1. Introduction
Despite considerable progress, as we move through the third decade of the 21st century, almost two in five people on the planet still cook with highly polluting, mainly solid fuels (e.g., biomass, coal), burned in their main homes or nearby buildings (Jeuland and Pattanayak, Reference Jeuland and Pattanayak2012; Jeuland et al., Reference Jeuland, Pattanayak and Bluffstone2015b; Stoner et al., Reference Stoner, Lewis, Martínez, Gumy, Economou and Adair-Rohani2021). This reliance, particularly on biomass fuels like fuelwood, agricultural waste and animal dung, is primarily concentrated in low-income developing countries (Jeuland et al., Reference Jeuland, Bhojvaid, Kar, Patange, Pattanayak, Ramanathan, Rehman, Tan Soo and Ramanathan2015a; Stoner et al., Reference Stoner, Lewis, Martínez, Gumy, Economou and Adair-Rohani2021), contributing to indoor and outdoor air pollution (Smith et al., Reference Smith, Frumkin, Balakrishnan, Butler, Chafe, Fairlie, Kinney, Kjellstrom, Mauzerall, McKone, McMichael and Schneider2013; WHO, 2022), forest degradation (Gebreegziabher and van Kooten, Reference Gebreegziabher and van Kooten2013) and climate change (Bailis et al., Reference Bailis, Drigo, Ghilardi and Masera2015).
Though across the world the percentage of people depending on biomass fuels has declined dramatically in the past decades and will continue to decline, the absolute number of people who depend on biomass fuels is expected to fall only slightly through 2030, to 2.4–2.7 billion people (IEA, 2020; Stoner et al., Reference Stoner, Lewis, Martínez, Gumy, Economou and Adair-Rohani2021). Biomass cooking fuel use is also expected to remain very significant in Ethiopia, which is the focus of this paper, because over 1 billion people in sub-Saharan Africa are projected to use solid fuels in 2025 (Stoner et al., Reference Stoner, Lewis, Martínez, Gumy, Economou and Adair-Rohani2021). In 2021, Ethiopia depended on biofuels and waste for about 88 per cent of its energy demand (IEA, 2024), most of which is used for cooking, though use of commercial fuels is expected to increase in the coming decades (Mondal et al., Reference Mondal, Bryan, Ringler, Mekonnen and Rosegrant2018).
Burning biomass cooking fuels causes indoor and outdoor air pollution (e.g., Lim et al., Reference Lim, Vos, Flaxman, Danaei, Shibuya, Adair-Rohani, Almazroa, Amann, Anderson and Andrews2013; Smith et al., Reference Smith, Frumkin, Balakrishnan, Butler, Chafe, Fairlie, Kinney, Kjellstrom, Mauzerall, McKone, McMichael and Schneider2013), forest degradation (e.g., Arnold et al., Reference Arnold, Köhlin and Persson2006; Boucher et al., Reference Boucher, Elias, Lininger, May-Tobin, Roquemore and Saxon2011; Gebreegziabher and van Kooten, Reference Gebreegziabher and van Kooten2013), which can create serious labor demands due to the need to collect fuelwood (e.g., see Cooke et al., Reference Cooke, Kohlin and Hyde2008) and climate change (e.g., Bailis et al., Reference Bailis, Drigo, Ghilardi and Masera2015). Traditional biomass cookstoves, such as the three-stone tripod typically used in Ethiopia, are only 5–15 per cent efficient (Mobarak et al., Reference Mobarak, Dwivedi, Bailis, Hildemann and Miller2012), implying they use more biomass than needed to cook meals and emit high levels of air pollution. To mitigate some of the problems associated with biomass fuels, improved biomass cookstoves (ICS) have been introduced around the world. These stoves seek to provide efficiencies by cooking meals with less fuel, resulting in less fuelwood consumption (Mondal et al., Reference Mondal, Bryan, Ringler, Mekonnen and Rosegrant2018) and potentially offering improved performance in other ways, such as reduced cooking time and improved indoor air quality compared with traditional technologies.Footnote 2 These technologies are typically not sophisticated, are often cheap (e.g., Hanna et al., Reference Hanna, Duflo and Greenstone2016), and in some cases require only minor changes in cooking habits (Jeuland and Pattanayak, Reference Jeuland and Pattanayak2012).
Many ICS were introduced in the 1980s, but often it was found that cooks did not regularly use those stoves (e.g., Gil, Reference Gil1987; Barnes et al., Reference Barnes, Openshaw, Smith and van der Plas1993), suggesting that benefits observed in labs either did not materialize in the field or were outweighed by additional costs to users. It soon came to be recognized that ICS adoption really has two parts. The first is willingness to try stoves, which implies distribution to households. The second and more important aspect of adoption is regular use over an extended period, which has come to be seen as the main litmus test of true adoption (Johnson et al., Reference Johnson, Edwards, Ghilardi, Berrueta, Gillen, Frenk and Masera2009; Bensch and Peters, Reference Bensch and Peters2015; Hanna et al., Reference Hanna, Duflo and Greenstone2016). Because of the critical role of longer-run regular use in determining whether potential benefits of ICS technologies are generated in households, in this paper we focus on regular use over an extended time period.
We examine adoption of the Mirt ICS, which is a low-cost (approximately $12), chimneyless manufactured stove made of six pieces of concrete joined together with mud.Footnote 3 It has been promoted in Ethiopia since 1998 and is used primarily to cook injera. Injera is a type of pancake that is eaten in most parts of Ethiopia, and it requires hot, even energy delivery to be properly baked. The average household uses about 20 kg of fuelwood per week to make injera, and injera baking has been estimated to use about half of the primary energy of the country (Tesfay et al., Reference Tesfay, Kahsay and Nydal2014).
Mirt has been estimated to use 50 per cent less wood in laboratory tests (GIZ, 2011), 40 to 50 per cent relying on self-reports in surveys (Megen Power Ltd., 2008; Dresen et al., Reference Dresen, DeVries, Herold, Verchot and Müller2014) and 20 to 30 per cent in field-based controlled cooking tests (Gebreegziabher et al., Reference Gebreegziabher, Beyene, Toman, Bluffstone, Dissanayake and Martinsson2018). Use of ICS in rural Ethiopia is limited (Mondal et al., Reference Mondal, Bryan, Ringler, Mekonnen and Rosegrant2018), with most households relying exclusively on three-stone tripods that waste approximately 90 per cent of the energy input (Alem et al., Reference Alem, Hassan and Kohlin2014).
Wassie and Adaramola (Reference Wassie and Adaramola2021) estimate a significant reduction in fuelwood consumption and the potential emissions reductions associated with use of improved cookstoves in Ethiopia. Gebreegziabher et al. (Reference Gebreegziabher, van Kooten and van Soest2017) also find a significant reduction in fuelwood and dung consumption from use of improved cookstoves. They also note the associated positive externalities in the form of reduced pressure on the forest as well as increased soil productivity from reduced dung use as fuel.
We extend our previous field experiment-based work on this topic, which examined regular use of Mirt over a period of about a year (Bluffstone et al., Reference Bluffstone, Beyene, Gebreegziabher, Martinsson, Mekonnen and Vieider2021), in light of three monetary treatments, which were randomly applied at baseline. Injera is not cooked every day, but regular baking would occur at least twice per week in areas such as our study regions, which lack refrigeration. Regular usage of the Mirt injera stove therefore means using it at least twice per week (Tesfay et al., Reference Tesfay, Kahsay and Nydal2014).
The three monetary treatments applied at baseline are providing the stoves: (1) free of charge; (2) at a monetary cost, and (3) with a one-off incentive to regularly use the Mirt ICS during the first approximately six-week monitoring period after distribution. In all cases, randomization was at the site level. As will be discussed in section 2 of this paper, monetary incentives have been very important as part of policy packages to promote new technologies in the Global South. For example, Hassen and Kohlin (Reference Hassen and Kohlin2017) examined the role of sunk cost and screening effects on Mirt stoves in rural Ethiopia.
In our previous paper (Bluffstone et al., Reference Bluffstone, Beyene, Gebreegziabher, Martinsson, Mekonnen and Vieider2021), which examined behaviors that occurred within the first year, we found that households in general regularly used the Mirt stove. We also found that average usage – and average regular usage defined as use at least twice per week – increased over time; and free distribution promoted both sustained high levels of regular use and increases in average usage over time. Paying for the Mirt ICS was found to promote increased usage mainly because average use in the beginning was very low compared to the other two monetary treatments. We therefore concluded after one year of evidence that the Mirt stove was on average regularly used and free distribution is the preferred treatment for maximizing uptake (Bluffstone et al., Reference Bluffstone, Beyene, Gebreegziabher, Martinsson, Mekonnen and Vieider2021).
In this paper, we test whether these results hold up in the longer run, meaning approximately three-and-a-half years after initial distribution of the Mirt stove. This period is about 70 per cent of the estimated lifespan of Mirt and allows us to examine common ICS program outcomes that were not observed during the first year. These outcomes include dis-adoption and abandonment due to breakage or poor fit with family cooking styles, which were not observed during our shorter-run analysis. We also evaluate if use behavior has changed during the approximately 2.5 years since the Mirt stoves were last monitored, and if there are any lingering effects on stove use from the monetary treatments applied in 2013.
A theory of change we consider in this paper is that the stove is expected to be more useful and the benefits for households would be larger if it is used over an extended period. The benefits of the improved stove for households include time and money saved due to reduced use of fuelwood, and reduced exposure to indoor smoke. Thus, extended use of the stove would have significant private benefits, especially because the cooking fuel is so important in terms of total energy consumption in Ethiopia. Moreover, for the benefits to be realized it may be necessary to incentivize use and it is important to examine the extent to which the stove is being used regularly. There are also benefits from reduced outdoor air pollution locally as well as global benefits from reduced GHG emissions. Thus, ascertaining how regularly the stove is used in the long run and whether incentives matter has important implications for policy.
We find that about 63 per cent of households still have their Mirt stoves in place after about 3.5 years, and over 80 per cent of those who no longer have them report that they were abandoned due to breakage. In our previous shorter-run analysis, no stoves were abandoned. In the longer run we find that, among those who retained their functioning stoves, Mirt stove use is high and on average in line with regular use (roughly twice per week). If anything, average usage frequency increased over time. We therefore conclude that in the longer run the Mirt stove continues on average to be regularly used.
We also find that usage intensity 3.5 years after introduction is not driven by the monetary incentives given at the time of the introduction. In contrast to our previous shorter-run analysis, while free distribution does no worse at promoting persistent longer-run regular usage, it is generally not more effective than the other treatments.Footnote 4 Persistent regular usage of the stove that we find in this study, combined with the reduction in fuelwood consumption and carbon emissions reported in Mekonnen et al. (Reference Mekonnen, Beyene, Bluffstone, Gebreegziabher, Martinsson, Toman and Vieider2022), is likely to imply that ICS policies should be continued or even expanded in Ethiopia because of the sustained benefits to households, as well as society at large through climate change mitigation benefits.
Our study contributes to the limited literature on adoption of improved stoves in the longer run, with Bensch and Peters (Reference Bensch and Peters2015) and Hanna et al. (Reference Hanna, Duflo and Greenstone2016) being the major studies. Bensch and Peters (Reference Bensch and Peters2015) randomly distributed an improved stove with expected lifespan between one year and three-and-a-half years free of charge and analyzed behavior more than one year after distribution. Hanna et al. (Reference Hanna, Duflo and Greenstone2016), in contrast, found that the usage rate and performance of the stove they studied declined over the four years examined, mainly due to limited or no maintenance investments by households. Section 2 discusses the key literature that is germane to our research topic. Section 3 discusses the methods and section 4 presents results. Section 5 discusses the findings and section 6 concludes.
2. Literature review
In this section, we briefly review previous literature on long-run adoption of new technologies, with a special emphasis on the monetary terms under which those technologies were provided. Of the two parts of ICS adoption – distribution and regular use – the second part is more ambitious and important. Financial incentives such as subsidies and rewards are often used to promote distribution, but sometimes concerns are raised that promoting distribution by offering stoves free or at low cost may waste resources and even reduce usage. Other instruments to induce behavioral changes and enhance adoption in diverse spheres of life, including energy conservation, smoking cessation, and adherence to medication, include social comparisons (e.g., Schultz et al., Reference Schultz, Nolan, Cialdini, Goldstein and Griskevicius2007; Nolan et al., Reference Nolan, Schultz, Cialdini, Goldstein and Griskevicius2008), and provision of nudges (Allcott and Rogers, Reference Allcott and Rogers2012, Reference Allcott and Rogers2014). Abrahamse et al. (Reference Abrahamse, Steg, Vlek and Rothengatter2005), Delmas et al. (Reference Delmas, Fischlein and Asensio2013), Karlin et al. (Reference Karlin, Ford and Zinger2015), and Andor and Fels (Reference Andor and Fels2018) provide useful analyses and reviews of the literature on behavioral interventions centered around energy conservation. For example, Abrahamse et al. (Reference Abrahamse, Steg, Vlek and Rothengatter2005) conclude that information does not necessarily result in behavioral changes or energy savings. They also find that payments for adoption of energy efficiency technologies encourage short-run adoption, but effects dissipate in the long-run.
With regard to monetary treatments, free or low-cost distribution of technologies may promote uptake (e.g., Cohen and Dupas, Reference Cohen and Dupas2010; Bensch et al., Reference Bensch, Grimm and Peters2015), but could also potentially mean that technologies go to those who will not use them (e.g., Ashraf et al., Reference Ashraf, Berry and Shapiro2010; Chassang et al., Reference Chassang, Padro and Snowberg2012). Payments could also potentially generate psychological sunk costs that promote use, but little evidence appears to exist of such effects (Ashraf et al., Reference Ashraf, Berry and Shapiro2010; Cohen and Dupas, Reference Cohen and Dupas2010).
Some of the most significant work on long-run adoption has been on insecticide-treated bed nets in Sub-Saharan Africa. The WHO analyzed free distribution of bed nets versus charging nominal prices. They find that free distribution is most effective (The Economist, 2008). Dupas (Reference Dupas2014) also shows that a one-time subsidy can boost the purchase of mosquito-repellent bed nets and those who received the bed nets under concessionary terms do not anchor their valuations on those prices.Footnote 5 Cohen and Dupas (Reference Cohen and Dupas2010) do not find a significant difference in the use of bed nets between households who paid, compared with those who received their bed nets for free. In a field experiment in Zambia, Ashraf et al. (Reference Ashraf, Berry and Shapiro2010) investigated the intensity with which chlorine to treat drinking water was used. They find that higher prices of chlorine increase the average probability that people use it by screening out low-value users. They find little evidence, however, that high sunk costs promote use.
A second potential monetary treatment is payment for performance. Cahill and Perera (Reference Cahill and Perera2011) review the effectiveness of incentives for smoking cessation. They argue that, though there seems to be some evidence that incentives work in the short run, generally the effects dissipate and there is no clear evidence that initial behavioral changes translate into routine practice. Kane et al. (Reference Kane, Johnson, Town and Butler2004) categorize behavioral changes into ‘simple’ (single actions at a point in time), such as taking a medication, and ‘complex’ (requiring effort over a sustained period of time) such as smoking cessation and weight loss to reduce obesity.
According to Giuffrida and Torgerson (Reference Giuffrida and Torgerson1997) and Aveyard and Bauld (Reference Aveyard and Bauld2011), incentives are effective for simple, but not for complex, behavior change. They argue that complex behavior change requires multiple interventions. Moreover, the effectiveness of interventions for complex behavioral change often declines when the incentive is withdrawn (Paul-Ebhohimhen and Avenell, Reference Paul-Ebhohimhen and Avenell2007). Adoption of a new cooking technology can potentially require significant adjustments by cooks, which could qualify it as a complex behavioral change. The literature therefore suggests that a one-shot monetary incentive which, along with free distribution and requiring payment is one of the treatments analyzed in this paper, would be no more effective at incentivizing longer-run adoption than other treatments.
3. Field experiment
3.1 Design
The field experiment involved distributing the Mirt stove under randomly-assigned monetary treatments to a sample of rural Ethiopian households in 2013. In our study area and throughout much of Ethiopia, households with no refrigeration typically bake injera at least twice per week (Tesfay et al., Reference Tesfay, Kahsay and Nydal2014). We therefore use baking injera at least twice per week as our definition of regular usage.
We identified cooking events by measuring the surface temperature of the Mirt stoves using electronic Stove Use Monitors (SUMs).Footnote 6 The SUMs we used are approximately the size of a watch battery. We set the SUM device to record surface temperature every ten minutes and the memory can record for approximately 60 days and tolerate temperatures up to 120° Celsius (C). Temperature, time and date data were then downloaded after the monitoring period was completed. Respondents were informed that the purpose of the SUM device was to record surface temperature of the stove and that enumerators would come back periodically to download the data and reinstall the SUM devices.
We have three randomly-assigned monetary treatments: (1) users received the stove for free; (2) users paid 25 Birr for the Mirt stove (about 13 per cent of the market price), and (3) users paid nothing for their stoves and received a 50 Birr,Footnote 7 one-time incentive payment if the SUMs that were affixed to their stoves indicated that Mirt stoves were used at least twice per week (i.e., were regularly used) during the first monitoring period.
Using signal processing and analysis (O'Haver, Reference O'Haver1997), we measured frequency of stove use for a monitoring period by simply counting the number of times the stove surface temperature exceeded a critical value. The average stove use in a period is calculated as the number of cooking events divided by total days temperature was measured. Because injera is not baked daily, we express this value per week and define regular use of the stove as average use of at least twice per week.
Based on readings from the SUMs, we identify a cooking event as having taken place if the recorded temperature exceeded 40°C. We chose 40°C as our cutoff, because it is considerably higher than the maximum in-home ambient temperature measured using thermometers, which was 35°C. We therefore reduce the possibility of counting normal temperature variations as cooking events by using the 40°C cutoff. We also conduct robustness checks with 45°C and 60°C cutoff points and find that results do not change.
The three-stone tripod is the traditional injera baking technology in rural Ethiopia and in our study sites. It is difficult to reliably monitor three-stone tripod cooking using SUMs, because the stones are often moved, fires are less controlled than with an enclosed stove, and SUMs may be destroyed. We therefore only analyze use of the ICS, which is the objective of this paper, implying that we are unable to observe effects on overall cooking behavior, including use of three-stone tripods. Images of the Mirt stove are included in online appendix figure A1.Footnote 8
3.2 Sampling and fieldwork
Our sampling frame is based on a random sample of 110 villages in local jurisdictions called kebeles. Footnote 9 We eliminated 29 of the 110 villages because they were part of our pilot survey, injera was not typically baked in villages or the normal three-stone tripod technology was not used. We then selected 36 villages using stratified proportional random sampling from the 81 villages that were left. Stratification was by regional state (Amhara, Oromia, and Southern Nations, Nationalities and Peoples (SNNP)), with forest area used as the state weight. Forest area was used because most fuelwood comes from forests and it also proxies well for total regional state population and land area. In total, 20 per cent of villages were from Amhara, 50 per cent from Oromia and 30 per cent from SNNP regional states. From the 36 village/site clusters, 360 households were randomly selected to receive the Mirt stove. A total of 10 households received the Mirt stove in each village.Footnote 10
Respondents were informed about the nature of the survey and experiment and formal oral consent was received from all respondents. Full information was provided individually to respondents on stove features, usage methods, the SUMs, and details about the monetary treatment randomly applied to them. Though all households were monitored using the SUMs, we acknowledge that knowing they were being monitored could itself potentially increase Mirt stove usage.
Respondents were also informed that stove use was not required to receive a stove, that the stove was anticipated to reduce fuelwood consumption and respondents were encouraged to use their Mirt stoves. While the strategy to address refusals was to replace refusing households by randomly choosing from remaining households in each site, there were no refusals during the random distribution of the stoves or abandonment until after the first year of data collection. There was very low attrition for the long-term study, with the end-line (2016) data collection including approximately 95 per cent of the 2013 households.
The baseline survey and user training, which were conducted during June–August 2013, were done concurrently with stove distribution, and included household characteristics and socioeconomic variables. In 2016, when the final monitoring period occurred, a second survey was conducted. This survey updated baseline household characteristics, but particularly focused on kitchen characteristics and use/disuse and abandonment of the Mirt stove. Field workers were experienced and were trained directly by investigators, with approximately four enumerators working under each of the five field supervisors. All of the supervisors and many of the enumerators who conducted the follow-up field work in 2016 also participated in the baseline research in 2013.
In addition to installing the stoves in the injera baking area at baseline (the first period of the study), the fieldworkers installed the SUMs on the stoves and initiated them in each of the five periods the study was conducted. The five monitoring periods were: June–August 2013 (period 1); August–October 2013 (period 2); March–May 2014 (period 3); May–July 2014 (period 4) and November–December 2016 (period 5). The use of multiple follow-up periods (four in this study) rather than just one, is expected to increase the power of the analysis (McKenzie, Reference McKenzie2012).
Randomization in the selection of Mirt stove recipients within each village was at the household level. Randomization of the monetary treatments was at the village level to ensure that all those in a village received their Mirt stoves on the same terms, with 12 villages receiving each treatment, for a total of 120 observations within each treatment and period.Footnote 11 Households in Ethiopian villages were likely to know the terms under which their neighbors received the improved stoves. Allocation of stoves using different treatments within a village would have likely created resentment, which would have been unethical and potentially biased usage intensity. This village-level distribution approach also reflected typical implementation of improved stove programs, where all individuals in villages would generally receive stoves under the same conditions.
In sum, therefore, three equally weighted monetary treatments were distributed randomly across villages using the stratification procedure noted above, and 10 households were randomly selected to receive Mirt stoves within each village, for a total of 360 observations in each period. As previously noted, there was very low attrition in this long-term study, but to analyze a consistent sample of households in all rounds, the sample is restricted to those households that could be reached and provided valid data in 2016, which was the end line. In the analysis of regular stove use and frequency of stove use, we include only period 5 observations from households who had not abandoned the stove.Footnote 12
4. Results
Our analysis starts with evaluating whether our randomly-assigned monetary treatments and controls may have contributed to households abandoning their stoves between July 2013 and November/December 2016 (up to 3.5 years after the initial distribution of the stoves). These results are then followed by analysis of longer-run usage frequency, including effects of treatments on use of stoves at least twice per week, both considering and ignoring dis-use. Our main findings are that almost 2/3 of stoves were still used after 3.5 years, those that were in place in 2016 were on average regularly used and our monetary treatments did not appear to affect longer-run regular usage.
Descriptive statistics for our outcome variables are given in table 1 for the pooled sample, and for households for which the stove was in place in 2016 (i.e., period 5). For the pooled sample, we find that average usage is 2.4 times per week and 47 per cent of households used the stove an average of two or more times per week. The corresponding figures for households in period 5 who had not abandoned the stove were slightly higher (2.5 times/week and 52 per cent regular usage).
Descriptive statistics of outcome variables
a This total does not include observations for period 5 in which the household abandoned the Mirt stove.
Indicator variables for the three monetary treatments (approximately 1/3 of sample each), which are our variables of interest, as well as one-time village-level group training (half of households) and controls are included in all models. Table 2 presents descriptive statistics for the controls, which were collected at baseline and are potential drivers of the decision to keep stoves in place in the long run, the frequency of use and also regular use of the Mirt stove. These variables include socioeconomic characteristics, such as gender, education and age of the respondent, indicators of economic status and cooking-related characteristics. As shown in table 2, about 87 per cent of respondents were men, a majority of whom were illiterate. The mean family size was five adult equivalents.
Descriptive statistics of baseline controls
Notes: Households with valid data in 2016 are included. Number of observations varies by variable.
a TLU is a standard method developed by the Food and Agriculture Organization (FAO) for combining different livestock types into one metric based on weight, with a cow = 1.
b Please see Vieider et al. (Reference Vieider, Beyene, Bluffstone, Dissanayake, Gebreegziabher, Martinsson and Mekonnen2018) for a discussion of the estimation method.
c As part of the study, households were randomly selected to participate in a controlled cooking test to compare the traditional tripod stove with the Mirt stove. This variable is included as a control variable to see if it had effects on use of the Mirt stove.
Tropical livestock units based on cow-equivalents (mean of about 5) and whether the house has a corrugated roof (>70 per cent) are potential indicators of economic status, as is whether the household uses pure teff flour. Teff flour is more expensive than other grains used to bake injera and, as shown in table 2, a minority of households use it. Distance to all weather roads measures market access and integration and we find that, on average, households live within an hour of walking time of roads.
We control for participation in controlled cooking tests to adjust for potential Hawthorne effects that could affect the dependent variables.Footnote 13 Whether the household uses the stove for purposes other than baking attempts to capture whether households are making full use of the Mirt stove, which can cook coffee and to some extent stews while simultaneously baking injera. Most respondents report only using their Mirt stoves for cooking injera. The number of injera baked during a session captures how intensively the stove is used at each baking, which can affect usage and abandonment, and we find a mean of 20 injera. Whether the stove was installed inside the house (as opposed to in a separate kitchen) also may be correlated with use and dis-adoption of the stove, because of undesirable smoke in the home and potentially higher chance for breakage. Almost 2/3 of households placed their Mirt stoves in separate kitchens rather than in the main house.Footnote 14
4.1 Dis-adoption of stoves prior to November/December 2016
At the end of 2016 (about three-and-a-half years after baseline), we conducted a visual inspection and survey to identify whether the stoves adopted in 2013 were still in place. About 63 per cent of those who received a stove were still using it 3.5 years after initial distribution, which is high considering that these stoves were approaching the end of their approximately five-year lifespans.
About 83 per cent of the 121 respondents who gave reasons for abandoning the stove said the driver of abandonment was breakage. Other reasons included ‘I do not need it, because I do not bake injera’ and ‘I have a better improved stove.’ In our model of abandonment, we also included the average number of times households used their Mirt stoves per week in period 4, which is the period immediately preceding the last period of data collection (and hence predetermined), to adjust for intensity of use, which may affect dis-adoption (mean = 2.4 times/week).
Table 3 presents results of probit regressions to examine potential drivers of dis-adoption 3.5 years after the introduction, where the outcome is coded as 1 if the stove is still in place and 0 otherwise. The results in columns 1 and 2 show that none of the treatments are correlated with retention of the stove when compared to the reference category of Free Treatment and results are unaffected when other covariates are added as shown in columns 3 and 4. Households with larger family size are more likely to dis-adopt, perhaps because additional members contribute to breakage of the stoves. Households who were randomly selected to participate in controlled cooking tests were more likely to still be using the stove in 2016 (p < 0.05), as were households that bake more injera at a time, perhaps because they are more reliant on the technology.
Probit regression of whether Mirt stove was in place in 2016
Notes: Standard errors are clustered at the site level. Dependent variable is stove use, which is 1 if stove is still in use and 0 otherwise.
a TLU converts the different livestock types into one unit.
Mean frequency of Mirt stove use per week by period and treatment
Notes: Numbers of observations are reported in parentheses. For periods 1 to 4, the number of observations is for households with valid data for period 5. For period 5, the number of observations is for households who did not abandon the stove in period 5.
4.2 Stove use in the longer run
In our analysis of stove use in the longer run, we look at the extent of usage by households who provided valid data for period 5 and also use Heckman selection models to adjust for disuse. Our analysis focuses attention on period 5, with the objective being to analyze longer-run effects. Outcome variables and controls used are presented in tables 1 and 2 respectively.
Table 4 shows frequency of Mirt stove use by period and treatment, measured using the SUMs for households with valid observations in 2016, but excluding period 5 observations for households who had abandoned the stove (mainly due to breakage). The numbers represent average times the stove was used per week during each of the approximately 6-week monitoring periods. Results are broadly in accord with normal injera cooking patterns and the stated average usage reported in Gebreegziabher et al. (Reference Gebreegziabher, Beyene, Toman, Bluffstone, Dissanayake and Martinsson2018). Average use per week was higher than two times per week in period 5, which is in line with regular usage. The data show an increase in average usage from period 1 to 2 (p < 0.03) and period 2 to 3 (p < 0.01), a decline from period 3 to 4 (p < 0.08) and a statistically insignificant increased usage from period 4 to 5 (p < 0.26). Average usage increased from period 1 to 5 (p < 0.01).Footnote 15
Table 5 presents household random effects probit regression coefficient estimates to examine whether treatments by period are related to regular stove use, which is defined as using the stove on average at least twice per week, with various temperature threshold definitions of cooking events (40°C, 45°C, and 60°C). We define use in terms of regular use to focus on adoption in the sense that households use the Mirt stove in accordance with normal injera baking frequency. Free Treatment period 5 is the omitted category.Footnote 16
Household level random effects probit model of regular use of Mirt stove (at least twice per week) with three cooking event thresholds (in degrees °C)
Notes: Standard errors in parentheses. Free treatment period 5 is the reference group.
As a robustness check, we apply a Heckman selection model to analyze usage intensity after accounting for possible selection bias that may arise due to exclusion of those who do not use the stove (table 6). As an exclusion restriction, we include random assignment of a household to participate in controlled cooking tests as an explanatory variable only in the selection equation under the assumption that it does not affect cooking frequency. Tests of this exclusion restriction are shown in the online appendix, tables A2 and A3, where participation in controlled cooking tests is shown to be unrelated to regular usage. As demonstrated in table A5 (online appendix), participation in controlled cooking tests is positively related to whether the Mirt stove was used (p < 0.01). Tables 5 and 6 do not include details on socioeconomic and cooking characteristics, but these are included in tables A2 and A5 in the online appendix.
Heckman selection model assuming 40°C definition of cooking events
Notes: Standard errors are clustered at site level. Free treatment in period 5 is the reference group.
a Except participation in controlled cooking test, which is only included in the selection equation.
We find that for the 40°C definition of cooking events, estimates are statistically significant for Pay Treatment in periods 1 and 2 and Free Treatment in period 1. In all these cases the coefficient estimates are negative, suggesting that regular usage was generally higher for Free Treatment in period 5 compared with earlier periods. All other estimates are statistically insignificant. These results hold regardless of cooking event definition except that for the 45°C definition Incentive Treatment in periods 1 and 4 are also significant, but only at the 10 per cent level, and for the 60°C definition Free Treatment in period 1 is not statistically significant. Probit analysis with clustered standard errors at the site level for the 40°C definition shows similar results, except that Free Treatment in period 1 is statistically insignificant and Pay Treatment in period 2 is significant only at the 10 per cent level (online appendix table A3). A similar probit analysis only for period 5 with clustered standard errors at the site level shows that none of the treatments are statistically significant (online appendix table A4).
These results suggest some evidence of increased regular use over time, which we term persistence. This long-run persistence compared with early periods is similar to our previous results, which suggested that regular usage on average increased through period 4 for Free Treatment and also for Pay Treatment, which had very low regular usage in period 1 (Bluffstone et al., Reference Bluffstone, Beyene, Gebreegziabher, Martinsson, Mekonnen and Vieider2021). Results from period 5 suggest that average usage leveled out, but there was little slippage in terms of regular usage.
We present the Heckman selection model, which adjusts for non-use of the stove, in table 6. Outcome and selection equations both include socioeconomic and cooking characteristics (table A5, online appendix). Results are generally similar to the random effects probit model. In the Heckman model, Incentive Treatment usage in periods 1 and 5 is lower compared with Free Treatment in period 5 (p < 0.05). Incentive Treatment usage in period 4 and Pay Treatment usage in period 1 are higher than Free Treatment in period 5. Usage with other treatments and periods are not statistically significantly different from Free Treatment in period 5. Thus, the Heckman selection model generally shows longer-run persistence, but no other significant results.
5. Discussion
Sustained use of improved technologies, such as biomass cookstoves, is critical to actually generating benefits from ICS promotion programs. Our results suggest that 3.5 years after receiving the stove, which is close to the expected 5-year lifespan, about 63 per cent of households covered by the survey still had their Mirt stoves in place, which appears to be quite high. Among those who still had their stoves, the average frequency of stove use in period 5 (i.e., 2016) is no different than the latter part of the first year (periods 3 and 4), but in several cases greater than shortly after initial distribution (periods 1 and 2). If users retain stoves as long as marginal benefits exceed marginal costs, these findings suggest that even after 3.5 years, the Mirt stove provides positive net benefits to users.
We find little systematic relationship between treatments and retention of the stove, which suggests that abandonment is due to reasons other than our randomized treatments. Based on our end line survey, the vast majority of those who dis-adopted their stoves report that it was because their Mirt stoves broke and therefore either became too inconvenient to use or could not be used at all. People therefore do not seem to have abandoned the Mirt stove simply because they did not like it.
Analysis of regular stove use, defined as use of the stove at least twice per week, suggests no differences across treatments 3.5 years after initial distribution. We acknowledge that our relatively small sample size in each period could mean we were simply unable to detect effects that actually existed, but that we were able to statistically identify effects with fewer observations over only one year (Bluffstone et al., Reference Bluffstone, Beyene, Gebreegziabher, Martinsson, Mekonnen and Vieider2021) gives us some confidence that indeed any treatment effects dissipated in the longer run.
We examine effects across time by treatment and find that between the two earliest periods and the end line, households who were randomly required to pay a monetary price for the stove show larger increases in regular usage, regardless of model or definition of cooking events (40°C, 45°C or 60°C). This result may simply be because Pay Treatment households initially on average used the Mirt stove so much less than those who randomly received the other two treatments. By period 5, average Pay Treatment usage was between the other two treatments.
6. Conclusions and policy implications
Achieving sustained, regular use of ICS technologies has been perhaps the major stumbling block for ICS promotion policies and programs. If households are not found to voluntarily regularly use improved stoves after distribution, those technologies cannot produce the benefits that were engineered into stoves, or perhaps after considering costs, benefits are not sufficient to warrant true adoption by households. This paper analyzes the extent to which the Mirt improved cookstove for baking injera continued to be regularly used by 360 randomly selected households in three regional states in rural Ethiopia, 3.5 years after it was distributed to them. We also examine if there are any lingering effects on stove usage of different monetary treatments commonly used in stove distribution programs (providing the stove for free, requiring a monetary payment and providing an incentive payment for stove use). The data come from a randomized field experiment begun in summer 2013.
The main conclusion of this paper is that households who randomly received the stove on average continued to regularly use it 3.5 years after distribution – about 70 per cent of the stove's expected life – unless it had broken.Footnote 17 This finding points to positive net benefits that users received from the stove, compared with exclusive use of the three-stone tripod for both injera baking and other cooking tasks. Over the longer-term period of stove use that we study in this paper, monetary treatments appear to be relatively unimportant. While requiring monetary payments from Pay Treatment users could potentially have created disgruntlement early-on, depressing usage, regular usage appears in the longer-run to be driven by factors other than our monetary treatments.
As the end of the Mirt stove life approaches, regular usage is similar across treatments and remains high. Persistent regular usage, combined with the reduction in fuelwood consumption and carbon emissions reported in Mekonnen et al. (Reference Mekonnen, Beyene, Bluffstone, Gebreegziabher, Martinsson, Toman and Vieider2022), is likely to imply sustained benefits to households, as well as society at large through climate change mitigation benefits. These benefits are dependent on voluntary usage, which our analysis suggests continues into the long run. To ensure long-term use of Mirt stoves, it is important to have a plan on how to replace broken stoves. This could, for example, be done by direct replacement or by giving spare stoves to kebeles where broken stoves could be replaced.
Though over time the energy mix will shift to non-biomass fuels, the contemporary overwhelming dominance of biomass cooking fuels in Ethiopia suggests that ICS may be an important bridge technology and interim energy policy focus. As most households have chosen to regularly use Mirt for most of its useful life, the stove is low-cost and offers a variety of private and social benefits, policymakers should strongly consider investing more in the Mirt stove. It is unfortunate that the useful life of Mirt is only about five years, because with sometimes weak supply chains and rough roads, replacing stoves in rural Ethiopia can be a challenge. There are also recent developments to improve the technology to further reduce indoor air pollution. Policy and research attention should therefore at least partially focus on increasing durability or taking other steps to extend the expected stove life beyond the current five years.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355770X25000117.
Acknowledgements
Financial support for this work was provided by the World Bank. The World Bank had no other involvement with the work done, which reflects solely the efforts of the authors. The authors acknowledge comments from participants of the Sustainable Energy Transition Initiative (SETI) workshop held in Santiago, Chile in May 2019.
Competing interest
None.
Open access
