Policy Research Working Paper 8930 Do Improved Biomass Cookstoves Reduce PM2.5 Concentrations? ? If So, for Whom? Empirical Evidence from Rural Ethiopia Randall Bluffstone Daniel LaFave Alemu Mekonnen Sahan Dissanayake Abebe Damte Beyene Zenebe Gebreegziabher Michael A. Toman Development Economics Development Research Group June 2019 Policy Research Working Paper 8930 Abstract Improved biomass cookstoves have been promoted as air pollution by 64 percent and median household air pol- important intermediate technologies to reduce fuelwood lution by 78 percent—although the resulting household consumption and possibly cut household air pollution in air pollution levels are still many times greater than the low-income countries. This study uses a randomized con- World Health Organization’s guideline. These large percent- trolled trial to examine household air pollution reductions age reductions may reflect decreased emissions due to less from an improved biomass cookstove promoted in rural use of fuelwood, given Mirt’s energy-efficient design, and Ethiopia, the Mirt improved cookstove. This stove is used the likelihood that higher-emissions three-stone cooking is to bake injera, which is very energy intensive and has a very moved outside the main living area once a Mirt improved particular cooking profile. In the overall sample, the Mirt cookstove is installed. Households in the subsample who improved cookstove leads to only minor reductions in mean experience a greater decline in household air pollution tend household air pollution (10 percent on average). However, to be less wealthy and more remotely located and burn for those who bake injera in their main living areas, the less-preferred biomass fuels, like agricultural waste and Mirt improved cookstove reduces average mean household animal dung, than households that cook in a separate area. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The lead author and task team leader may be contacted at bluffsto@pdx.edu and mtoman@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Do Improved Biomass Cookstoves Reduce PM2.5 Concentrations? If So, for Whom? Empirical Evidence from Rural Ethiopia Randall Bluffstone*, Daniel LaFave, Alemu Mekonnen, Sahan Dissanayake, Abebe Damte Beyene, Zenebe Gebreegziabher, Michael A. Toman Key Words: Household Air Pollution, Indoor Air Quality, Ethiopia, Improved Stove JEL Codes: I15, Q55, Q23, Q42                                                         * Corresponding author: Bluffstone, Portland State University, Portland, OR, USA bluffsto@pdx.edu Dissanayake: Portland State University, Portland, OR, USA; Beyene: Ethiopian Development Research Institute, Addis Ababa, Ethiopia; Gebreegziabher, Ethiopian Development Research Institute and Mekelle University, Addis Ababa and Mekelle, Ethiopia; LaFave: Colby College, Waterville, ME, USA; Mekonnen, Ethiopian Development Research Institute and Addis Ababa University, Addis Ababa, Ethiopia; Toman, The World Bank, Washington DC, USA. Financial support for this research from the World Bank and the Knowledge for Change Trust Fund is gratefully acknowledged.   Do Improved Biomass Cookstoves Reduce PM2.5 Concentrations? If So, for Whom? Empirical Evidence from Rural Ethiopia 1. Introduction Approximately 40% of the world, or almost 3 billion people, rely on solid fuels, such as coal, wood and charcoal for cooking (WHO, undated; IEA, 2017). This proportion is much higher (68% to over 90% of the population) in Sub-Saharan Africa and is expected to increase through 2030 (IEA 2017, Rehfuess et al., 2006; Smith et al., 2004). In Ethiopia, which is the focus of this paper, the percentage is about 90% (Beyene et al, 2015a), which is typical for low-income countries (WHO, undated). Particularly because of the human respiratory effects of burning solid fuels in homes, but also because of environmental effects (Masera et al., 2015), shifting households from biomass to commercial energy, such as gas and electricity, is a key goal. Such commercial fuels are used for cooking by about three-fifths of the world (Smith et al., 2013, IEA, 2012), but as of 2010, about 1.3 billion people did not have access to electricity, mainly in rural areas of developing countries (MacCarty et al., 2008; Jeuland and Pattanayak, 2012), and many more find commercial fuels too expensive or irregularly supplied to use for cooking and heating. Reliance on biomass fuels is likely to continue for the foreseeable future in Sub-Saharan Africa and elsewhere in low and lower-middle income countries (IEA 2017). Cookstoves burning gas and electricity produce little or no household air pollution (HAP), but at present such technologies are very rarely used in rural areas of Ethiopia due to their high costs and insufficient access to electricity. Relatively simple improved cookstove (ICS) technologies that use less biomass and may require only minor changes in household cooking habits are therefore potentially important intermediate technologies (Jeuland and Pattanayak, 2012) if they are truly appropriate for users (Mobarak et al., 2012). In this paper, we analyze the Mirt (“best” in Amharic language) ICS, which is promoted in Ethiopia, and used to make injera, which is a type of pancake or crepe made from teff, which is one of the main staple grains grown in Ethiopia. In areas without refrigeration, injera is generally baked at least twice per week. That cooking does not occur every day makes injera a rather unique product. It is also very energy intensive per cooking event and overall; injera baking is the end-use for approximately 50% of all primary energy consumed in the country (Bizzarri, 2010; Tesfay 2014). Though not nearly as flexible or portable as the traditional technology, which is the three- stone tripod, the Mirt stove is more efficient than the traditional stove and has been estimated to use 2    50% less wood in laboratory tests (GIZ, 2011), 40% to 50% based on surveys (Megen Power, 2008; Dresen et al., 2014), and 20% to 30% less in field-based controlled cooking tests (Gebreegziabher et al., 2018). The Climate Resilient Green Economy strategy of the Federal Government of Ethiopia (FDRE, 2011) proposes to distribute ICS that will be used by about 20 million households by 2030. Reducing the demand for biomass by increasing fuel efficiency is indeed currently one of the strategic priorities in the Ethiopian energy sector (FDRE, 2014). Our paper builds on previous work based on a randomized controlled trial rolled out in 2013 and examines the effects of the Mirt ICS promoted in Ethiopia on HAP in cooking areas, measured as 72-hour mean, median and maximum PM2.5 concentrations. We also investigate the types of households that receive HAP benefits. In previous papers we have reported on satisfaction and per- meal fuelwood savings (Gebreegziabher et al., 2018), actual in-field usage (Beyene et al., 2015b), and learning (Bluffstone et al., 2017). This paper extends the analysis to HAP, using a cross-sectional randomized controlled trial (RCT) implemented in 2016. We find average, maximum and median PM2.5 concentrations that are very much in line with the literature (e.g. MacCracken and Smith, 1998; Chen et al., 2016), and several orders of magnitude greater than the WHO guideline concentration values. Our analysis suggests that the Mirt ICS on average has limited effects on HAP, with average of mean concentrations only 10% lower in households randomly assigned to receive a Mirt stove. Digging deeper, we echo findings of Langbein et al. (2017) that cooking location matters for ICS results. We find large reductions in HAP, including average mean reductions of 64% and median reductions of 78%, compared with those who did not receive Mirt stoves, for households that have their cooking/baking areas in their main houses;1 for these households, adoption of the Mirt ICS may offer significant health benefits. We analyze the characteristics of these households and find they tend to be poorer by several measures, live farther away from all-weather roads, and are more likely to cook with less preferred, smoky fuels like animal dung and agricultural wastes than households who have a separate cooking and baking area. Though our study design does not allow us to identify the reasons for the large percentage reductions in HAP due to Mirt use by households who cook/bake in their main living areas, they may reflect the decreased emissions made possible by greater stove energy-efficiency and                                                         1  Though many kitchen areas are distinct rooms for the purpose of cooking and baking, many households combine cooking/baking with living space in small houses. To avoid giving the impression that cooking/baking areas are always “kitchens” in the sense of separate rooms, we generally use the term “cooking/baking areas” rather than “kitchen.”   3    reduced use of fuelwood, combined with the possibility that traditional three-stone cooking is moved out of the main living area once a Mirt stove is installed. The paper is organized as follows. In the next section we present key literature. Section 3 discusses methods, including data collection, sampling and empirical methods. Section 4 presents results and section 5 discussion and conclusions. 2. Key Literature Particulate matter (PM) is a general term for small solids and liquid droplets suspended in the atmosphere. In Africa, biomass fuel burning constitutes the major source of ambient PM (Karagulian et al. 2015). Particles with a diameter size between 2.5 and 10 μm (PM2.5–PM10) are considered coarse, and those with diameter 0.1 μm - 2.5 μm are fine and those with diameter less than 0.1 μm are considered ultra-fine particles (Anderson et al 2012). PM2.5 constitutes 50–70% of PM10 in most locations in Europe (WHO-ROE, 2013). PM with a diameter greater than 10 μm have relatively short suspension half-lives and are largely filtered out by the nose and upper airway. As the largest health damages come from particles less that PM10, much of the focus has been on fine and ultra-fine particles (Jeuland et al., 2015). There is little question that HAP from solid fuel burning in homes around the world is a critical health hazard, especially for women and children (Smith et al., 2004). The important global burden of disease effort, focusing primarily on particulates for which the best evidence exists, argued that about 3.5 million premature deaths are caused each year by HAP due to the indoor combustion of solid fuels (Lim et al., 2013).1 An additional 0.5 million deaths are attributable to the particle emissions emitted by homes into the outdoor environment, where they represent 16% of total outdoor concentrations (Smith et al., 2013). The WHO estimates that 4.3 million people annually die prematurely due to HAP-related respiratory illnesses, which is more than the 3.7 million total premature deaths attributable to ambient air pollution. All but 20,000 of these HAP-related deaths are in low and middle-income countries, with 580,000 in Africa (Jeuland et al., 2015; Martin et al., 2011; WHO, 2014). In their 2017 meta-analysis, Achilleos et al. (2017) find a mean 0.89% increase in overall mortality, a 0.80% increase in cardiovascular deaths, and a 1.10% mean increase in respiratory                                                         1 Six percent of the global burden of disease is due to lower respiratory infections, which may be linked to HAP, which is second only to ischemic heart disease. In 2000 and 2011, lower respiratory infections were the primary cause of reduced health-related life satisfaction measured as disability adjusted life years (WHO, 2014, WHO, 2013).   4    mortality per 10 μg/m3 increase in PM2.5. They particularly implicate elemental carbon and potassium as critical elements of PM2.5 leading to increased mortality. Salvi (2015) documents that chronic obstructive pulmonary disorder (COPD) is the third most important cause of mortality worldwide, that COPD incidence is very high in Sub-Saharan Africa, and women often spend 3 to 5 hours per day in smoky environments due to indoor cooking with biomass fuels. The author argues for additional attention to reducing the incidence of COPD in Sub-Saharan Africa. The literature suggests that ICS, even if households continue to use biomass fuels, can reduce HAP. Quansah et al. (2017) conduct a systematic review of the literature and meta-analysis across a variety of interventions. Based on standardized mean difference measures, they find ICS has the largest effects for total particulate matter, with average reductions of 18% in daily average personal PM exposures, and larger effects for children. In their review of the literature, Pope et al. (2017) find large (39%) average effects of ICS on PM2.5 even for improved biomass stoves without chimneys. In their extensive lab-in-field study conducted in rural Nepal, Ojo et al. (2015) also find large average reductions in PM2.5 concentrations when biomass ICS are introduced. They identify very high baseline mean concentrations, typically exceeding 3,000 μg/m3and reductions in HAP concentrations of 50% - 60% due to the two improved biomass cookstoves they evaluated. We contribute to this literature on the effects of ICS on HAP using an RCT involving the Mirt stove, which started in Ethiopia in 2013 and continued until 2016. In this paper we focus on the effects on PM2.5 concentrations of the intent to treat with the Mirt ICS. Health effects are investigated in related research (LaFave et al, 2018). 3. Methods 3.1 Experimental Design In this section we present the experimental method we use to identify the effect of the Mirt ICS on HAP. The treatment is provision of the Mirt ICS to randomly-selected households in 36 villages across three regional states in Ethiopia. The control is not receiving the Mirt stove and using only the traditional stove, which is the three-stone tripod. We emphasize that in virtually all households who received the Mirt stove, households continue to actively use their traditional cookstove for non-injera cooking. This is because Mirt is highly specialized for cooking injera and the main burner is approximately 50 cm in diameter, which is much larger than for cooking other foods. It is therefore necessary to have a second stove to cook foods other than injera, such as stews 5    and coffee.2 We also note that cooking injera, though occurring every 2-3 days, uses a majority of the fuelwood consumed by households, suggesting that improving the cooking of injera could lead to HAP improvements. The design of the stove is shown in Figures 1 and 2. Figures Figure 1 Figure 2 Mirt Stove with Injera Cooking Cook pouring Injera batter on Mirt Stove   Source: ethiopiaethos.files.wordpress.com/2010/06/comp5.jpg Source: energypedia.info/wiki/Baking_with_Improved_Ovens In June-July 2013, 360 Mirt stoves were distributed to 360 rural households in Ethiopia. Households were randomly selected to receive the stove using stratified random sampling at the regional state level to be representative of the state population and forest cover. The three regional states are Amhara, Oromia and Southern Nations, Nationalities and Peoples Regional State, which contain approximately 70 percent of the Ethiopian land area and 80 percent of the population. Households in these regional states generally cook injera and the traditional technology is the three- stone tripod; the cuisines and baseline cooking technology are therefore similar across households. A total of 10 stoves were distributed to 10 randomly selected households in each of 36 randomly selected villages.3 As of November-December 2016, when data collection for this paper was conducted, 121 stoves were no longer in use. Approximately ¼ of these were in storage and all except one of the others were discarded, with virtually all abandoned due to breakage. This finding is not surprising, because the stove has an estimated five-year lifespan.                                                         2 The Mirt stove also has a second burner to cook stews and coffee, which can be used in conjunction with when injera is being cooked. 3 Stoves were distributed under 3 different monetary treatments – given for free, pay a subsidized cost and paid if used regularly during first 6 weeks. We observed no differences in initial uptake across the monetary treatment arms (Beyene et al., 2015b), so we group all households distributed a Mirt stove in a single treatment for this paper. 6    At the same time as the initial stove disbursement in 2013, 4 control households were randomly selected in each village, for a total sample size of 504 households. Baseline demographic, socioeconomic and other data were collected from both treatment and control households in June- July 2013 and confirmed balance across the treatment and control groups. No baseline HAP measurements were collected and therefore we utilize a cross-sectional randomized study design to compare air pollution levels across households who randomly received a Mirt and those that did not. In November-December 2016, 6 of the 14 households under study in each village were randomly selected for HAP monitoring. Three of the households in each village were treatment and three control, for a total sample size of 216. In the field research design, if a randomly selected household was given a Mirt stove that was subsequently abandoned, another household from the same village was randomly selected; only households currently with Mirt stoves in place (though perhaps not actively used) were compared to households that never received a Mirt stove.4 This sample size was chosen based on standard conventional power calculations in the HAP measurement literature (see Edwards et al., 2007). These standard conventions include achieving a statistical power of 0.80, a p value of 5% in two-tailed tests and detecting a 30% HAP reduction. A reliable, Ethiopia-based, estimate of the coefficient of variation in HAP reduction was not available prior to our study to compute the minimum sample size. We therefore used a conservative COV estimate of 0.7 (Edwards et al., 2007), which, given our cross-sectional study design, gave a minimum sample size of 86 households in each arm (total 172). Allowing for possible monitoring equipment failure, we utilized a sample size of 216 (108 in each arm). We collected valid data from over 95 percent of the target households, with missing data due to HAP monitoring equipment failures and a small number of households where all members were in the field for coffee harvest on the day of data collection. We note, though, that these failures were random across the treatment and control households and such issues were not unexpected. The sample size remains sufficient to achieve the standard level of power with a modest HAP reduction detection level. Maximum, mean and median PM2.5 levels were measured in the 204 households using Particle and Temperature Sensor Plus (PATS+) light-scattering particle sensors developed by Berkeley Air Monitoring Group of Berkeley, California. The PATS+ devices were in place in                                                         4In the appendix we show that based on 23 observables the sample is balanced across those who were subject to HAP monitoring and those who were not monitored. The subsample of households who experienced HAP monitoring is therefore quite representative of our overall sample. 7    households’ cooking/baking areas for 72 hours. Because households typically cook injera at least twice per week, all households should have baked injera during the 72-hour monitoring period.5 PATS+ results were calibrated using gravimetric filters co-located for exactly 24 hours in 50 of the sample households. Half of these households received Mirt stoves and half did not. Such calibration is necessary when using light scattering particle sensors (Berkeley Air Monitoring Group, 2017; Chow et al., 2002). Based on these findings, OLS regression was used to estimate an adjustment factor of 0.8065, which was applied to all PATS+ measurements. In gathering HAP information, field enumerators were instructed to follow strict protocols developed by Berkeley Air Monitoring Group. These protocols, which focus on placement of the equipment in cooking areas, maintenance of the equipment and timing of the samples, are included in the Appendix to this paper. In addition to measuring mean, median and maximum PM2.5 concentrations, the PATS+ monitors also measured humidity and temperature. Key data on cooking/baking area characteristics, including size, shape, presence of windows and location (in the main living area or separate), were collected as well as information on the status of Mirt stoves, including if they were broken or unused, and whether traditional stoves were still in place. We also collected updated information on fuels used.6    3.2 Empirical Methods With sufficient sample size and a randomized intervention, if the randomization is successful across the treatment and control groups, we can simply compare sufficient statistics of those in the treatment and control groups to identify the treatment effect. To evaluate the quality of the randomization we compare households receiving Mirt stoves with those who did not receive stoves along 39 dimensions, with a special emphasis on cooking/baking area and environmental characteristics that are very likely to affect HAP concentrations. We also test for balance on smoking behaviors, socioeconomic characteristics, and cooking/eating frequency. We find that only one of the 39 variables is statistically different across the groups, a rate that is just as likely to be due to chance given a 5 percent significance threshold.                                                         5 Only 2 households (<1%) report on average cooking injera less than twice per week. 6 The research design allowed for the possibility that households cooked in two places in their homes (e.g. in main house and a separate kitchen), but no households had two kitchens. Two HAP-monitored households reported they did not use their Mirt stove and one had purchased a replacement. 8    In the following section we describe the sample, randomization balance checks and treatment effect estimates. We find only minor reductions in mean HAP due to Mirt in the overall sample, but significant differences in HAP depending on the location of baking injera, where the choices are in the main house or inside, but in a separate cooking/baking area.7 We conduct a number of robustness checks, including running regressions, and conclude that for those who cook/bake in their main living areas, the Mirt stove reduces the average of mean HAP by 64% and median HAP by 78%. Finally, we examine the decision to cook/bake in main living areas using probit models. We find that variables consistent with poverty, isolation and less preferred fuels are most correlated with cooking/baking in households’ primary residences. 4. Data and Estimation Results 4.1 Descriptive Statistics The households in our sample mainly cook in separate, indoor cooking/baking areas, about half of which are round shaped and half that are rectangular. Very few cooking/baking areas have windows and whether or not a Mirt stove was provided in 2013, households continue to use three- stone tripods for cooking foods other than injera. About ¾ of all households cook and bake injera in cooking/baking areas that are separate from their main houses. Average kitchen size is larger for the ¼ of households who cook in their main houses, but those cooking/baking areas are multi- purpose. The average house only has 2.6 rooms and the average household has 9.3 adult equivalent members, so rooms in main houses are very unlikely to be completely dedicated to cooking. Virtually all households use fuelwood, but substantial minorities also use smaller branches, dung and agricultural wastes; commercial fuels are rarely used and never for injera baking. About 55% of respondents are Ethiopian Orthodox and 25% are Muslim. The remainder are other Christian denominations. A total of 6.86% of households self-identify as female-headed. Means for treatment and control households are provided in the Appendix.                                                         7 Only one household cooks/bakes outside in an uncovered location. 9    4.2 Randomization Balance Tests As shown via balance tests in Appendix Table A2, the subsample of households who were subjected to HAP monitoring (n=204) are representative of the overall sample (n=5048). We test balance across treatment and control households using 39 variables for the subsample that had HAP monitoring. A total of 23 variables are from the baseline survey and 16 are from the endline survey. Though randomization occurred prior to the baseline, we add endline variables related to temperature, humidity, cooking/baking area floor area, other cooking area characteristics and cooking characteristics from 2016 that were not collected in 2013, because these variables may affect 2016 HAP measurements and are invariant to whether a household is in the treatment group. We find that all 2016 variables are balanced across treatment and control subsamples, suggesting that HAP monitoring will yield an unbiased estimate of the effect of Mirt on HAP. From the 2013 baseline survey, only one variable – existence of metal roofs on main houses - is not balanced at the 5% significance level, which is a rate that is just as likely to be due to chance given the 5% significance threshold; though a majority of both treated and control households have metal rather than thatched roofs, treated houses are more likely to have metal roofs. All balance test results are presented in the Appendix. 4.3 Treatment Effects of the Mirt ICS on Household Air Pollution Table 1 presents the basic findings from the HAP monitoring. As shown in Table 1, consistent with the literature (Ojo et al., 2015; Chen et al., 2016; MacKracken and Smith, 1998), the mean and maximum values are extremely high. The average of the mean concentrations across the sample is 1,231 g/m3. This figure is about 50 times the WHO (2015; 2006) 24-hour guideline of 25g/m3.9 The average of the mean concentrations for households with Mirt stoves is slightly lower than for those who did not receive a Mirt stove and use only the traditional technology (1,162 g/m3versus 1,297 g/m3). The average of measured median PM2.5 concentrations is 149.39g/m3, with Mirt stove households again a bit lower at 145.12g/m3.                                                         8 In the endline, when HAP monitoring was conducted, enumerators could not contact 23 households. Attrition was balanced across treatment and control arms of the study, and uncorrelated with household characteristics likely to determine HAP. In balance tests, samples sizes are therefore less than 504. 9  We omit three observations that could be outliers, because the mean is over 19,000 g/m3 during the three-day monitoring period.  10    Table 1 Mean, Median and Maximum PM2.5 Concentrations in Cooking/baking areas of Households who did and did not Receive Mirt Stoves in 2013 Mean Median Maximum Mirt, Generally with 1162.11 (1883.21) 145.12 (789.81) 45,465.21 (36,595.46) Traditional Too [98] [99] [98] Traditional Only 1297.03 (1990.24) 149.39 (710.42) 49,649.25 (45,976.89) [103] [105] [105] Standard deviations in parentheses and sample size in brackets. Though the stoves may not be used, all treated households have Mirt Stoves in place. Consistent with other studies in similar environments, maximum concentrations are extremely high in areas where biomass is burned. The average of the maximum concentrations is 47,629 g/m3.10 Households that received a Mirt stove have lower average maximum concentrations (45,465 g/m3 versus 49,649 g/m3). In the overall sample statistics reported in Table 1, we do not find statistically significant evidence that the Mirt ICS reduces PM2.5 concentrations (average mean, maximum and median). Using two-tailed t-tests that assume normality and allow for unequal variances and Kruskal-Wallis rank-sum 2 tests with ties that allow for non-normality, we do not find evidence of statistically significant differences.11 We next consider differences in the impact of the Mirt stove based on the location of the cooking environment, focusing on those who cook in their primary living areas. Approximately ¾ of households (n=151) in the sample have located their cooking/baking areas outside the main house where people generally live. All but one of these cooking/baking areas are indoors, though in a separate building. The remaining 53 households have cooking/baking areas located in the house where people live, so these areas serve many purposes besides cooking. There is no difference in treatment assignment depending on whether injera baking occurs in the main house or a separate cooking/baking area (i.e. Mirt stoves did not tend to be differentially installed in main houses or separate cooking/baking areas). We therefore do not conflate cooking location with our treatment effect. Results are available from the first author. We test randomization across treatment and control subsamples, but now restrict our analysis to the subsample of households who cook in their main living areas, again using the 23                                                         10  We omit one observation that may be an outlier, because the measured maximum concentration was over 400,000 g/m3.  11 Based on Shapiro-Wilk W tests for normality, we are able to reject normality at < 0.01 level. 11    variables collected in 2013 and the 16 variables from 2016 used to test overall balance. Out of 39 variables, no variables from 2016 were unbalanced and only three from 2013 were not balanced at the 5% level based on Kruskal-Wallis tests. Respondents with Mirt stoves were more likely in 2013 to have metal roofs (p-value = 0.0019), which is the same as the overall sample, are less likely to report being food secure over the previous year (p-value = 0.049), and are more likely to drink alcohol (p-value = 0.052). We find significant effects of the Mirt stove on average PM2.5 concentrations if households have their primary baking area in their main house, but not for households with separate cooking/baking areas. When cooking/baking is done in the main living areas, average mean concentrations of PM2.5 are estimated to be 64% lower for those who have the Mirt ICS than households without Mirt stoves (592 g/m3 versus 1622g/m3, two-tailed t-test p value=0.03, Kruskal-Wallis rank sum test with ties prob >2=0.10). There are no statistically significant differences in maximum concentrations, but in percentage terms, effects are very large if central tendency is measured at the median. We find that the average of the median PM2.5 concentrations is less than ¼ of what they are in houses that do not have a Mirt stove (19 g/m3 versus 85 g/m3, Kruskal-Wallis rank sum test prob >2 =0.01). As the distributions of all three PM2.5 concentration measures are skewed, medians may be considered better measures of central tendency than means. By this measure, the Mirt stove reduces PM2.5 concentration by 78%, which is at the higher end of ICS interventions and may offer health benefits to those in baking areas (Quansah et al., 2017). Table 2 Mean, Median and Maximum PM2.5 Concentrations in Cooking/Baking Areas of Households who did and did not Receive Mirt Stoves in 2013 and who Bake and Cook in their Main House Rather than in a Separate Cooking/Baking Area Mean Median Maximum Mirt, Generally with 595.85 (157.23) 19.12 (9.25) 45,579.62 (6828.47) Traditional Too (21) (21) (21) Traditional Only 1622.93 (421.22) 85.18 (48.77) 43,796.72 (6828.47) (32) (32) (32) Standard deviations and N in parentheses It is important to emphasize that there are no statistically significant differences in average PM2.5 concentrations when comparing our three concentration measures by cooking location only (main house versus separate cooking/baking area) rather than treatment vs. control assignment 12    (prob >2 0.33 to 0.62). This suggests that the effect on HAP is not due to cooking location. Instead, the difference appears to be due to Mirt conditioned on cooking/baking area location. We also note that the sample size of those households who cook in their main houses (n=53), is not atypical for ICS studies and is sufficiently powered to detect a moderate HAP effect (e.g. see Edwards et al., 2007; Quansah et al., 2017; Balakrishnan et al., 2002; Grabow et al, 2013). To detect the observed 64% reduction in HAP with 80% power and estimated COV=0.7, we would need approximately 20 households in each arm compared to the 21 treatment and 32 control observations we have in this study.12 We next estimate a difference-in-difference type specification to test whether cooking location is indeed important for effectiveness of the Mirt stove. The dependent variable is mean PM2.5 concentrations, and independent variables include dummies for whether a household received a Mirt stove, whether households cook/bake in the main house and these two variables interacted. The interaction term is our variable of interest and measures the additional change in mean PM2.5 for those households who use the Mirt in the main house relative to those who use Mirt elsewhere. In a second model, we add an indicator for roof type, the one variable out of 39 that is unbalanced in the full sample. Robust standard errors are clustered at the Kebele (peasant association) level. Table 3 reports the difference-in-difference regression results. Based on the coefficient of the interaction term, we find that cooking with Mirt in the main house is estimated to reduce average HAP concentrations by approximately 1,000g/m3, which is similar to the simple mean comparison estimate in Table 2. In both models, the coefficient estimate is significant at the 5% level. We therefore conclude that it is neither the Mirt stove nor baking in main houses alone that reduces HAP, but the two variables together that yield large effects.                                                         12 Of course, because by definition people live in their main houses, their choice of cooking location affects HAP exposures. This issue is addressed in a related paper by LaFave et al. (2018). We note that injera baking and cooking of other foods, such as stews, coffees and breads, can in principle take place in different locations. As discussed above, at baseline no household had two cooking areas. The 2013 reported locations of baking injera and cooking other foods are virtually perfectly correlated. 13    Table 3. OLS Regressions. Dependent Variable Mean HAP (g/m3) Model 1 Model 2 Mirt Stove (0/1) 166.405 259.812 (279.95) (264.279) Bakes in Main House (0/1) 472.787 69.300 (495.029) (463.304) Mirt Stove * Bakes in Main House (0/1) -1193.486** -1039.694** (529.668) (508.99) Metal Roof on Main House (0/1) -1104.681* (602.294) Constant 1150.140 1456.082 (927.024) N 201 201 R2 0.019 0.044 Robust standard errors clustered at kebele level. Statistically significant estimates in bold. ***=significant at 1% level, **=significant at 5% level, *=significant at 10% level. 4.4 Characterizing Those Who Cook/Bake in Main Living Areas We now characterize the approximately ¼ of sample households that bake and cook in their main living areas, where Mirt seems to significantly reduce PM2.5 concentrations. We estimate probit regressions to explain the choice to cook and bake in the main house using a combination of variables collected at the time of the initial stove disbursement in 2013 and during the HAP monitoring in 2016. Marginal effects from the probit models are reported in Table 4. Column 1 estimates cooking location as a function of cooking/baking area characteristics. We add basic socioeconomic variables in Column 2. Column 3 then adds additional welfare proxies, such as cooking/baking and eating frequency and distance to an all-weather road, and the full model in Column 4 includes wealth variables, such as agricultural land area, livestock, number of rooms in main houses and roof material. We emphasize that we are not alleging causality, but only correlation to try to characterize the types of households and household circumstances that are closely related to the decision to cook in main houses and therefore potentially identify the types of households who might benefit from the Mirt stove. Across specifications, we find that cooking/baking areas in main houses tend to have more area (p-value<0.01). This result is not surprising given that main living areas need to fulfill many functions along with cooking, such as sleeping, working and keeping warm, compared to separate areas where only cooking/baking takes place. As already noted, in the full sample, households have a mean of 2.6 rooms for an average of 9.3 adult equivalent household members, making it very unlikely that only cooking occurs in the main house cooking/baking area. Cooking/baking areas in 14    main houses have more windows and those that are circular in shape are less likely to be in main houses (p<0.05). Houses that are made of sticks and mud (>90%) in Model 4 are less likely to have a kitchen in the main house (p<0.05). Ethnic (Oromo, Amhara) and religious variables (e.g. Muslim) are correlated with cooking/baking in main houses. It is also of interest that in several models those who have higher levels of variables representing more assets report cooking/baking in separate areas rather than in their main living quarters. Those who have metal rather than thatched roofs (p<0.01), more agricultural land (p<0.10) and livestock (p<0.10), and who have used a bank in the past year (p<0.05) are more likely to cook in separate cooking/baking areas. Households that live in more remote areas – farther away from all-weather roads – are more likely to cook and bake in their main houses (p<0.01). Larger households with more members are also less likely to cook in main houses in two of three models (p<0.01). Households with more members have more labor and offer a number of other benefits, including old-age insurance, which could suggest they are more resilient than smaller households. With regard to cooking and eating behaviors, those who bake injera more frequently (a possible sign of well-being, because we also adjust for household size) are more likely to cook in separate cooking/baking areas (p<0.10). Finally, households who use less-preferred fuels like dung and agricultural wastes, which also tend to be smokier than fuelwood, are more likely to cook in their main living areas. This result especially seems to hold for agricultural wastes in Model 4 (p<0.01). Table 4. Marginal Effects of Probit Regressions of the Decision to Locate Cooking/Baking areas in Main Houses Data Model 1 Model 2 Model 3 Model 4 Year Kitchen floor area 2016 0.014*** 0.010*** 0.010*** 0.002** any shape (m2) (0.009) (0.011) (0.011) (0.022) Flat Ceiling in 2016  -0.12 -0.089 -0.108 -0.024 Kitchen (0/1) (0.585) (0.562) (0.576) (0.574) Number of 2016  0.011 0.117 0.146* 0.062*** windows in kitchen (0.258) (0.314) (0.332) (0.382) Kitchen is a circle 2016  -0.23** -0.188** -0.215*** -0.029*** rather than (0.338) (0.305) (0.308) (0.675) rectangular shape Can see light 2016  0.082 0.042 0.041 0.007 through the walls (0.220) (0.268) (0.307) (0.655) of baking area (0/1) 15      House is made of 2013 0.061 -0.056 -0.080 -0.439** sticks and mud (0.679) (0.799) (0.870) (0.867) Households size 2013  -0.037*** -0.039*** -0.003 (adult equivalent) (0.041) (0.045) (0.056) Respondent is 2013  -0.33*** -0.374*** -0.19*** Oromo ethnic (0.588) (0.605) (0.663) group (0/1) Respondent is 2013  -0.305*** -0.314*** -0.106*** Amhara ethnic (0.491) (0.614) (0.832) group (0/1) Respondent used a 2013  -0.057 -0.073 -0.046** bank in the past (0.306) (0.327) (0.701) year (0/1) Respondent is 2013  0.000009 0.035 0.041 member of equb (0.262) (0.331) (0.325) savings group (0/1) Age of respondent 2013  -0.001 -0.001 0.0002 (years) (0.009) (0.009) (0.020) Respondent is at 2013  -0.0.38 -0.058 -0.007 least literate (0/1) (0.200) (0.198) (0.363) Respondent is a 2013  0.042 0.090 0.003 man (0/1) (0.445) (0.461) (0.497) Ethiopian 2013  -0.129 -0.221** -0.006 Orthodox (0/1) (0.310) (0.421) (0.453) Muslim (0/1) 2013 0.263 0.237 0.406*** (0.612) (0.584) (0.631) Uses dung for fuel 2016 0.179* 0.152* 0.032 (0/1)13 (0.370) (0.355) (0.361) Uses agricultural 2016 0.014 0.046 0.066*** waste for fuel (0/1) (0.294) (0.316) (0.312) Two-way walking 2013 0.0009* 0.0004*** distance to all- (0.002) (0.003) weather road (mins) Average times per 2016  -0.124 -0.020 day those over 10 (0.357) (0.371) years eat14 Average times per 2016  0.018 -0.008 day cooking occurs (0.138) (0.195) Average times per 2016  -0.011* -0.004* week injera baking (0.024) (0.038) occurs                                                         Fuelwood is not included, as virtually 100% of households use fuelwood. 13 The frequencies at which younger people eat are available, but not reported, because 25% of the observations have no 14 data. 16    Agricultural land 2013  -0.018* area (ha.) (0.179) Livestock (tropical 2013  -0.006* livestock units) (0.070) Roof of main 2013  -0.855*** house is made of (0.917) metal (0/1) Number of rooms 2013  -0.012 in main house (0.199) Toilet in house 2013  0.018 (0/1) (0.480) N 199 198 197 192 Prob > Wald 2 0.000*** 0.000*** 0.000*** 0.000*** Pseudo R2 0.177 0.343 0.383 0.669 Log pseudo -93.25 -74.32 -69.487 -35.718 likelihood Marginal effects reported. Robust standard errors (clustered at kebele level) of coefficient rather than marginal effect estimates reported as recommended by Greene (2008, p. 487). Statistically significant estimates in bold. ***=significant at 1% level, **=significant at 5% level, *=significant at 10% level. 5. Discussion and Conclusions The use of biomass for cooking is expected to increase in Sub-Saharan Africa, at least until 2030. Improved biomass cookstoves have been promoted as an important intermediate technology to reduce fuelwood consumption (Gebreegziabher, 2018) and possibly indoor air pollution, while costing less than stove and fuel combinations involving electricity or LPG. We examine the Mirt improved biomass injera cookstove, because injera cooking is known to be very energy intensive, consuming about 50% of primary energy in Ethiopia. Looking at mean, median and maximum PM2.5 concentrations across households reveals PM2.5 concentrations that are at extremely high levels compared to WHO standards, but consistent with the literature. On average across all households, the Mirt stove reduces HAP, but by only a very small amount that is unlikely to affect human health. However, while most of our sample households cook in separate cooking/baking areas, a substantial minority cook inside their primary living areas. We examine this distinction and find that Mirt has significant HAP effects within the subsample that cooks in main houses, with central tendency reductions in PM2.5 concentrations in the 63% - 78% range. We emphasize that HAP concentrations are still many times the WHO guideline in these households, but households that cook inside their primary living areas who randomly received a Mirt stove have average mean and median concentrations that are substantially less than those who only have the traditional technology. 17    We run OLS regressions to explain PM2.5 concentrations in terms of the interactions of variables, adjusting for the one variable out of 39 that is unbalanced across treatment and control subsamples. We find that so long as Mirt is located in a household where people cook and bake in their main living areas, mean average HAP is about 1,000 µg/m3 lower than in houses without a Mirt stove; the Mirt stove therefore appears to have significant effects for those who combine their living and cooking/baking areas. As HAP reductions are well over 60%, Mirt may offer health benefits to those who cook in their main living areas. We also find that these households tend to be less wealthy, more remotely located and more likely to burn less-preferred, smokier biomass fuels like agricultural waste and animal dung than households that cook and bake in separate kitchen areas. Examining the HAP reducing properties of a technology that substitutes for such an energy- intensive process could yield insights into successes and failures around the world. However, extrapolating to very different contexts should be done with caution, because injera baking has a very different cooking profile than many other foods cooked around the world. While other dishes may be cooked multiple times per day, injera is typically cooked 2-3 times per week. Though we have tried to show that our treatment and control households are balanced based on a variety of observables, we acknowledge that our cross-sectional study design does not allow us to completely rule out that exogenous factors affecting HAP do not confound our estimates. Our study design does not allow us to identify the reasons for the large percentage reductions in HAP due to Mirt use by households who cook/bake in their main living areas. One explanation is the decreased emissions made possible by greater stove energy efficiency, but it may also be that households cooking in their main living area who adopt Mirt subsequently move their traditional three-stone cooking outside their main living areas. We did not measure footprint of houses (only cooking areas), but poorer households, who we find tend to cook/bake in main living areas, generally have smaller houses.15 Adopting Mirt may therefore lead to easily-moved three- stone stoves being shifted out of main living areas into new, informal, covered cooking areas that did not exist in 2013. Such shifts would reduce indoor air pollution compared with the control households who still use only less-efficient three-stone stoves inside their main homes. If this explanation is correct, then of course the HAP becomes outdoor ambient air pollution, with all that implies. Even so, because the ambient air pollution is less concentrated and more subject to dispersion outdoors than inside, adopting Mirt may improve respiratory health.                                                         As shown in Table 4, number of rooms does not affect the probability of cooking/baking in the main living area. 15 Adult equivalent household size per room also has no effect. 18    Mirt is considered a prestige product by many households in rural Ethiopia and we know from related research that it is overwhelmingly positively viewed by users (Gebregziabher, 2018). Faced with a choice whether to have Mirt or the traditional three-stone tripod stove in their main living areas, households with limited means and little space may choose Mirt. It is therefore possible that Mirt reduces HAP and perhaps improves respiratory health by forcing traditional, inefficient, but highly mobile traditional cooking technologies out of main living quarters. This possibility echoes key conclusions of Langbein et al. (2017) that researchers and policy makers must think broadly about what constitutes kitchens in low-income countries and remember that traditional cooking technologies are often extremely mobile. 19    Bibliography Achilleosa, S. M-A Kioumourtzoglou, C-D Wuc, J. D Schwartz, P. Koutrakisa, S. I. Papatheodorou. 2017. “Acute Effects of Fine Particulate Matter Constituents on Mortality: A Systematic Review and Meta-Regression Analysis,” Environment International 109: 89–100 Anderson, J. O. J.G. Thundiyil and A. Stolbach. 2012. ‘Clearing the Air: A Review of the Effects of Particulate Matter Air Pollution on Human Health’, Journal of Medical Toxicology 8:166–175. Balakrishnan, K, J. Parikh, S. Sankar, R. Padmavathi, K. Srividya, V. Venugopal, S. Prasad and V. L. Pandey. 2002. “Daily Average Exposures to Respirable Particulate Matter from Combustion of Biomass Fuels in Rural Households of Southern India Environmental Health Perspectives 110: 1069-1075. Berkeley Air Monitoring Group. 2017. Technical Note: Adjusting PATS+ particulate matter concentrations using an in-field co-located gravimetric calibration. May 2017. Mimeo. Beyene, A, R. Bluffstone and A. Mekonnen. 2015a. “Community Forests, Carbon Sequestration and REDD+: Evidence from Ethiopia” Environment and Development Economics 21: 249-272. Beyene, A, R. Bluffstone, Z. Gebreegziabher, P. Martinsson, A. Mekonnen, F. Vieider. 2015b. “The Improved Biomass Stove Saves Wood, But How Often Do People Use It? Evidence from a Randomized Treatment Trial in Ethiopia,” World Bank Policy Research Working Paper 7297. Available at http://documents.worldbank.org/curated/en/2015/06/24603652/improved-biomass- stove-saves-wood-often-people-use-evidence-randomized-treatment-trial-ethiopia Bizzari, M. 2010. Safe Access to Firewood and Alternative Energy in Ethiopia: An Appraisal Report. Prepared for the World Food Program. http://www.genderconsult.org/uploads/publications/doc/SAFE_Ethiopia_Appraisal_Rep ort_Final_Draft_2.pdf Accessed May 12. Bluffstone, R. Z. Gebreegziabher, A. Beyene and A. Mekonnen. 2017. “Learning, Experience, Fuelwood and Time Savings from Improved Biomass Cookstoves: Evidence from Randomized Controlled Cooking Tests in Ethiopia.” Paper presented at the Second Meeting of the Sustainable Energy Transitions Initiative, Duke University. Chen, C. S. Zeger, P. Breysse, J. Katz, W. Checkley, F. C. Curriero, J. M. Tielsch. 2016. “Estimating Indoor PM2.5 and COConcentrations in Households in SouthernNepal: The Nepal Cookstove InterventionTrials.” PLOS One July 2016. 0157984: 1-17. Chow, J.C., Engelbrecht, J.P., Freeman, N.C.G., Hashim, J.H., Jantunen, M., Michaud, J.P., Saenz de Tejada, S., Watsom, J.G., Wei, F., Wilson, W.E., 2002. Chapter one: exposure measurements. Chemosphere 49, 873-901. Dresen, E, B. DeVries, M. Herold, L. Verchot and R. Müller. 2014. “Fuelwood Savings and Carbon Emission Reductions by the Use of Improved Cooking Stoves in an Afromontane Forest, Ethiopia.” Land3: 1137-1157. Edwards, R, A. Hubbard, A. Khalakdina, D. Pennise and K. Smith. 2007. “Design Considerations for Field Studies of Changes in Indoor Air Pollution due to Improved Stoves.” Energy for Sustainable Development, XI (2): 71-81. FDRE, 2011. Ethiopia’s Climate-Resilient Green Economy: Green Economy Strategy. 20    FDRE 2014. Ethiopia’s Climate-Resilient Green Economy Climate Resilience Strategy: Water and Energy, Draft, Ministry of Water, Irrigation and Energy, Federal Democratic Republic of Ethiopia (FDRE). Gebreegziabher, Z. A. Beyene, R. Bluffstone, P. Martinsson, A. Mekonnen, M. Toman. 2018. “Fuel Savings, Cooking Time and User Satisfaction with Improved Biomass Cookstoves: Evidence from Controlled Cooking Tests in Ethiopia.” Resource and Energy Economics 52: 173-185. GIZ-ECO (Energy Coordination Office). 2011. Mirt Stove Ethiopia. GIZ-ECO Ethiopia, Addis Ababa, November 2011. Greene, W. 2008. “Discrete Choice Modeling,” in The Handbook of Econometrics: Vol. 2, Applied Econometrics, Part 4.2., ed. T. Mills and K. Patterson, Palgrave, London. Grabow, K, D. Still and S. Bentson. 2013. “Test Kitchen Studies of Indoor Air Pollution from Biomass Cookstoves” Energy for Sustainable Development 17: 458-462. IEA 2017. World Energy Access Outlook 2017: From Poverty to Prosperity. World Energy Outlook Special Report. International Energy Agency. Paris: IEA. IEA 2012. World Energy Outlook 2012. International Energy Agency. Paris: IEA. Jeuland, M. A, S. Pattanayak, and R. Bluffstone. 2015. “The Economics of Household Indoor Air Pollution.” Annual Review of Environmental and Resource Economics. 7: 81-108. Jeuland, M. A. and S. Pattanayak. 2012. “Benefits and Costs of Improved Cookstoves: Assessing the Implications of Variability in Health, Forest and Climate Impacts.” PloS one, 7, e30338. Karagulian, F., C. A. Belis, C. F. C. Dora, A. M. Prüss-Ustün, S. Bonjour, H. Adair-Rohani, M. Amann. 2015. ‘Contributions to Cities' Ambient Particulate Matter (PM): A systematic review of local source contributions at global level’, Atmospheric Environment 120: 475-483. LaFave, D, A. Beyene, R. Bluffstone, S. Dissanayake, Z. Gebreegziabher, A. Mekonnen, M. Toman. 2018. “Impacts of Improved Biomass Cookstoves on Child and Adult Health: Experimental Evidence from Rural Ethiopia” Paper prepared for the World Bank. Contract 7178322. Langbein, J. JPeters and C. Vance. 2017. “Outdoor Cooking Prevalence in Developing Countries and its Implication for Clean Cooking Policies. Environmental Research Letters 12 115008: 1-11. Lim, S. S., T. Vos, A. D. Flaxman, G. Danaei et al. 2012. "A Comparative Risk Assessment of Burden of Disease and Injury Attributable to 67 Risk Factors and Risk Factor Clusters in 21 Regions, 1990–2010: a Systematic Analysis for the Global Burden of Disease Study 2010." The Lancet 380 (9859):2224-2260. MacCarty, N, D. Ogle, D. Still, T. Bond, C. Roden. 2008. “A Laboratory Comparison of the Global Warming Impact of Five Major Types of Biomass Cooking Stoves,” Energy for Sustainable Development 12: 5-14. MacCracken, J. P. and K. R. Smith. 1998. “Emissions and Efficiency of Improved Woodburning Cookstoves in Highland Guatemala.” Environment International, Vol. 24, No. 7, pp. 739-747. Martin II, W. J., R. I. Glass, J. M. Balbus, F. S. Collins. 2011. “A Major Environmental Cause of Death.” Science 334(6053): 180-181. Masera. P. R. Bailis, R. Drigo, A. Ghilardi and I. Ruiz-Mercado. 2015. “Environmental Burden of Traditional Bioenergy Use” Annual Review of Environment and Resources 40: 121-150. 21    Megen Power Ltd. 2008. Final Report: Impact Assessment of Mirt Improved Biomass Injera Stoves [sic] Commercialization in Tigray, Amhara and Oromiya National Regional States, Submitted to the MoARD/GTZ SUN Energy Programme, Addis Ababa. Mobarak, A. M., P. Dwivedi, R. Bailis, L. Hildemann and G. Miller. 2012. “Low Demand for Nontraditional Cookstove Technologies.” Proceedings of the National Academy of Sciences109: 10815-10820. Ojo, K. D., S. I. Soneja, C. G. Scrafford, S. K. Khatry,S. C. LeClerq, W. Checkley, J. Katz, P. N. Breysse andJ. M. Tielsch. 2015. “Indoor Particulate Matter Concentration, Water Boiling Time, and Fuel Use of Selected Alternative Cookstoves in a Home-Like Setting in Rural Nepal.” International Journal of Environmental Research and Public Health, 12: 7558-7581 Pope, D. N. Bruce, M. Dherani, K. Jagoe, E. Rehfuess. 2017. “Real-Life Effectiveness of ‘Improved’ Stoves and Clean Fuels in ReducingPM2.5 and CO: Systematic Review and Meta-Analysis.” Environment International 101: 7–18 Quansah, R., S.S emple, C. A. Ochieng, S. Juvekar, F. A. Armah, I. Luginaah, J. Emina. 2017. “Effectiveness of Interventions to Reduce Household Air Pollution and/or Improve Health in Homes Using Solid Fuel in Low-and-Middle Income Countries: A Systematic Review and Meta-Analysis,” Environment International 103:73–90. Rehfuess, E., S. Mehta, A. Prüss-Üstün. 2006. “Assessing Household Solid Fuel Use: Multiple Implications for the Millennium Development Goals.” Environmental Health Perspectives 114. Salvi, S. 2015. “Comment: The Silent Epidemic of COPD in Africa.” The Lancet. January, 2015: e6- e7. Smith, K.R., S. Mehta, M. Maeusezahl-Feu. 2004. “Indoor Air Pollution from Household Use of Solid Fuels.” In: M. Ezzati, A. D. Lopez, A. Rodgers, C. J. L. Murray (Eds.). Global and Regional Burden of Disease Attributable to Selected Major Risk Factors. World Health Organization, Geneva, pp. 1435 -1494. Smith, K.R., K. Dutta, C. Chengappa, P. P. S. Gusain, O. Masera, V. Berrueta, R. Edwards, R. Bailis, K. N. Shields. 2007. “Monitoring and Evaluation of Improved Biomass Cookstove Programs for Indoor Air Quality and Stove Performance: Conclusions from the Household Energy and Health Project.” Energy for Sustainable Development 11: 5 -18. Smith, K. R., H. Frumkin K. Balakrishnan, C. D. Butler, Z. A. Chafe, I. Fairlie, P. Kinnery, T. Kjellstrom, D. L. Mauzerall and T. E. McKone. 2013. “Energy and Human Health.” Annual Review of Public Health, 34, 159-188. Tesfay, A. H, M. B. Kahsay, O. J. Nydal. 2014. “Solar Powered Heat Storage for Injera Baking in Ethiopia.” Energy Procedia 57: 1603-1612. WHO (World Health Organization). 2006. Air Quality Guidelines: “Global Update 2005: Particulate matter, Ozone, Nitrogen Dioxide, and Sulfur Dioxide” Copenhagen, Denmark: World Health Organization. WHO. 2015. “WHO Guidelines for Indoor Air quality: Household Fuel Combustion.” Geneva: World Health Organization. 22    WHO 2013. “WHO Methods and Data Sources for Global Burden of Disease Estimates 2000-2011. Global Health Estimates Technical Paper WHO/HIS/HSI/GHE. Geneva, Switzerland: World Health Organization. WHO 2014. “Burden of Disease from Household Air Pollution for 2012.” Available at www.who.int/phe. (Accessed 20 October 2014): World Health Organization. WHO. Undated. “Global Health Observatory data repository: Population Using Solid Fuels (Estimates).”http://apps.who.int/gho/data/view.main.1681?lang=en. Accessed June 5, 2018. WHO-ROE (Regional Office for Europe). 2013 Health Effects of Particulate Matter: Policy Implications for Countries in Eastern Europe, Caucasus and Central Asia, Copenhagen 23    Appendix Field Protocols for taking HAP Measurements 24    25    26    27    28    29    30    31    Appendix Table A1: Balance Tests across Treated and Control Households Variable N Mean of Mean of Two-Tailed Kruskal-Wallis Rank- Treated Control t-test p value Sum Test with ties p (unequal 2) value 2013 Data Livestock in TLU 201 5.164 4.689 0.311 0.192 Households size (adult 204 9.340 9.310 0.945 0.752 equivalent) Number of children under 15 187 3.054 2.798 0.236 0.222 years Respondent smoker in 2013 (0/1) 203 0.061 0.076 0.661 0.661 Ethiopian Orthodox religion 204 0.576 0.514 0.361 0.380 (0/1) Respondent is at least literate 204 0.646 0.619 0.687 0.686 (0/1) Muslim religion (0/1) 204 0.286 0.212 0.226 0.226 Respondent is a man (0/1) 204 0.899 0.943 0.250 0.245 Respondent is married (0/1) 204 0.942 0.039 0.917 0.966 Age of respondent (years) 204 42.94 42.95 0.964 0.930 House has metal roof (0/1) 204 0.828 0.619 0.001*** 0.001*** Number of rooms in 204 2.72 2.56 0.217 0.178 respondent’s house Respondent has toilet inside 204 0.909 0.923 0.706 0.705 house (0/1) Total household land (ha.) 204 1.87 1.73 0.517 0.614 If farmer, food produced is 201 0.531 0.486 0.521 0.577 sufficient for year (0/1) Respondent used a bank in the 204 0.242 0.171 0.211 0.214 past year (0/1) Respondent is a member of an 203 0.112 0.086 0.530 0.527 equb mutual savings group (0/1) Altitude of house above mean sea 203 2205.14 2220.99 0.772 0.908 level (m) Two-way walking distance to all- 203 46.91 60.89 0.180 0.198 weather road in minutes House is made of sticks and mud 203 0.970 0.962 0.761 0.760 Respondent drinks alcohol (0/1) 204 0.545 0.488 0.164 0.164 Respondent is Oromo ethnic 204 0.343 0.381 0.580 0.578 group (0/1) Respondent is Amhara ethnic 204 0.232 0.229 0.950 0.950 group (0/1) 2016 Data Mean temperature (degrees C) 204 20.847 20.831 0.972 0.978 Mean % humidity 204 48.170 48.696 0.721 0.679 Kitchen is a circle rather than 204 0.556 0.657 0.139 0.138 32    rectangular shape Kitchen floor area any shape (m2) 198 28.19 20.21 0.386 0.993 Number of windows in kitchen 201 0.203 0.194 0.897 0.623 Kitchen ceiling is flat rather than 203 0.051 0.086 0.328 0.331 peaked (0/1) Can see light through the walls of 202 0.778 0.825 0.401 0.399 baking area (0/1) There is a three-stone tripod in 204 0.96 0.990 0.164 0.155 kitchen (0/1) Bakes injera in main house (0/1) 201 0.212 0.305 0.132 0.133 Other household members 204 0.071 0.067 0.894 0.894 smoked in 2016 (0/1) Number of meals eaten per day 204 2.94 2.91 0.603 0.608 by household members >= 10 years Number of times cooked per day 204 2.404 2.438 0.782 0.785 Number of times injera baked per 204 5.242 5.514 0.746 0.374 week Uses fuelwood for baking injera 204 0.980 0.990 0.533 0.527 Uses animal dung for baking 204 0.444 0.438 0.928 0.928 injera Uses agricultural waste for baking 204 0.192 0.257 0.266 0.266 injera Appendix Table A2: Balance Tests across Households Subject to HAP monitoring and those not Monitored Baseline (2013) Data Variable N Mean of Mean of Two-Tailed Kruskal-Wallis Rank- Treated Control t-test p Sum Test with ties p value value (unequal  ) 2 Livestock in TLU 475 4.921 5.093 0.606 0.707 Households size (adult 481 9.33 9.177 0.628 0.456 equivalent) Number of children under 15 434 2.925 3.085 0.277 0.345 years Respondent smoker in 2013 481 0.686 0.578 0.631 0.627 (0/1) Ethiopian Orthodox religion 481 0.544 0.502 0.360 0.359 (0/1) Muslim religion (0/1) 481 0.250 0.318 0.102* 0.106 Respondent is at least literate 481 0.632 0.599 0.462 0.462 (0/1) Respondent is a man (0/1) 480 0.922 0.880 0.131 0.141 Respondent is married (0/1) 481 0.943 0.892 0.048** 0.058* Age of respondent (years) 480 42.90 41.78 0.353 0.374 33    House has metal roof (0/1) 481 0.721 0.664 0.185 0.188 Number of rooms in 481 2.64 2.59 0.526 0.543 respondent’s house Respondent has toilet inside 481 0.917 0.845 0.014*** 0.019** house (0/1) Total household land (ha.) 481 1.800 1.820 0.896 0.987 If farmer, food produced is 477 0.507 0.442 0.159 0.158 sufficient for year (0/1) Respondent used a bank in the 481 0.206 0.238 0.398 0.400 past year (0/1) Respondent is a member of an 477 0.099 0.142 0.142 0.151 equbmutual savings group (0/1) Altitude of house above mean 480 2213.26 2201.24 0.731 0.585 sea level (m) Two-way walking distance to all- 480 54.07 69.01 0.084 0.895 weather road in minutes House is made of sticks and mud 481 0.966 0.960 0.756 0.759 Respondent drinks alcohol (0/1) 204 0.495 0.512 0.705 0.704 Respondent is Oromo ethnic 204 0.363 0.404 0.354 0.355 group (0/1) Respondent is Amhara ethnic 204 0.230 0.256 0.513 0.514 group (0/1) 34