Improving indoor air quality for poor families: a controlled

experiment in Bangladesh

Introduction

According to the WHOGlobal and Regional Burden ofDisease Report, 2004 (http://www.who.int/publica-tions/cra/en), acute respiratory infections from indoorair pollution (IAP – pollution from burning wood,animal dung and other biofuels) are estimated to kill amillion children annually in developing countries.Although IAP is a complex mixture of solid particles,liquid droplets, and gases, recent epidemiological studieshave reported that exposure to particulates, particularlysmall particulates [with diameter <10 lm (PM10)], isstrongly associated with respiratory illness and death.1

The design of cost-effective IAP reduction strategiesin developing countries has been hindered by lack ofinformation about actual PM concentrations in poorhouseholds. Data have been scarce because monitoringin village environments is difficult and costly.2 Rela-tively small-scale studies of indoor PM10 exposurefrom wood and coal combustion have been conducted

Abstract The World Health Organization�s 2004 Global and Regional Burden ofDisease Report estimates that acute respiratory infections from indoor air pol-lution (pollution from burning wood, animal dung, and other bio-fuels) kill amillion children annually in developing countries, inflicting a particularly heavytoll on poor families in South Asia and Africa. This paper reports on anexperiment that studied the use of different fuels in conjunction with differentcombinations of construction materials, space configurations, cookinglocations, and household ventilation practices (use of doors and windows) aspotentially-important determinants of indoor air pollution. Results from con-trolled experiments in Bangladesh were analyzed to test whether changes in thesedeterminants can have significant effects on indoor air pollution. Analysis of thedata shows, for example, that pollution from the cooking area is transported intoliving spaces rapidly and completely. Furthermore, it is important to factor inthe interaction between outdoor and indoor air pollution. Hence, the optimalcooking location should take �seasonality� in account. Among fuels, seasonalconditions seem to affect the relative severity of pollution from wood, dung, andother biomass fuels. However, there is no ambiguity about their collectiveimpact. All are far dirtier than clean (LPG and Kerosene) fuels. The analysisconcludes that if cooking with clean fuels is not possible, then building thekitchen with permeable construction material and providing proper ventilationin cooking areas will yield a better indoor health environment.

S. Dasgupta1, D. Wheeler2,M. Huq3, M. Khaliquzzaman4

1World Bank – Research-DECRG/Ru, NW Washington,DC, USA, 2Center for Global Development, Washington,DC, USA, 3Development Policy Group, Dhaka,Bangladesh, 4World Bank, Dhaka, Bangladesh

Key words: Indoor air; PM10 concentration; Ventilation.

S. DasguptaWorld Bank – Research-DECRG/IEMSN # MC-3-300 1818 H StreetNW, Washington, DC 20433USATel.: +1 202 473 2679Fax: +1 202 522 1151e-mail: sdasgupta@worldbank.org

Received for review 18 October 2007. Accepted forpublication 9 June 2008.� Indoor Air (2008)

Practical ImplicationsSeveral village-level measures could significantly reduce IAP exposure in Bangladesh. All would require arrangementsand the assert of male heads-of-household: negotiated bulk purchases of higher cost, cleaner fuels; purchase of morefuel-efficient stoves; peripheral location of cooking facilities; building the kitchen with permeable constructionmaterial; rotation of women in cooking roles, to reduce their exposure; and ventilation of smoke through a stack tallenough to disperse smoke over a relatively broad area. It is expected that village men and women will agree to thesemeasures if they become convinced that IAP poses a serious risk to health, and their actions will significantly reducethe risk. The keys to success are effective public education about the sources and risks of IAP, and financial andtechnical assistance for changes in cooking arrangements.

1Recent attention focuses on even finer particles (PM2.5).

2Two databases on measured indoor air pollution: (i) measurements of var-

ious air pollutants in urban and rural residences in China extracted from 110

published papers between 1980 and 1994, and (ii) measured indoor air pol-

lution in about 250 communities extracted from 71 studies published between

1968 and 2002 excluding studies already covered in the China database are

currently available on the WHO website. http://www.who.int/indoorair/

health_impacts/databases_china/en/index.html We are thankful to anony-

mous reviewer for this reference.

Indoor Air 2009; 19: 22–32www.blackwellpublishing.com/inaPrinted in Singapore. All rights reserved

� 2008 The AuthorsJournal compilation � Blackwell Munksgaard 2008

INDOOR AIRdoi:10.1111/j.1600-0668.2008.00558.x

22

(Dasgupta et al., 2006a,b). These studies have yieldedtwo main conclusions: natural gas and kerosene are farless pollution-intensive than biofuels such as wood anddung, and use of improved stove designs can signifi-cantly reduce indoor pollution from biofuels.Unfortunately, nationwide energy surveys in many

developing countries have revealed that poorhouseholds almost always use �dirty� biomass fuels,particularly in rural areas. This is often because cleanfuels (for e.g., LPG and Kerosene) are not available.Even where a clean fuel is available, most poorhouseholds use dirty fuels because the relative priceof the clean fuel is simply too high. Improved stoves forbiomass combustion could help, but studies in Asiaand Latin America have found almost no adoptionof improved stoves, despite widespread promotionalefforts. Households report non-adoption for a varietyof reasons, including capital and maintenance costs,inconvenience, and incompatibility with food prepara-tion traditions. Thus, neither clean fuels nor improvedstoves offer strong prospects for reducing IAP in ruralareas in the near future.Some of the previous studies have alluded to

construction materials, space configurations, cookinglocations and household ventilation practices (use ofdoors and windows) as potentially-important determi-nants of IAP (Brauer and Saxena, 2002; World Bank,2002; Freeman and Saenz de Tejada, 2002; Heltberget al., 2003; Moschandreas et al., 2002). In theory,many innovations in construction, space configurationand cooking practices could have significant effects onIAP. However, systematic research on IAP seems tohave focused primarily on the use of clean fuels andimproved stoves. We are not aware of any controlled,scientifically-monitored research on the relationshipsbetween structural arrangements and IAP in develop-ing countries.To promote better understanding of these relation-

ships, we have conducted a series of controlled,scientifically-monitored experiments in Bangladesh totest whether changes in construction materials, spaceconfigurations, and cooking locations can havesignificant effects on IAP. This paper summarizes ourresults. The remainder of the paper is organized asfollows. Section 2 discusses our controlled experiments,and describes how indoor air and the ambient envi-ronment have been monitored for this exercise. InSections 3 and 4, we present our experimental results.Section 5 provides a summary of conclusions.

Controlled experiments in Bangladesh

Previous World Bank research in Bangladesh, using thelatest air monitoring technology and a national house-hold survey, has found that IAP is dangerously high formany poor families. Concentrations of 300 lg/m3 orgreater for respirable airborne particulates (PM10) are

common in Bangladeshi households, implying wide-spread exposure to a serious health hazard (Dasguptaet al., 2006a). The findings also suggest wide variationin indoor air quality based on fuels, cooking locations,construction materials, and ventilation practices.The potential importance of these factors prompted

our follow-on program of direct experimentation, toovercome uncertainties about causation that are inevi-tably associated with cross-sectional survey analyses.The experiments focus on structural arrangements thatare already common among poor households in Ban-gladesh, as the literature on interventions to promoteclean fuels and improved stoves is replete with disap-pointments stemming from reluctance of poor familiesto adopt innovations that are unfamiliar, unsupportedby existing services, and potentially costly to maintain.Conceptually, our direct experimental approach

clarifies the assessment of structural changes by elim-inating the effects of unobserved, correlated variables.At the same time, we recognize that complete exper-imental control of household cooking arrangements isextremely difficult to achieve. In our experimentalarrangement, village women from other householdsagreed to cook identical meals at preset times, using thesame quantity of each fuel employed. In practice, therewere clearly variations in all these dimensions. Someundoubtedly reflected disregard of the agreement;others were inevitable from a pragmatic perspective.In either case, however, we have no reason to supposethat the variations were not random and uncorrelatedwith the exogenous changes in structures and fuels thatconstituted our experiments.The controlled experiments were conducted in Bur-

umdi village of Narayanganj district. Village Burumdiis located approximately 27 km away from the centerof Dhaka, the capital of Bangladesh in South-Eastdirection. At the time of the experiments, there wereapproximately 290 houses in the village with 1600inhabitants. Most of the villagers are either self-employed in non-agricultural sectors (e.g., fishermanand rickshaw puller) or are service providers. There areno obvious sources of outdoor air pollution. Thenearest secondary road, which is lightly traveled byvehicular traffic, is 3 km away. Burumdi residents, likemost other rural inhabitants of Bangladesh, do notpossess motor vehicles for private transportation.Neither do they use any motorized agricultural equip-ments, such as tractors or harvesters. The nearestmanufacturing industrial unit is again 3 km away, andother potentially polluting manufacturing industriesare beyond a radius of 10 km from the village.Architects familiar with climatic conditions and

cultural constraints faced by Bangladeshi householdsstudied building materials, housing configurations, andconstruction techniques in different regions of thecountry, and developed an appropriate set of structuraloptions. Local workers at each site were then hired to

A controlled experiment in air quality improvement

23

construct experimental houses that are indistinguish-able from structures actually used by poor families inthe area, using standard local building practices. Thefour sets of houses built had walls made of thatch,mud, corrugated iron (tin) and bricks (henceforthreferred to as Thatch, Mud, Tin, and Brick houses,respectively). The floor material varied between cementconcrete and mud; the Thatch, Mud and Tin houseshad mud flooring on rammed earth and the Brickhouse had cement concrete flooring following conven-tion. The walls and floors of each house were perma-nent, while roofing materials were altered to produce avariety of standard combinations. In the final experi-mental set, the Thatch house had tin and thatch roofs;the Mud, Tin and Brick houses had tin roofs, and in alater experimental stage the Brick house roof waschanged to concrete. Prevailing winds in the region arefrom south to north in the summer and north to southin the winter, so the axes of the houses were aligned inthe north-south direction to capture varying windconditions. Houses were furnished to create life settingsfor the experiments. See Appendix for details.Provisions were made for cooking fuel combustion in

four configurations that simulate common cookingarrangements: inside the house (within-dwelling); in aspace attached to it (attached kitchen); in a spaceenclosed by walls and a roof at a little distance fromthe house (detached kitchen); and in the open air. Thewall materials of attached and detached kitchens variedamong thatch, mud, tin and brick, floor materialsvaried among cement concrete and mud, while theroofs varied among thatch, tin and concrete. Thedesigns of the houses were flexible enough to permitalterations in building materials and kitchen configu-rations from time to time, as the experimental programdictated, without much delay.Arrangements were made with local residents; and

housewives cooked the standard daily meals3 for theirfamily of four on traditional stoves during mid-day forthe controlled experiments. Experimental combustionused diverse energy sources: clean fuels (kerosene andLPG), wood, cow dung, and other biomass fuels (ricehusks, jute etc.). However, only one type of fuel wasallowed per cooking session. The cooks were givencertain amount of fuels for cooking and actual amountof fuels used was recorded in a log book. The focus ofthe experiment was on preparing a standard meal for afamily of four, and we did not fix the cooking time orfuel use. In each case, a very small amount of kerosenewas used for easy initial ignition following the localpractice in Bangladesh. To eliminate other sources ofpollution, smoking, lighting of candles/oil lamps/ker-osene lamps, burning of mosquito/insect repellents

were not allowed. The experiments also allowed forturning on a ceiling fan in the living space, followingthe local custom. Table 1 summarizes the distributionof conducted experiments. Combinations of cookingarrangements, kitchen configurations and fuel wererestricted to cases generally observed in Bangladesh.4

Cooking and indoor air quality monitoring wereconducted during April 2005–June 2006, with theexception of the monsoon period (July 2005–Septem-ber 2005). In order to capture seasonal variation, thismonitoring period was specifically selected to includethe high-dust and low-dust season of Bangladesh,following local convention. Within our monitoringperiod, high-dust season spanned from November to

Table 1 Experimental configurations

Wall Roof FloorNo. ofexperiments

(a) By house typeBrick Concrete Concrete 37

Tin Concrete 63Tin Tin Mud 112

Thatch 0Mud Tin Mud 117

Thatch 0Thatch Tin Mud 89

Thatch Mud 80

Kitchen wall Kitchen roof FloorNo. ofexperiments

(b) By kitchen (construction material)Brick Concrete Concrete 16

Tin Concrete 24Mud 10

Tin Tin Concrete 8Mud 126

Thatch Mud 33Mud Tin Mud 46

Thatch Mud 34Thatch Tin Mud 99

Thatch Mud 53

Kitchen typeNo. ofexperiments

(c) By kitchen typeWithin dwelling 59Attached 165Detached 225Open 49

Fuel typeNo. ofexperiments

(d) By fuelClean 54Firewood 186Cow dung 100Others 158

3Standard daily meal of a rural family in Bangladesh includes rice, lentil,

vegetable curry and fish, and was followed for the controlled experiments to

reflect local reality.

4For example, cooking with cow dung/rice husk/jute inside houses is not

common in rural Bangladesh, and hence been excluded from our experi-

ments.

Dasgupta et al.

24

March, when humidity is low and rainfall is rare, andthe low-dust season covered April–June and October,when pre-monsoon thunderstorms and post-monsoonrainfall are frequent.The concentrations of airborne PM10 in the areas

surrounding the houses were also monitored at regularintervals during the controlled experiments. As theexperimental houses were located among standardhouses in villages, these measured concentrationscapture the impact of two potentially-important fac-tors: the volume of smoke emitted by other houses inthe neighborhood, and seasonal variations in rainfall,wind and humidity that affect the dispersion rate ofairborne particulates.

Monitoring PM10 concentrations

Our controlled experiments used two types of equip-ment: air samplers that measure 24-h average PM10

concentrations, and real-time monitors that recordPM10 at 2-min intervals.

• The Airmetrics MiniVol Portable Air Sampler (Air-metrics, 2004) is a more conventional device thatsamples ambient air for 24 h. While the MiniVol isnot a reference method sampler, it gives results thatclosely approximate data from U.S. Federal Refer-ence Method samplers. The MiniVols were placed inthe center of the room, were programmed to draw airat 5 l/min through PM10 particle size separator(impactors) and then through Pure Teflon filters. Theparticles were caught on the filters (height of theintake was about 1 m), and the filters were weighedpre- and post-exposure with a microbalance at Air-Metrics Laboratory in USA. In order to avoid con-tamination, the filters were transported to and fromBangladesh in air-tight petrislides via diplomaticpouch service.5

• The other instrument used in the study is a real-timemonitoring instrument: the Thermo Electric personalDataRAM (pDR-1000; Thermo Electron, 2004). ThepDR-1000 (pD-RAM) uses a light scattering pho-tometer (nephelometer) to measure airborne particleconcentrations. The operative principle is real-timemeasurement of light scattered by aerosols, inte-grated over as wide a range of angles as possible. At

each location, the instrument operated continuously,without intervention, for a 24-h period to recordPM10 concentrations at 2-min intervals.6

The readings of the pD-RAM and MiniVol airsamplers provided a detailed record of IAP concentra-tion in each controlled experiment.In addition, ambient PM10 concentrations for 24 h

were monitored 76 times during the experiments withMiniVol air samplers. MiniVols were placed at threelocations inside the village (not on periphery oroutside) to approximate the ambient PM10 of thevillage. The locations were carefully selected so thatthey are far from any kitchen yet within the inhabitedpart of the village. At each location, the MiniVolwas placed on a stool of about 20 inches height. Thereadings (reference: Table 2) revealed wide inter-aswell as intra-season variation in PM10 in outdoorenvironment.

Regression analysis of experimental data

Given the level of emissions from fuel use, theparticulate concentration in a space depends on thelength of time the emitted particles remain in the space,as well as ambient (outdoor) concentration. The extentand duration of smoke in the kitchen, and the amountof smoke leaking from the kitchen to the outdoors orinto other living spaces, may depend on severalstructural factors: the location of the kitchen, theextent of ventilation, and the permeable nature ofmaterials used to construct the roof and walls of thekitchen.We have conducted multivariate regression analysis7

with dummy variables to investigate the roles of threebasic determinants of 24-h average PM10 concentrationin indoor air: Kitchen configurations, building mate-rials, and fuels. In particular, the focus of ourregression analysis was on kitchen configurations(within-dwelling, attached, detached, and open), build-ing materials (brick, mud, tin, and thatch), and fuels:clean (kerosene and LPG), wood, dung and otherbiomass. As previously described, these elements have

5For reliable operation of MiniVol samplers, it is essential to maintain the

designed flow rate to maintain the cut off diameter of the particulate in the

impactor at the inlet of the samplers. The flow rate of all the samplers was

checked using a NIST traceable flow calibration unit before the start of any

sampling, so as to ensure proper flow rate during data collection. Inspection

and maintenance of the pump components were done regularly to ensure

proper pump operation at all times. Field operators were trained to record

and report any change in the sound level as changes in the sound level were a

good indicator for the malfunction of the pumps. Standard operations pro-

cedures for the MIniVols were followed according to the guidelines available

at http://www.airmetrics.com/products/minivol/basicop.html. For details on

calibration and QC, see Dasgupta et al., 2004.

6Two pD-RAMs used in the experiments were calibrated by comparing the

readings of pD-RAMS and MiniVols in co-located measurements. The

scatter plot of data obtained from recordings of two pD-RAMs over a period

of 24 h at 10 min logging intervals were checked for a slope value of unity,

low intercept (close to zero) and high R-squared value (>0.95). In case of a

significantly different value from unity for the slope, calibration of one of the

units was adjusted. The calibration of the pD-RAM which provided values

closer to MiniVol monitors was left unchanged and the other was adjusted to

yield unit slope in the plots. In case, the intercepts were significantly different

from zero, the zeroing of the units was done using the procedure in the

instrument manual using the zero air pouch supplied by the manufacturer.

The calibration process for the two pD-RAMs was repeated every 2 weeks or

so. For details on calibration and QC, see Dasgupta et al., 2004.7When a variable is determined jointly by multiple explanatory variables,

bivariate statistics produces biased estimates due to omission of relevant

variables (Johnston, 1963).

A controlled experiment in air quality improvement

25

been varied under fixed experimental conditions, withprescribed burn times for fuels, in both low-dust andhigh-dust seasons.In this section, we report regression results for

houses in which 24-h average PM10 concentrations inkitchens and living rooms were concurrently monitoredwith MiniVol equipment.

Kitchen results

In multivariate regression analysis, reliability (consis-tency and efficiency) of estimated coefficients dependson avoidance of collinearity, and proper treatment ofheteroskedasticity and seasonality. In our case, theexperiments conducted provided information on allfeasible combinations of structural arrangements andfuels. This design avoids the potential collinearityintroduced by field surveys of villages, where culturalor economic factors may produce intercorrelatedclusters of fuels and structures (Dasgupta et al.,2006a). We control for heteroskedasticity constructingregression standard errors from robust (Huber/White/Sandwich) variance estimates. We take into account ofseasonality by running separate regressions for high-and low-dust seasons,8 as well as variations in kitchenconfigurations that reflect seasonality. As no one cooksoutside during the low-dust season when rain isfrequent, open kitchen option has been precluded fromthe controlled experiments in the low-dust season. Toensure sample comparability, we include high-dust-season results for houses without open and detachedkitchen arrangements, as well as full results thatinclude open and detached configurations.

In Table 3, all the results highlight the importance ofseasonal conditions. The first three columns are for thehigh-dust season. The complete high-dust-season sam-ple is 225 experimental results, which falls to 177 forexperiments that are directly comparable to low-dust-season experiments (exclusion of detached and openkitchens). In the low-dust season, the sample is 250experimental results. Regression R-squared values varyfrom 0.29 to 0.41, indicating substantial randomvariation in ambient conditions.The overall impact of seasonality is captured by the

constant terms in the regressions. In all three high-dust-season regressions, the constant term is over150 lg/m3 [average of the three relevant constantterms from equations (1), (2), (3) being 155.29], whilein the low-dust season it drops to about one-third ofthat level (55 lg/m3). This seasonal difference of100 lg/m3 highlights the importance of seasonal vari-ations in village ambient (outdoor) air quality indetermining household-level IAP.As in any multivariate regression with dummy

variables, in order to avoid the problem of dummyvariable trap, for each set of determinant, we had toexclude one category from all categories (Johnston,1963). Hence, following convention, the regressioncoefficients should be interpreted as effect relative tothe excluded category. In our analysis, the excludedcategories from kitchen configurations, building mate-rials and fuels are open kitchen, thatch as wallmaterial, thatch and concrete as roof material, andstraw, respectively.The results for kitchen configurations provide an

interesting counterpoint to the conventional assump-tion that, controlling for fuels and building materials,IAP should be highest in the interior/within-dwellingkitchen (because it is the most contained cooking space)

Table 2 PM10 (lg/m3) concentrations recorded by (a) MiniVol Portable Air Samplers (24-h average) and (b) pD-RAM (2-min intervals). (c) Ambient PM10 (lg/m3) concentrations recorded byMiniVol Portable Air Samplers (24-h average)

Season Location of the monitor Mean s.d. Median Minimum Maximum

(a)High-dust Kitchen 222 79 213 39 473High-dust Living room 161 56 156 45 320Low-dust Kitchen 129 52 125 30 311Low-dust Living room 69 29 64 22 184

(b)High-dust Kitchen 585 1238 404 8 93,900High-dust Living room 468 2813 370 1 403,200Low-dust Kitchen 284 1060 146 1 109,900Low-dust Living room 843 9740 132 1 195,500

No. of readings

(c)High-dust 41 171 46 168 82 274Low-dust 35 54 23 48 15 125

8t-test of comparison of kitchen concentrations between high- and low-dust

season (t = 15.195) rejected the null hypothesis of equality of means and

justifies estimation of separate regressions.

Dasgupta et al.

26

and decline with progressive opening to the outside.However, this view reflects the implicit assumption thatoutdoor air is �clean�. In reality, the high-dust-seasonreverses configuration effects because intervening struc-tures filter the contaminated external air. Interior/within-dwelling kitchens, furthest removed from theoutdoors, have ambient concentrations 187 lg/m3

lower than open kitchens (excluded category) ordetached kitchens (difference of which is statisticallyinsignificant in comparison with the excluded category).Attached kitchens, next removed from the outdoors,have concentrations 49 lg/m3 less than open ordetached kitchens. In the low-dust season, the signreverses for attached kitchens although interior/within-dwelling kitchens exhibit no differential effect.In the high-dust season, we find consistent differen-

tial effects for kitchen wall materials. Significantpositive coefficients of brick and mud, as comparedto insignificant coefficient of tin-with thatch beingexcluded category implies concentrations are signifi-cantly higher for brick and mud than for thatch andtin. Quantitative estimates of the regression coefficientsfurther reveal that brick accounts for the highestincrement over thatch and tin – about 60 lg/m3 – whilemud has an increment of about 34 lg/m3. Again, theseresults are significantly changed by conditions in thelow-dust season: The increment for brick is small andinsignificant, and the increment for mud retains itmagnitude but changes signs. Tin remains insignificant.For roofing materials, tin accounts for a negative,

significant increment from thatch during the high-dustseason, but the significance disappears during the low-dust season.Even the contaminating effects of fuels are affected

by seasonal conditions, although the results retain theappropriate signs and high levels of significance in bothseasons. In the high-dust season, both wood and dungaccount for increments of around 70 lg/m3 relative toclean fuels. Other biomass fuels add an even greaterincrement – around 90 lg/m3. The order of effectschanges substantially during the low-dust season, withdung accounting for the greatest increment(109 lg/m3) over clean fuels, followed by wood(62 lg/m3) and other biomass fuels (57 lg/m3).

Living room results

Table 4 reports regressions for living rooms that aremonitored concurrently with kitchens using MiniVols.9

We introduce the kitchen PM10 concentration to controlfor dispersion of smoke from cooking. We estimate thekitchen concentration effect using both Ordinary LeastSquare (OLS) and Instrumental Variable (IV) regressionmethods, the latter to correct for possible pollutantbackflow from living spaces to kitchens (Greene, 1993;Maddala, 1988).

Table 3 Regression results for kitchen pollution

Dependent variable

Kitchen PM10 concentration (lg/m3, MiniVol, 24 h)

High-dust season Low-dust season

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

Kitchen layout (open excluded)Within-dwelling )187.314 (7.65)** )186.996 (8.81)** )187.067 (8.60)** 6.085 (0.85)Attached )49.251 (2.46)* )48.931 (3.03)** )49.155 (2.92)** 26.336 (4.12)**Detached )0.491 (0.03)

Kitchen wall materials (thatch excluded)Brick 60.400 (2.88)** 60.274 (2.93)** 62.073 (2.94)** 18.433 (1.83)Mud 34.348 (2.40)* 34.138 (2.56)* 35.519 (2.47)* )41.444 (5.22)**Tin 8.808 (0.64) 8.641 (0.67) 10.231 (0.73) 10.903 (1.67)

Roof materials (thatch and concrete excluded)Tin )23.395 (2.17)* )23.572 (2.30)* )22.945 (2.10)* )6.024 (0.87)

Fuels (straw excluded)Wood 70.593 (4.61)** 70.573 (4.62)** 71.139 (4.61)** 61.995 (6.66)**Dung 67.460 (3.71)** 67.459 (3.71)** 74.787 (3.85)** 109.297 (9.71)**Other 92.975 (5.75)** 92.967 (5.77)** 90.378 (5.61)** 56.516 (5.78)**Constant 156.079 (8.03)** 155.898 (8.81)** 153.886 (7.78)** 55.398 (4.21)**Observations 225 225 177 250R-squared 0.29 0.29 0.36 0.41

Column (1) presents results for all experimental variables and housing configurations in the high-dust season. Column (2) retains the sample but drops the control for detached housing,which does not attain marginal significance in any of the regressions. Column (3) replicates column (2), but for a sample that drops open-kitchen configurations for direct comparability withthe low-dust-season sample. Finally, column (4) provides results for the low-dust season.Absolute value of Robust t statistics in parentheses.*Significant at 5%; **significant at 1%.

9t-test of comparison of living room concentrations between high- and low-

dust season (t = 23.839) rejected the null hypothesis of equality of means

and justifies estimation of separate regressions.

A controlled experiment in air quality improvement

27

As before, comparisons of the constant terms inregression results indicate the importance of seasonality.Regression fits differ substantially between the twoseasons, with R2 near 0.60 during the high-dust seasonand 0.15 during the low-dust season.10 The constanteffects are again highly significant and very different, butfor the living room their role is reversed. In the IVestimates, the high-dust-season constant is 52 lg/m3,while during the low-dust season it doubles to 105lg/m3.During the high-dust season, as revealed by the

significant positive coefficient of Kitchen PM10 con-centrations in the regressions (1) and (2), dispersion ofkitchen pollution is a large, highly significant determi-nant of living-room pollution. Ceteris paribus, livingroom pollution increases by 0.5–0.6 lg/m3 for eachincrease of 1 lg/m3 in kitchen pollution. In the low-dust season, however, this effect essentially disappears.The OLS estimate is positive, small and significant,while the IV estimate is negative, small and marginallysignificant.For building materials, inter-seasonal results are

mixed. Relative to thatch, both mud and tin havenegative increments that are significant in both sea-sons. They are roughly constant in magnitude acrossseasons for mud, while they are substantially greater inthe high-dust season for tin. Brick-related incrementsare insignificant in both seasons. The effect of tinroofing relative to thatch is significant but contradic-tory across seasons. It is small in both seasons, butpositive during the high-dust season and negativeduring the low-dust season.The regressions also measure the effect of operating a

fan in the living room. During the high-dust season,

significant negative coefficient of the fan impliesoperating fan has a modest but significant negativeeffect on IAP. During the low-dust season, however,the effect disappears.

Variations in exposure patterns: pD-RAM results

Even if total particulate exposure is the same during a24-h period, brief, highly-concentrated exposures mayaffect health differently than sustained exposures. Inthis context, our experimental data provide usefulinformation on the relationship between exposurepatterns, building materials, fuels, and kitchen confi-gurations. Our MiniVols only record total 24-h expo-sures, so they cannot track exposure variation.However, our pD-RAM samplers measure pollutionat regular intervals over the 24-h cycle. We use thepD-RAMdata for a standardized assessment of kitchenexposure patterns during the mid-day meal preparationperiod. From each experiment, we draw 150 regular-interval pD-RAM observations covering five mid-dayhours, centered on the hour of maximum IAP exposure.We draw inferences about distribution patterns fromthe mean, maximum, minimum, standard deviation,and median for each experiment. Distributions withshorter but more extreme exposures have higher max-ima, means (pulled upward by higher maxima), andstandard deviations. Distributions with more uniformexposures have higher minima and medians.Table 5 provides regression results for 101 experi-

ments. We expect considerable random variations inthe exposure data, which combine variable cookingtimes with significant stochastic �noise� introduced bylarge, random spikes in the pD-RAM observations.Nevertheless, we obtain significant R-squared�s

Table 4 Regression results for living room pollution

Dependent variable

Living room PM10 concentration (lg/m3)

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

High-dust season Low-dust season

OLS IV OLS IV

Kitchen PM10 (lg/m3) 0.610 (12.82)** 0.516 (7.11)** 0.107 (3.20)** )0.114 (1.96)Living room wall materials (thatch excluded)

Brick 6.531 (0.90) 1.570 (0.18) )2.404 (0.39) )9.590 (1.33)Mud )22.002 (3.54)** )23.942 (3.57)** )15.390 (3.25)** )20.568 (3.98)**Tin )46.856 (5.13)** )41.781 (4.41)** )11.883 (2.39)* )11.217 (2.12)*

Living room roof materials (thatch excluded)Tin 11.058 (2.29)* 11.266 (2.10)* )15.207 (2.22)* )13.454 (1.78)Living room fan on )22.543 (2.54)* )19.087 (2.04)* )0.701 (0.11) )0.209 (0.03)Constant 31.199 (2.82)** 52.077 (3.20)** 75.277 (10.71)** 104.777 (9.96)**Observations 225 225 250 250R-squared 0.60 0.59 0.15

OLS, ordinary least square; IV, instrumental variable regression.Absolute value of Robust t statistics in parentheses.*Significant at 5%; **significant at 1%.

10R2 could not be computed for the low-dust-season IV equation.

Dasgupta et al.

28

(around 0.26), as well as high levels of significance forseveral of the right-hand variables. The results suggestthat building materials do not have significant effectson exposure patterns. However, wood and dungcombustion seem to generate more exposures thanother biofuels (the excluded fuel variable). Theirparticulate distributions have higher means, maximaand standard deviations; several estimated parametersare highly significant, and all are at least marginallysignificant. Attached Kitchen configuration also seemsto promote more exposures. The means, maxima andstandard deviations for Attached kitchen distributionsare all much higher than those for the excludedconfiguration: inside-dwelling, and with very highsignificance. In the opposite vein, Detached Kitchenconfiguration seems to promote more sustained expo-sure. Minimum and median particulate concentrationsfor Detached Kitchens are significantly higher thanthose for the excluded (inside-dwelling) configura-tion.11 The median result for the high-dust season isalso worth noting here. It suggests an overall increaseof about 86 lg/m3 during this season – a figure which isclose to the MiniVol result for kitchens.

Conclusions

Analysis of our experimental data yields a number ofconclusions that have significance for policy and villagecooking practices. First, outdoor air pollution is ahighly-significant determinant of indoor ambient pol-lution levels, particularly in the high-dust season. Theimportance of outdoor pollution implies a collectiveproblem with promoting the use of chimneys to expelcooking smoke. Thus, successful mitigation of IAP willrequire area and community-based approaches.12 Sec-ond, given the importance of outdoor air pollution,seasonality is very important in determining optimal

cooking arrangements. During the polluted high-dustseason, our results suggest that interior kitchens areactually much better than outdoor or detached facil-ities. This is not true during the low-dust season, butthen it is difficult to cook outside in any case. Third,building materials do make a significant difference forindoor pollution in the high-dust season. As far asquality of indoor air is concerned, in kitchen areas,brick walls are significantly more air-trapping thanmud walls, which are, in turn, significantly more air-trapping than thatch or tin walls. For kitchens, tinroofs are significantly better than thatch roofs. In livingrooms, tin walls are significantly better than mud walls,which are in turn better than brick or thatch walls. Insummary, with reference to IAP in Bangladesh, tinseems to be the best building material, followed bythatch and mud. Fourth, operating a living room fanalso provides significant benefits in the high-dustseason. Fifth, our results show that IAP from cookingsmoke can be a serious problem in living rooms as wellas kitchens. By implication, males� avoidance ofcooking areas does not protect them from IAP(Dasgupta et al., 2006b).13 Finally, among fuels, sea-sonal conditions seem to affect the relative severity ofpollution from wood, dung, and other biomass fuels.However, there is no ambiguity about their collectiveimpact: All are far dirtier than clean fuels, and use ofclean fuels should certainly be encouraged.

Acknowledgements

Financial support for this studyhas beenprovidedby theWorld Bank through Research Support Budget andTrust Funds administered by ESMAP. The findings,interpretations, and conclusions are entirely those of theauthors. They do not necessarily represent the view oftheWorldBank, its ExecutiveDirectors, or the countriesthey represent.

Table 5 Regression results for exposure distribution statistics

Mean Maximum Minimum s.d. Median

Brick 249.683 (1.78) 5765.622 (1.94) 24.741 (1.10) 642.883 (1.83) 27.575 (0.48)Mud 36.592 (0.27) )635.003 (0.22) )5.529 (0.25) )85.740 (0.25) 41.120 (0.74)Thatch 201.391 (1.50) 2405.807 (0.84) 31.963 (1.48) 339.155 (1.01) 46.685 (0.85)Wood 322.702 (2.45)* 8782.403 (3.15)** )14.011 (0.66) 1045.098 (3.17)** )9.783 (0.18)Dung 203.429 (1.70) 4407.647 (1.74) )14.287 (0.74) 590.197 (1.97)* )28.595 (0.58)Kitchen layout attached 257.893 (2.17)* 6584.797 (2.62)** 27.972 (1.46) 669.463 (2.25)** 67.878 (1.40)Kitchen layout detached 274.299 (2.01)* 3480.145 (1.20) 63.093 (2.87)** 400.850 (1.17) 181.843 (3.25)**High dust 88.887 (0.86) 498.348 (0.23) 22.689 (1.36) 42.903 (0.17) 85.526 (2.02)*Constant 49.277 (0.32) )1847.06 (0.57) 46.929 (1.90) )121.12 (0.32) 73.523 (1.17)Observations 101 101 101 101 101R-squared 0.20 0.27 0.29 0.24 0.31

Absolute value of t statistics in parentheses.*Significant a 5%; **significant at 1%.

11We found pD-RAMs hard to operate in the open due to high deposition

and insect entry into the optical chamber.12Community-based sanitation approaches have proven to be successful in

Bangladesh and other developing countries, Kar (2005).

13Our findings are in line with uniform daily exposure of adult men and

women to TSP reported in the Garhwal Himalaya (Saksena et al., 1992).

A controlled experiment in air quality improvement

29

Appendix

References

Airmetrics. (2004) MiniVol Portable AirSampler, description at http://www.airmetrics.com/products/minivol/index.html.

Brauer, M. and Saxena, S. (2002) Accessibletools for classification of exposure toparticles, Chemosphere, 49, 1151–1162.

Dasgupta, S., Huq, M., Khaliquzzaman, M.,Meisner, C., Pandey, K. and Wheeler, D.(2004) Guidance Document: ‘‘MonitoringIndoor Air Pollution’’, http://econ.worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/EXTPROGRAMS/EXTIE/0,,contentMDK:21311101~menuPK:4228103~pagePK:64168182~piPK:64168060~theSitePK:475520,00.html.

Dasgupta, S., Huq, M., Khaliquzzaman, M.,Pandey, K. and Wheeler, D. (2006a)Indoor air quality for poor families: newevidence from Bangladesh, Indoor Air, 16,426–444.

Dasgupta, S., Huq, M., Khaliquzzaman, M.,Pandey, K. and Wheeler, D. (2006b) Whosuffers from indoor air pollution:evidence from Bangladesh, Health PolicyPlan., 21, 444–458.

World Bank. (2002) ‘‘India: HouseholdEnergy, Indoor Air Pollution, and Health’’,ESMAP/South Asia Environment andSocial Development Unit, November.

Freeman, N.C.G. and Saenz de Tejada, S.(2002) Methods for collecting time/activ-ity pattern information related to expo-sure to combustion products,Chemosphere, 49, 979–992.

Greene, W.H. (1993) Econometric Analysis,New Jersey, Prentice Hall, Inc.

Heltberg, R., Kojima, M. and Bacon, R.(2003) Household Fuel Use and FuelSwitching in Guatemala, Joint UNDP/World Bank Energy Sector ManagementAssistance Program, World Bank\ESMAP, ESMAP Technical Paper 036,June.

Johnston, J. (1963) Econometric Methods,Tokyo, Kogakusha, Ltd, McGraw-Hill.

Kar, K. (2005) Practical guide to triggeringCommunity Lead Total Sanitation, KamalKar, UK, Institute of DevelopmentStudies, University of Sussex, http://www.ids.ac.uk/ids/booshop/wp/Wp257%20pg.pdf.

Maddala, G.S. (1988) Introduction toEconometrics, New York, MacmillanPublishing Company.

Moschandreas, D.J., Watson, J., D�Aberton,P., Scire, J., Zhu, T., Klein, W. and Sax-ena, S. (2002) Methodology of exposuremodeling, Chemosphere, 49, 923–946.

Saksena, S., Prasad, R., Pal, R.C. and Joshi,V. (1992) Patterns of daily exposure toTSP and CO in the Garhwal Himalaya,Atmos. Environ., 26A, 2125–2134.

Thermo Electron. (2004) Personal Data-RAM, pDR-1000, description at http://www.thermo.com/eThermo/CMA/PDFs/Product/productPDF_18492.pdf.

Thatch house

Thatch house:

Living space: 1400 Cu.Ft. (volume below false ceiling)500 Cu.Ft. (volume above false ceiling)

Attached / Detached Kitchen: 267 cu.Ft.

Floor: Mud on rammed earth

Dasgupta et al.

30

Tin house

Tin house:

Living space: 1780 cu.Ft. (volume below false ceiling)386 Cu.Ft. (volume above false ceiling)

Attached kitchen : 223 cu.Ft.

Detached kitchen: 220 cu.Ft.

Tin house with indoor kitchenLiving space: 1110 cu.Ft. (volume below false ceiling)

Indoor kitchen: 661 cu.Ft.

Mud houseLiving space: 2692 cu.Ft. (volume below false ceiling)

768 Cu.Ft. (volume above false ceiling)

Attached kitchen: 235 cu.Ft.

Detached kitchen: 284 cu.Ft.

Floor: Mud on rammed earth

Mud house:

A controlled experiment in air quality improvement

31

Brick house

Brick house:

Brick house (tin roof)Living space: 2295 cu.Ft. (volume below false ceiling)

500 Cu.Ft. (volume above false ceiling) Attached / Detached kitchen: 300 Cu.Ft. Floor: Cement Concrete

Brick house (tin roof) with indoor kitchenLiving space: 1473 Cu.Ft. (volume below false ceiling)Indoor kitchen: 788 Cu.Ft. Floor: Cement Concrete

Brick house (R.C.C. ROOF)Living space: 2295 Cu.Ft. Attached / Detached kitchen: 300 Cu.Ft. Floor: Cement Concrete

Brick house (R.C.C. ROOF) with indoor kitchenLiving space: 1473 Cu.Ft. Indoor kitchen: 788 Cu.Ft.Floor: Cement Concrete

Dasgupta et al.

32