggest that its role is secondary to the role of ventilation factors for Bangladeshi 5 The available evidence is for outdoor particulates in industrial economies, and may have limited relevance for indoor air pollution from biomass fuels. Studies of short exposures to outdoor particulate concentrations suggest some impact on heart rate variability and the rate of heart attacks. However, a recent study in Palm Springs, California suggests that the short-period effect disappears when 24-hour average exposure is controlled for. Similarly, average exposures seem to dominate day-to-day variations in daily time series studies. Our thanks to Dr. Bart Ostro, Chief, California Office of Environmental Health Hazard Assessment, for his insights. 38 households. Moving from indoors to an open kitchen lowers the PM10 concentration in cooking and living areas by almost the same magnitude as switching from firewood to kerosene. Switching from mud walls to other materials lowers PM10 far more than any other factor, and use of thatch roofs also has a large effect. (5) In a representative sample of households, how serious is the indoor air quality problem for poor families in Bangladesh? For our assessment, we adopt the Indian rural PM10 exposure standard: a 24-hour average of 100 ug/m3. Our results for six Bangladeshi regions suggest that indoor PM10 concentrations are quite high for many poor families. For all rural families with per capita incomes below $1.00/day, we estimate a mean PM10 concentration of 275 ug/m3 for kitchen spaces -- nearly three times the Indian standard. This falls somewhat in peri-urban and urban areas (to 226 and 193 ug/m3, respectively), but remains much higher than the standard. (6) Are there significant geographic variations in indoor air quality? Our results suggest great geographic variation, even for households in the same per capita income group. This variation reflects local differences in fuel use and, more significantly, construction practices that affect ventilation. For the poorest households, rural PM10 concentrations vary from 410 ug/m3 in Cox's Bazar to 202 ug/m3 in Faridpur. Even in urban areas, concentrations differ by almost 100 ug/m3 between the highest areas, Jessore and Rajshahi, and the lowest, Sylhet. The poorest households in Rangpur face the same mean indoor concentration (198 ug/m3) as the highest-income households in Cox's Bazar. (7) What are the prospects for clean stoves and clean fuels? Our survey suggests that limited information may be a significant impediment to adoption of improved stoves. Only 15% of our sample households regard improved 39 stoves as a viable option, either because they have not heard of them or because they do not think they are locally available. Even among families that have considered the option, however, improved-stove use appears quite limited because of concerns about convenience or initial investment cost. The intermediate-term prospects for clean-fuel use appear more hopeful in urban and peri-urban areas, if economic growth continues. Although our sample evidence is limited, it suggests very high adoption rates among families whose daily per capita incomes exceed $2.00. Only 30% of extreme-poverty families use clean fuels, even in urban areas where their prices are relatively low. In rural areas, our evidence offers little hope for adoption of clean fuels in the near future. However, our analysis also suggests that poor families may not have to wait for clean fuels or clean stoves to enjoy significantly cleaner air. Within our sample household population, some arrangements are producing relatively clean conditions, even when "dirty" biomass fuels are used. Since these arrangements are already within the means of poor families, the potential for cost-effective improvements may be larger than is commonly believed. 40 References Airmetrics, 2004, MiniVol Portable Air Sampler, description at http://www.airmetrics.com/products/minivol/index.html. Balakrishnan, K., J. Parikh, S. Sankar, R. Padmavathi, et al., 2002, "Daily average exposures to respirable particulate matter from combustion of biomass fuels in rural households of Southern India," Environmental Health Perspectives, 110(11): 1069-1975. Beer, T., T. Grant, G. Morgan, J. Lapszewicz, P. Anyon, J. Edwards, P. Nelson, H. Watson and D. Williams, 2000, "Comparison of Transport Fuels: Final Report (EV45A/2/F3C) to the Australian Greenhouse Office on the Stage 2 study of Life-cycle Emissions Analysis of Alternative Fuels for Heavy Vehicles," CSIRO, University of Melbourne, Southern Cross Institute of Health Research. Boleij, J. , P. Ruigewaard, F. Hoek, et al., 1989, "Domestic air pollution from biomass burning in Kenya," Atmospheric Environment, 23: 1677-1681. Brauer, M. and S. Saxena, 2002, "Accessible Tools for Classification of Exposure to Particles," Chemosphere, 49: 1151-1162. Bruce, N., 1999, Lowering Exposure of Children to Indoor Air Pollution to Prevent ARI: the Need for Information and Action, Capsule Report (3), Environmental Health Project, Arlington VA. Caceres, D., M. Adonis, G. Retamal, P. Ancic, M. Valencia, X. Ramos, N. Olivares, L. Gil, 2001, "Indoor air pollution in a zone of extreme poverty of metropolitan Santiago, Revista Medica Chilena, 129:33-42. Campbell, H., 1997, "Indoor air pollution and acute lower respiratory infections in young Gambian children," Health bulletin, 55: 20-31. Freeman, N. C. G. and S. Saenz de Tejada, 2002, "Methods for Collecting Time/Activity Pattern Information Related to Exposure to Combustion Products," Chemosphere, 49: 979-992. Galassi, C., B. Ostro, F. Forastiere, S. Cattani, M. Martuzzi and R. Bertollini, 2000, "Exposure to PM10 in the Eight Major Italian Cities and Quantification of the Health Effects," presented to ISEE 2000, Buffalo, New York, August 19-22. Kammen, D., G. Wahhaj, and M. Yadom, 1998, Broad-Search Annotated Bibliography on ARI and Indoor Air Pollution, Environmental Health Project, Arlington, VA.(http://www.ehproject.org/PDF/ARIbibs/ARIBib2.pdf). 41 Kirkwood, B., S. Gove, S. Rogers, J. Lob-Levyt, P. Arthur and H. Campbell, 1995, "Potential Interventions for the Prevention of Childhood Pneumonia in Developing Countries: A Systematic Review," Bulletin of the World Health Organization, 73: 793-798. Leaderer, B., L. Naeher, T. Jankun, K. Balenger, T. Holford, C. Toth, J. Sullivan, J. Wolfson and P. Koutrakis, 1999, "Indoor, Outdoor, and Regional Summer and Winter Concentrations of PM10, PM2.5, SO42-, H+, NH4+, NO3-, NH3, and Nitrous Acid in Homes with and without Kerosene Space Heaters," Environmental Health Perspectives, 107 (3), March. Macro International, Demographic and Health Surveys, Beltsville, MD. McCracken, J., and K. Smith, 1997, Annotated Bibliography: ARI and Indoor Air Pollution, Environmental Health Project, Arlington VA. McCracken, J., R. Albalak, E. Boy, N. Bruce, J. Hessen, M. Schei, K. Smith,1999, "Improved stove or inter-fuel substitution for decreasing indoor air pollution from cooking with biomass fuels in highland Guatemala?", Indoor Air, 3:118-123. Moschandreas, D.J., J. Watson, P. D'Aberton, J. Scire, T. Zhu, W. Klein, and S. Saxena (2002) "Methodology of Exposure Modeling," Chemosphere, 49: 923-946. Murray, C. and A. Lopez (eds.), 1996, The Global Burden of Disease, Cambridge MA: Harvard School of Public Health, WHO, World Bank. Naeher, L., K. Smith, B. Leaderer, et al., 2001, "Carbon monoxide as a tracer for assessing exposures to particulate matter in wood and gas cookstove households of highland Guatemala," Environmental Science and Technology 35: 575-581. Pandey, K., K. Bolt,, U. Deichmann, K. Hamilton, B. Ostro and D. Wheeler, 2002, "The Human Cost of Urban Air Pollution: New Estimates for Developing Countries," World Bank Development Research Group Working Paper (forthcoming). Parikh, J., K. Balakrishnan, V. Laxmi, B. Haimanti, 2001, "Exposures from cooking with biofuels: pollution monitoring and analysis for rural Tamil Nadu, India, Energy 26: 949- 962. Pino, P., M. Oyarzun, T. Walter, D. von Baer, I. Romieu, 1998, "Indoor air pollution in South East Santiago," Revista Medica Chilena, 126: 367-374. Rosemarin, A., 2002, China Human Development Report 2002: Making Green Development a Choice, (Oxford: Oxford University Press), September. Saksena, S. and V. Dayal, 1997, Energy Environment Monitor, 13, 93-102. 42 Santos-Burgoa, C., L. Rojas-Bracho, I. Rosas-Perez, A. Ramirez-Sanchez, G. Sanchez- Rico, S. Mejia-Hernandez, 1998, "Modelling particulate exposure and respiratory illness risk in a general population," Gaceta Medica Mexicana,134(4): 407-417. Smith, K., 2000, "National Burden of Disease in India From Indoor Air Pollution," Proceedings of the National Academy of Science USA, (97) 13286-13293. Smith, K. and S. Metha, 2000, The Global Burden of Disease from Indoor Air Pollution in Developing Countries: Comparison of Estimates, Prepared for the WHO/USAID Global Technical Consultation on Health Impacts of Indoor Air Pollution in Developing Countries. Smith, K., 1993, "Fuel Combustion, Air Pollution Exposure, and Health: the Situation in Developing Countries," Annual Review of Environment and Energy, 18:526-566. Smith, K., Y, Liu, J. Rivera, et al., 1993. "Indoor air quality and child exposures in highland Guatemala," Proceedings of the 6th International Conference on Indoor Air Quality and Climate, University of Technology, Helsinki, 1:441-446. Thermo Electron, 2004, Personal DataRAM, pDR-1000, description at http://www.thermo.com/eThermo/CMA/PDFs/Product/productPDF_18492.pdf. WHO, 2001, Informal Consultation on Epidemiologic Estimates for Child Health 11-12 June 2001, Department of Child and Adolescent Health Development, WHO, Geneva (http://www.who.int/child-adolescent-health/New Publications/Overview/). WHO, 1999, World Health Report 1999: Making a Difference, Geneva. WHO, 1997, Health and Environment in Sustainable Development, WHO/EHG/97.8, Geneva. World Bank, 2001, Making Sustainable Commitments: An Environment Strategy for the World Bank, July. World Bank, 2002, "India: Household Energy, Indoor Air Pollution, and Health," ESMAP / South Asia Environment and Social Development Unit, November. World Resources Institute, 1998, World Resources 1998-1999: a Guide to the Global Environment, Oxford University Press. 43 Appendix Cooking and Ventilation Behavior, Structural Characteristics and Building Materials: Data Recorded for the Day of PM10 Monitoring and Tested for PM10 Impact via Regression Analysis 1. Characteristics of house No of stories in house No of rooms in the house (excluding toilet, kitchen and lawn) Is there a chimney? For cooking events during the monitored day: Cooking Period Number Cooked For Cooking Time Time Fire Continued After Cooking 1 2 3 ... 2. Characteristics of Living Area Roofing material 1. Tile 2. Thatched 3. Concrete 4. Corrugated Iron Wall construction material 1. Brick wall 2. Thatched 3. Concrete 4. Corrugated Iron 5. Wood 6. Mud 3. Characteristics of kitchen Roofing material 1. Tile 2. Thatched 3. Concrete 4. Corrugate Iron Wall construction material 1. Brick wall 2. Thatched 3. Concrete 4. Corrugated Iron 5. Wood 6. Mud 44 4. Location of kitchen Stove-1 Stove-2 Stove-3 Stove-4A Stove-4B Stove-5 Stove-6 45 5. House ventilation related factors: For kitchen: No of walls No of windows No of doors Location of windows/doors allows cross-ventilation? Any screen doors/windows? Doors, windows open after midday meal? Doors, windows open after evening meal? Vertical surface area of kitchen Length Width Height Vertical surface area of openings (doors/windows/other) out of kitchen ID Doors Windows Others please specify__________ Height Width Height Width Height Width 1 2 3 Ventilation area between walls and roof in kitchen? Number of walls with opening to the outside of the house: Number of walls with opening to the inside of the house: If there are two or more openings in the walls, are openings on opposite walls? For living area: No of walls No of windows No of doors Location of windows/doors allows cross-ventilation? Any screen doors/windows? Doors, windows open after midday meal? Doors, windows open after evening meal? Vertical surface area of living area: Length Width Height (Non-kitchen monitor site) Vertical surface area of openings (doors/windows/other) out of living area ID Doors Windows Others please specify__________ Height Width Height Width Height Width 1 2 3 46 Ventilation area between walls and roof in living area? Number of walls with opening to the outside of the house: Number of walls with opening to the inside of the house: If there are two or more openings in the walls, are openings on opposite walls? 6. Fuel Used on Monitoring Day 1. Firewood 2. Sawdust 3. Tree residue 4. Straw 5. Rice husk 6. Jute Sticks 7. Bagasse 8. Other crop residue 9. Briquette 10. Animal residue 11. Charcoal 12. Kerosene 13. Piped natural gas 14. LPG/LNG 15. Bio gas 7. Other Sources of Smoke on Monitoring Day Kitchen Cigarettes smoked Lanterns lit (time, fuel) Mosquito coils lit (time) Living Area Cigarettes smoked Lanterns lit (times, fuels) Mosquito coils lit (times) 47