Energy Economics 36 (2013) 605–613

Contents lists available at SciVerse ScienceDirect

Energy Economics

j ourna l homepage: www.e lsev ie r .com/ locate /eneco

Clean fuel-saving technology adoption in urban Ethiopia☆

Abebe D. Beyene, Steven F. Koch ⁎Department of Economics, University of Pretoria, Private Bag X20, Hatfield 0028, South Africa

☆ We are very grateful to Ato Melesaw Shanko, manaLtd., and Hiwote Teshome from GTZ for making the dataMekonnen for helping us connect with these individuatheir willingness to discuss improved biomass cook stwould also like to thank Economic Research Southern Aresearch. Finally, thanks to the anonymous reviewers fothat their comments have greatly improved this manare solely the responsibility of the authors.⁎ Corresponding author. Tel./fax: +27 12 420 5285.

E-mail addresses: abebed2002@yahoo.co.uk (A.D. Be(S.F. Koch).

0140-9883/$ – see front matter © 2012 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.eneco.2012.11.003

a b s t r a c t

a r t i c l e i n f o

Article history:Received 6 July 2011Received in revised form 14 October 2012Accepted 3 November 2012Available online 9 November 2012

JEL classification:Q41Q42Q55Q56

Keywords:Improved stovesDuration analysisAdoptionUrban Ethiopia

The heavy dependence and inefficient utilization of biomass resources have contributed to the depletion offorest resources in Ethiopia, while the use of traditional cooking technology has also been linked to indoorair pollution and poor health. In response, the government and other institutions have pushed for the adop-tion of new cooking technologies, with limited success. This research examines the reasons underpinning thelack of widespread adoption, via duration analysis, correlating the speed of adoption of Mirte and Lakechcook stoves – two examples of new cooking technologies – in urban Ethiopia to socioeconomic factors.According to the duration analysis, adoption rates have steadily increased over time, while economic factors,such as product price, household income and household wealth, are, for the most part, important determi-nants of adoption behavior. There is also evidence that the availability of substitute technologies tends to hin-der adoption, and that there are large regional differences in adoption rates, suggesting the need for a moredetailed regional analysis of adoption decisions.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

According to the International Energy Agency (IEA, 2002), tradi-tional stoves using dung and charcoal are inefficient and emit largeamounts of carbon monoxide (CO) and other noxious gases. As a re-sult, poor people in the developing world are constantly exposed toindoor particulate matter and carbon monoxide concentrationsmany times higher than the World Health Organization (WHO) stan-dards. Surveys by Bruce et al. (2002), Emmelin and Wall (2007),Fullerton et al. (2008) and Smith et al. (2004), summarize thestrength of association between indoor air pollution (from, especially,biomass fuel use) and a wide range of illnesses and diseases. Associa-tions are shown to exist for acute lower respiratory tract infection,low birth weight, nutritional deficiency, interstitial lung disease,chronic obstructive lung disease and lung cancer, tuberculosis, car-diovascular disease, and cataracts; similar information can be found

ging director of MEGEN Poweravailable, as well as to Alemu

ls. We gratefully acknowledgeoves in Ethiopia. The authorsfrica for their support of thisr their comments. We believeuscript. Any remaining errors

yene), steve.koch@up.ac.za

rights reserved.

in WHO (2006). These health problems tend to be greater where tra-ditional cooking technology is more common (Masera et al., 2007;Smith and Mehta, 2003; Tasleem et al., 2007), such as in Ethiopia.

Ethiopians, like citizens in many developing countries, are highly de-pendent on biomass energy sources: fuel wood, charcoal, animal dungand crop residues. As noted by Geist and Lambin (2003) and Vance andIovanna (2006), socioeconomic factors, such as poverty, force people indeveloping countries, including Ethiopia, to exploit forest resources forboth domestic energy consumption and commercial gains. The afore-mentioned biomass energy sources, according to the Ethiopian Environ-mental Protection Agency (EPA, 2004), account for more than 90% oftotal domestic energy demand — approximately 99% of demand in ruralhouseholds compared to 94% of demand in urban households. Giventhe high levels of dependence, biomass sourceswill continue to dominateenergydemand inboth rural andurbanEthiopia in the foreseeable future.

Ethiopian dependence on biomass fuels impacts on the health ofits citizens, especially women and children. TheWorld Health Organi-zation (WHO, 2002) estimates that fumes from indoor biomass fueluse kill 1.6 million women and children in developing countries,each year, accounting for 3% of the global burden of disease. More re-cent information contained in WHO (2009) suggests that 1.1 millionfemale and 0.9 million male deaths worldwide (0.5 million total inAfrica) can be attributed to indoor smoke from solid fuels, such thatbiomass fuel use contributes 3.3% of the global burden of disease.The figures for Ethiopia, though, given its dependence on biomassfuels, are proportionately worse. According to the WHO (2002) re-port, with 95% of households using biomass fuels as their primary

606 A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

energy source, 4.9% of the Ethiopian burden of disease can be attrib-uted to solid fuel use for cooking, heating and lighting; nearly50,000 deaths can be attributed to the same cause.

In order tomitigate the adverse impact of indoor air pollution and re-duce pressure on forests, the Ethiopian government has devised a num-ber of strategies. Of particular relevance to this research is the promotionof alternative modern fuels and support for improved biomass cookstoves (Cooke-St. Clair et al., 2008). The Lakech and Mirte stoves,discussed below, are two such examples. The realization that improvedcook stove technology has the potential to alleviate the pressure on bio-mass resources and improve health led to improved cook stove pro-grams in a number of developing countries, including Ethiopia.1

Similarly, a number of institutions, e.g., the Ethiopian Rural EnergyDevelopment and Promotion Center (EREDPC), and other organizations,such as the German Technical Corperation (GTZ), have been involved inthe development and dissemination of different types of biomass cookstove technologies since the early 1970s in Ethiopia (EPA, 2004). Mostrecently, in December of 2010, the US EPA and theUS Peace Corps signeda Memorandum of Understanding (MoU), and that MoU includedsupport for the Global Alliance for Clean Cook Stoves in Ethiopia.2

Unfortunately, as discussed in Barnes et al. (1994) and Shanko et al.(2009), the efforts to disseminate various types of fuel-saving technol-ogies have faced different problems at different times. For example,some of the stove programs were not successful, due to problems relat-ed to the stove itself (technical problems). Other programs were notsuccessful, due to a lack of understanding of consumer tastes, whilesome programs were not successful, due to the lack of an appropriatepromotion strategy. In addition to implementation problems, thereare real concerns that the expected forestry benefits may not obtain.Specifically, the rebound effect – intuitively, better technology results ina decrease in the price of inputs yielding scale effects – has been observedin a number of locations. Nepal et al. (2010), for example, find that im-proved cook stoves in Nepal do not yield reductions in the demand forfirewood. Sorrell et al. (2009) provide amore detailed reviewof literaturein relation to the rebound effect.

Although it is not clear that new cook stove technology will alleviateforest dependence, there is a strong evidence of significant potentialhealth benefits for households that adopt improved cooking technology.Given the expected household benefits, the failures of earlier programsand the renewedemphasis on invigoratedpromotion efforts, an examina-tion of adoptiondecisions at the household level deserves attention.How-ever, most available studies related to technology adoption, especiallythose related to improved biomass cook stove technologies – Amacheret al. (1992), Gebreegzihabher et al. (2005) and Jan (2012)–have focusedonly on the dichotomous decision to adopt new technologies, and havenot considered the timing of adoption. Although informative, these binaryanalyses are static and ignore the dynamic nature of the adoption process.Furthermore, the aforementioned studies have focused on rural house-holds. Therefore, this research makes two contributions to the literature.

First, the available limited studies focus on rural areas, such thatthe urban sector is under-represented. However, the high depen-dence of urban dwellers on biomass resources also contributes to en-vironmental and health problems. For example, charcoal, theproduction of which is one of the main causes of deforestation in Ethi-opia, is almost exclusively used in urban areas, irrespective of thelevel of living standards. Moreover, since many households cannot af-ford modern energy sources, such as kerosene, liquefied petroleumgas (LPG) and electricity, a substantial portion of the urban poorwill continue to rely on fuel wood and charcoal. Therefore, focusing

1 Barnes et al. (1994) provide an excellent survey of the programs put in place before1994, as well as the lessons that could be learned from those programs, whileBhattacharya and Abdul-Salam (2002) provide a detailed description of programs inIndia and China.

2 In addition, Ethiopia's Climate-Resilient Green Economy (CRGE) document (EPA,2011) includes a plan to distribute up to 9 million improved stoves by 2015, in orderto reduce GHG emissions from fuel wood consumption.

on urban households is useful, from the viewpoint of protecting forestcover, as well as reducing the ill effects of biomass fuel use on health.

Second, the commonly applied binary dependent variable analysis,which considers only adoption or non-adoption, does not account foradoption over time, since it does not allow for differences in the time toadoption by the households. This analysis, therefore, employs durationanalysis, rather than static analysis, and, as far as we are aware, is thefirst to do so,within the context of improved cook stove technology adop-tion. Themain objective of this research is to examine and understand thedeterminants of the speed of adoption of fuel-saving technologies, espe-cially for Mirte and Lakech cook stoves, in urban Ethiopia. Though manyfactors, such as the technical design of the stove, are likely to affect thespeed of adoption, the data available for this study allows us only toaddress socioeconomic factors associated with the dissemination ofimproved biomass cook stoves in urban Ethiopia.

For the analysis, a series of proportional hazard models are esti-mated to examine the timing of the adoption of improved biomasscook stoves, based on data collected amongst urban households inEthiopia. The timing of adoption is, unfortunately, based on recalldata collected at one point in time, rather than on panel data collectedover time, which would have been preferred, since recall is rarely per-fect. Importantly, it was possible to include price information fromthe year of adoption to control for some changes that occurred inthe market for improved biomass cook stoves in Ethiopia. However,the rest of the variables are time-invariant, and generally collectedpost-adoption, due to the nature of the survey. The results of the anal-ysis point to adoption rates that increase over time, as well as eco-nomically appealing price and income effects – demand curves aredownward sloping and cook stoves are normal – however, the eco-nomically appealing results are not statistically significant in all cases.

The analysis unfolds in the usual fashion. The next section outlinesthe empirical methodology. Section 3 discusses the stove technolo-gies examined in the analysis, as well as the survey data used forthe analysis. The variables used in the analysis, as well as the litera-ture underpinning the choice of these variables, receive special atten-tion in this section. The results of the empirical analysis are presentedin Section 4, while Section 5 concludes.

2. Methodology

The analysis of duration data, commonly referred to as survivalanalysis, has been applied in anumber of situations in economics, demog-raphy and medicine. In terms of medical research, the focus is often onpatient survival following either disease diagnosis (Brookmeyer et al.,2002) or the administration of a medical treatment (Locatelli et al.,2001). In demography, survival analysis is often applied in the examina-tion of mortality rates and relates to the length of time a child survivesfrom birth, or the time that a mother survives following childbirth;some examples include Abou-Ali (2003), Handa et al. (2010) and Lavyet al. (1996). Within economics, unemployment duration and the dura-tion of strikes have often been examined via duration models, such asKennan's (1985) and Jaggia's (1991) analyses of strike duration in theUSmanufacturing sector. Most relevant to this study, though, is the anal-ysis of technology adoption – Burton et al. (2003), Dadi et al. (2004), andFuglie and Kascak (2001) – and Lee's (2003) adoption of privatizationpolicy. Burton et al. (2003) suggest that duration analysis has strengthscompared to the conventional bivariate outcomemodels, since simple bi-variate outcomemodels cannot capture the inter-temporal nature of theadoption process. Under these circumstances, the use of durationmodelsis superior to the analysis of adoption at any one point in time.

Survival analysis depends primarily on the distribution of dura-tions, or the length of survival times, in the population. Followingthe standard formulation, let T≥0 denote the duration, while t de-notes a particular value of T. In our case, duration is the length oftime, measured in years, until the household adopts the new technol-ogy; the formulation of this measure is described below. The

Table 1Sample location information.

Region (total) Town Sample size Percent

Amhara (580) 36.80% Bahirdar 424 26.89AmbaGiorgis 60 3.80Dagolo 96 6.09

Oromiya (667) 42.30% Atnago 66 4.19Goba 409 25.94Kofele 192 12.18

Tigrai (330) 20.93% Hiwane 51 3.23Mehoni 177 11.22AdiDaero 102 6.47

607A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

cumulative distribution function (CDF) of T is defined as F(t)=P(T≤ t), assuming t≥0. Assuming that T is continuous, the survivorfunction is defined as S(t)=1−F(t)=P(T> t), and describes theprobability that a duration will last longer than t, assuming survivalup to t. In our analysis, survival refers to the length of time to adop-tion of either of the new technologies by the household, while thehazard, discussed below, is the actual adoption.

One of the central concepts in the analysis of duration data is the haz-ard function. Assuming an individual occupies a given state up to time t,the probability that such an individual exits from the state within an in-terval Δ, at or after t is P(tbT≤t+Δ|T≥t). Therefore, the average proba-bility of leaving the state at or after t, per unit of time Δ, can be used tocreate the hazard function. Assuming a differentiable CDF, that hazardfunction is defined in Eq. (1); see Cameron and Trivedi (2005).

h tð Þ ¼ limΔ→0

PtbT≤t þ ΔjT≥t

Δ

� �¼ f tð Þ

S tð Þ ð1Þ

As defined before, S(t) is the survival function, while f(t) is theprobability density function. The hazard function specifies the instan-taneous rate of completion of a spell at T= t, conditional upon surviv-al up to time t. In our case, the hazard function, therefore, representsthe probability that a household adopts the improved stove at time t,given that it has not adopted before t. In other words, higher hazardrates indicate higher rates of adoption.

A variety of functional forms have been proposed for the durationmodels; Kiefer (1988) presents a very detailed summary of the differentdistributional assumptions behind these models. The two most widelyused parametric distributions are the exponential distribution and theWeibull distribution. The exponential distribution is characterized by aconstant hazard function, h(t)=λ, where the constant parameter, λ>0,implies that the passage of time does not influence the hazard rate. Dueto the assumption that subjects fail at the same rate through time,hazards of this sort are referred to as memoryless hazards. However, itmay be preferable to allow for a hazard with memory. The other com-monly applied distribution, the Weibull distribution, is characterized bythe hazard function h(t)=λptp−1, with λ>0 and p>0. Given theWeibull specification, the hazard rate is constant, monotonically increas-ing or monotonically decreasing depending on p. It is monotonicallyincreasing if p>1, and decreasing if pb1. If p=1, the Weibull hazardcollapses to the exponential hazard, and is, therefore, constant.

Assuming the duration for each individual, ti, is independent andnot censored, the log-likelihood function for completed spells isgiven in Eq. (2), where θ=(λ,p) is the vector of parameters and Xi

is a vector of household i explanatory variables.

‘ðθ Xj Þ ¼ ∑N

i¼1lnf ti θ;Xij Þ:ð ð2Þ

However, in this analysis, aswithmost analyses of durationdata, thereare censored observations, especially right-censored observations, inwhich case, it is only known that adoption has not occurred during theobservable time horizon. Therefore, the density function in Eq. (2) cannotbe applied; instead, it must be modified to allow for censoring. Thus, thelog-likelihood function contains two components, one for non-censored,di=1, observations and another for censored observations, di=0; K inEq. (3) represents the number of non-censored observations.

‘ðθ Xj Þ ¼ ∑K

i¼1di lnf ðti θ;Xij Þ þ ∑

N

j¼Kþ11−dið Þ lnS ti θ;Xij Þ:ð ð3Þ

The preceding discussion, although conditioning on additional co-variates, has not yet outlined the inclusion of these covariates withinthe likelihood structure; an oversight rectified in what follows. Eachof the popular models previously discussed, the exponential andWeibull models, are members of the proportional hazard family.

This model family incorporates separation between the time compo-nent and the contribution of the other covariates: h(t,X,θ,β)=h0(t,θ)g(X,β), where β is a vector of parameters to be estimated, h0 is thebaseline hazard, and g is the relative hazard. The most common func-tional specification for the relative hazard, also used here, is g(X,β)=exp(Xβ); the specification ensures non-negativity of the underlyinghazard function. Furthermore, this proportional specification allowsfor easy interpretation of the results, since the marginal effect of achange in any x∈X is simply the coefficient times the original hazard.

Anothermember of the proportional hazard family is the Cox (1972)proportional hazard model. One of the most attractive features of Cox'smodel is that the baseline hazard need not be estimated. Further, it isonly assumed that the hazard function is the same for each subject,and that given the covariates, the hazard between one subject and theother differs only by a multiplicative constant, based on the relativehazard. Given that this model does not specify the underlying hazard,it is also used, below, to check the robustness of the results.

The specifications described up to now assume that all observationsbehave identically. However, it is likely that there are unobservable fac-tors influencing household decisions. This problem, unobserved hetero-geneity, can create bias in the estimates. Mathematically, the easiestsolution is to multiplicatively append a stochastic term to the hazardfunction, h(t,X,θ,β)=h0(t,θ)g(X,β)ε, and assume a distribution for thatstochastic term. One common assumption applied in the literature isthat the stochastic term follows the gammadistribution. Below,we con-sider theWeibull–Gammamixturemodel, as described in Cameron andTrivedi (2005) and applied byGutierrez (2002), and test for unobservedheterogeneity. Unfortunately, the Weibull–Gamma mixture model forthe Lakech stove duration did not converge, so we were not able totest for unobserved heterogeneity; therefore, this test only applies tothe Mirte stove model.

3. Variables, data and hypotheses

The data for the analysis comes from the ‘Mirte Biomass InjeraStoves Market Penetration and Sustainability’ study conducted byMegen Power, Limited, in 2009. A number of explanatory variablestaken from the survey are included in the analysis, based, in part, onthe literature, discussed in the following subsection. Descriptive sta-tistics and the definitions of those variables are available in Table 3.

The survey was conducted in the Amhara, Oromiya and Tigrai Re-gions. Three towns from each region were selected for the survey. Forthe purpose of sampling, towns were classified into three categories:high-sales towns, low-sales towns, and non-project towns. The samplesize for each region and town was determined, proportionately, basedon the total number of households. Finally, based on sampling frames(determined by lists of households) obtained from the respectiveKebeles, households were selected using a simple random samplingtechnique. The towns selected for the study are presented in Table 1.The number of sampled households was 1577. The questionnaire wasfurther refined prior to fieldwork, through discussion and joint reviewwith enumerators; pre-testing of the questionnaire was undertakenwith a few households before the main sample interviews.

Table 3Descriptive statistics of the covariates of fuel-saving technologies and their expectedsigns (N=1557).

Variable Mean S.D. Min Max

Household head is male (−) 0.68 0.47 0 1Age of HH head at the time of the survey (+) 44.88 13.50 18 102Household head is illiterate or has no education(base)

0.21 0.41 0 1

Household head can read and write elementaryeducation (+)

0.42 0.49 0 1

Household head education between grades 9 and12 (+)

0.20 0.40 0 1

Household head has at least high school (+) 0.17 0.37 0 1Number of children aged 5 or younger (+) 1.75 1.54 0 14Number of adult members of the family (+/−) 3.38 1.87 1 15Household privately owns dwelling (+) 0.72 0.45 0 1Household ownership of separate kitchen (+) 0.75 0.44 0 1Price of Lakech (−) 13.56 1.88 9.12 40.65Price of Mirte (−) 58.87 0.67 35.46 70.92Monthly income is less than ETB500 (base) 0.57 0.49 0 1Monthly income is between ETB501 and ETB1499(+)

0.30 0.46 0 1

Monthly income is between ETB1500 and ETB2499(+)

0.09 0.29 0 1

Monthly income is above ETB2500 (+) 0.04 0.20 0 1Household owns electric Mitad, a substitute forMirte (−)

0.08 0.27 0 1

Household owns metal stove, a substitute forLakech (−)

0.48 0.50 0 1

Household owns clay stove 0.31 0.46 0 1Tigrai Region (+) 0.21 0.41 0 1Amhara Region (+) 0.37 0.48 0 1Oromiya Region (−) 0.42 0.49 0 1

Note: (+) and (−) denote the expected regression sign. ETB stands for Ethiopian Birr,which exchanged at the rate of 1USD=11.21ETB.

Table 2Mirte and Lakech biomass cook stove adoption by sample region.

Stove type Total Oromiya Tigrai Amhara

Mean S.D. Mean S.D. Mean S.D. Mean S.D.

Lakech 0.346 0.476 0.357 0.479 0.201 0.401 0.418 0.494Mirte 0.254 0.435 0.340 0.474 0.100 0.301 0.243 0.429Observations 1557 659 329 569

608 A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

The survey also includes information related to each household'ssocioeconomic characteristics, such as: income; the age, educationlevel, sex and occupation of the household head; type and ownershipof improved biomass cook stoves; type and ownership of substitut-able cook stoves3; children and adults in the household; house own-ership and characteristics of the house. Although the initial samplecontained 1577 observations, some observations were dropped dueto insufficient or missing data. Importantly, households not using bio-mass for injera baking are omitted. Moreover, household headsreporting their age to be less than 18 are also omitted from the anal-ysis. Thus, the total number of households used in this study is 1557.

3.1. Lakech and Mirte stoves

Various types of improved biomass cook stoves have been dissem-inated in both urban and rural Ethiopia. In this analysis, the two mostcommonly used types of improved stoves, called ‘Mirte improved bio-mass injera stove’ and ‘Lakech charcoal stove’ are discussed; Lakechand Mirte are local words meaning excellent and best, respectively.The Mirte stove, which is made from cement and pumice, wasdesigned by the Ethiopian Energy Studies Research Center in theearly 1990s; one of their goals was to alleviate environmental degra-dation (pollution and deforestation or forest degradation).4 Whenproperly utilized, it serves for approximately 8 years, and is used tocook injera, the staple food of Ethiopia. Injera baking is the mostenergy-intensive activity in Ethiopia, accounting for over 50% of allprimary energy consumption in the country, and over 75% of thetotal energy consumed in households.5 The Mirte stove has been pro-moted and widely distributed in the country, because it can achievefuel efficiency of up to 40% over the open fire stove (Shanko et al.,2009; Yosef, 2007). In addition to improved efficiency, reduced car-bon monoxide (CO) concentration during use is one of the expectedbenefits of the technology (Yosef, 2007).

The Lakech charcoal stove, on the other hand, is made from clay,sand, cement and sheet metal, the latter for cladding. Each Lakechstove is expected to save an average of 75 kg of charcoal per house-hold per year.6 Thus, according to EPA (2004), the Lakech stove yieldsa 25% savings over the traditional open fire stove.7 Furthermore, theEPA report suggests that if all Ethiopian rural and urban households

3 Note that the preparation of injera requires an appliance known as Mitad, a circularclay pan. The electric Mitad is relatively widely used in urban areas.

4 Various organizations were involved in the design and dissemination of both stovetypes. With the objective of addressing severe environmental degradation in the coun-try, Energy for Sustainable Development (ESD), a private UK firm, began working inEthiopia with the Ministry of Mines and Energy under the World Bank-funded BiomassFuels Supply and Marketing Review in 1988. ESD and its Ethiopian counterparts beganwork on designing an improved biomass injera stove in 1990. Other projects financedby the World Bank, such as the Cooking Efficiency Improvement and New Fuels Mar-keting Project (CEINFMP), were also involved in Ethiopia's improved stove programs.These programs were executed by the previous Ethiopian Energy Authority (EEA),which is now the Ethiopian Rural Energy Development and Promotion Centre(EREDPC).

5 See http://www.tve.org/ho/series1/reports_7-12/Mirte_Stoves_Ethiopia.html.6 Retrieved from http://stoves.bioenergylists.org/stovesdoc/Bess/Mirte.htm. According

to Bess (1998), the forest savings from the use of the Lakech was equal to the equivalentof over 2000 ha of important dryland forest in Ethiopia.

7 Studies, such as this one, are generally taken-up in kitchen lab settings, and, there-fore, these estimates are not necessarily accurate, with respect to the use of biomassfuels in household settings.

(approximately 14.44 million) shift to either the improved Lakechor Mirte stove, a savings of about 7,778,800 tons of fuel wood willbe achieved on an annual basis.

Table 2 presents a summary of the adoption rates, although notperiod specific adoption rates, for both stoves by sample region. Ascan be seen, there is not much difference in the average adoptionrates between the two types of stoves. The lack of differentiation be-tween average adoption rates is supported empirically, as well. Forexample, the predicted median time to adoption (from the Weibullmodel) of the Mirte and Lakech stoves is 15.66 and 16.94 years,respectively.

3.2. Analysis variables

Economically, affordability is likely to be an important consider-ation in the adoption decision. Amacher et al. (1992), concur; theirNepal study finds evidence that wealthier households are earlyadopters of improved cook stoves. Similarly, Jan (2012) argues thatan important factor in explaining the use of traditional cooking tech-nologies in rural areas is limited income. In the technology adoptionliterature, income is consistently a significant determinant (Burtonet al., 2003; Fuglie and Kascak, 2001). Therefore, categorical incomedata was included in the analysis. Four categories are available, andare referenced as quartiles for ease of discussion, although the catego-ries are not defined by quartile.

The energy ladder model of household energy demand also em-phasizes the role of income in the choice of modern fuels and technol-ogies for cooking and heating. The central idea of the energy ladderhypothesis is that households use more sophisticated fuels, such askerosene and electricity as their income increases (Mishra, 2008).However, many researchers question the energy ladder hypothesis,as the fuel choice is influenced by many other social and economicfactors (Heltberg, 2005; Jan et al., 2012; Masera et al., 2000). Anothercritical view of the energy ladder hypothesis argues that modern fuels

8 If the household was formed after the introduction of the technology, duration wascalculated from the year the household was formed. However, the survey does nothave any information on the year of marriage or the time the household was formed.Therefore, we chose the year of marriage to be the year in which the household headturned 18, which is the minimum age for marriage according to Ethiopian family law.

9 A few households, in the sample, report purchasing the Lakech charcoal stove be-fore 1991, which should not be; therefore, these households were removed from theanalysis. Moreover, some households do not provide a clear purchase year; thesehouseholds were also removed. A similar strategy was adopted for dealing with theMirte biomass cook stove.

609A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

are often used alongside traditional solid fuels, particularly in ruralareas and amongst the urban poor. Mekonnen and Köhlin's (2008)Ethiopian analysis provides evidence that multiple fuels are used si-multaneously. Although the energy ladder hypothesis argues that in-creases in income will affect household demand for energy type,Barnes et al. (1994) argue that the introduction of improved cookstove technology could be a new step in the energy ladder, lying be-tween traditional biomass stoves and modern fuels and appliances.Therefore, we assume that the Barnes et al.'s (1994) hypothesisholds; wealthier households are able to move up the energy ladderby adopting more efficient technologies. For that reason, we includetwo measures of wealth in the analysis, based on home ownershipand the availability of a separate cooking facility. Although a separatekitchen facility could have the confounding effect of reducing thehealth effects of biomass fuel use, thus reducing the demand for im-proved cook stoves, we hypothesize that the wealth effect dominatesthe health effect, such that households having the ability to access aseparate cooking facility are more likely to adopt improved cookstoves.

Affordability, however, is not only a function of income or wealth.Arif et al. (2011) uncover related financial constraints, finding thatadoption decisions are responsive to price in rural Bangladesh. Al-though all households were asked the price they paid for the stovethey purchased, it is clear that price data is not available for all house-holds; thus, uncovering an appropriate measure of the price of eachstove was a complex task. For the Lakech charcoal stove, it was possi-ble to obtain region-specific prices over the time period from theEthiopian Statistical Agency (ESA). However, the ESA did not collectinformation on the Mirte injera stove, leaving us with no choiceother than to make use of self-reported prices. For those that didnot buy, and, therefore, did not report a price, we used the medianvalue of the price that was paid by those purchasing the Mirtestove. However, it should be noted that the prices for the Mirte injerastove are not accurate reflections of the actual market price, asnon-governmental organizations (NGOs) were incentivizing the pur-chase of improved biomass cookstoves in these regions, which wouldhave resulted in unobserved market distortions. However, regionaldummy variables are also included to control for differences inother factors not directly included in the analysis, such as NGO in-volvement in the local markets. In all cases, prices were deflated to re-flect the real price at the time of purchase.

Another aspect of affordability relates to the costs of the inputs.Because improved cook stoves are technological substitutes for tradi-tional cooking activities – Amacher et al.'s (1992) Nepal study sup-ports this observation – households are more likely to adopt animproved stove when the fuel wood price is higher. Therefore, if pro-grams target areas where fuel wood is freely collected and consumersdo not perceive deforestation as a problem, the program will not besuccessful (Barnes et al., 1994). Instead, programs should targetareas in which fuel wood is perceived to be expensive. Mobarak etal. (2012) similarly suggest that the adoption of improved cook stovescould be increased if fuel-efficient cook stoves were introduced inareas where fuels are not easily available. However, although the sur-vey included questions related to household perceptions and trendsin biomass fuel prices, few households provided answers, and, there-fore, it was not possible to include this data in the analysis. Moreover,the responses related to household perceptions of biomass availabili-ty do not show significant variations; thus, even if these perceptionswere to be included, the results would be insignificant.

Other considerations are also expected to influence adoption deci-sions. Various studies in Asia and Africa, for example, indicate a rangeof factors hindering the dissemination and adoption of improved bio-mass cook stoves. On the supply side, programs were designed thatdid not reflect the interest of consumers, as the focus was primarilyplaced on the potential for improved stoves capable of reducing pres-sure on forests. Potential consumers should, instead, be the focus of

these programs. Slaski and Thurber (2009) suggest that an importantgoal of improved stove programs should be to convince consumersthat there are concrete and observable benefits to the adoption of im-proved cooking technologies, such as the potential to reduce the neg-ative health effects associated with indoor air pollution. Given theevidence on observable benefits described in EPA (2004), Shanko etal. (2009) and Yosef (2007), the ability of consumers to understandand incorporate these benefits is paramount. For that reason, educa-tion is expected to affect the adoption decision. For this analysis, theeducation of the household head is assumed to be associated with in-creased adoption rates. Furthermore, since one of the aforementionedbenefits relates to a reduction in the negative health effects caused bybiomass fuel use, while women and children are most likely to be ex-posed to these negative health effects, female-headed households andhouseholds with young children are expected to adopt more quicklythan either male-headed households or households without children.Therefore, the gender of the household head and the number of chil-dren in the household are included in the analysis. Unfortunately, ac-tual ages for the children are not available, so the best availableinformation is the number of children under the age of 15, ratherthan a much younger age, which would have been preferable.

Finally, individual and household valuations are likely to beinfluenced by the environment in which these stoves are to be used.Along these lines, Muneer and Mohamed (2003) discuss the impor-tance of social systems within the decision-making process, such asthe division of labor within the household and gender relations in so-ciety, rather than relying on generalizations from the literature, whenplanning the dissemination of innovations. Unfortunately, data di-rectly related to the division of labor and gender relations in societywas not available from the survey, or other sources, and, therefore,the analysis cannot contribute to our understanding of theseconcerns.

4. Results and discussion

The focus of this analysis is the length of time it takes a householdto adopt either of the two improved biomass cook stove technologies.For this analysis, the duration start date is defined as the date at whichthe improved biomass stovewasfirst introduced in the area.8 Accordingto a report by Shanko et al. (2009), EREDPC first developed the Mirtestove in the early 1990s, while, according to Bess (1998), commercialproduction of the Lakech stoves began in 1991.9 Also as part of thesurvey, households were asked if they had adopted, and if so, whenthey did adopt. Therefore, the duration end date was determined bythe household's adoption response. For households that had not yetadopted, the duration was right-censored at the year of data collection,2009. Combining these features, the dependent variable is the time thatit takes a household to adopt the new cook stove, which is measured inyears either from household formation or the time that the cook stoveswere first introduced in the region, whichever is later. For the Mirtestove, the introduction date was chosen to be 1994, while the Lakechstove introduction datewas taken to be 1991. In terms of interpretation,reduced time to failure refers to reduced time to technology adoption;the results, below, are interpreted with that feature in mind.

610 A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

4.1. Nonparametric analysis

Before undertaking parametric duration analysis, a simple test ofthe effect of income on the survival rate is performed, using the non-parametric Kaplan–Meier (KM) estimator (Kiefer, 1988), which is theratio of the number of survivors to the number of observations at riskin each time interval. The primary advantage of the KM estimator isthat it can easily accommodate right censoring in the data. Estimationrequires dividing the period of observation into a series of intervals,each containing one or more adoptions at its beginning. Fig. 1,below, illustrates the survival functions for both the Mirte and theLakech stoves by income level, based on the KM estimator. For bothstoves, the survival function for larger income values lies below thesurvival function for lower income values, suggesting that income in-creases the speed of adoption. Both the logrank and Wilcoxon tests,not reported here, confirm this ranking; thus, the speed of adoptionrises with income.

4.2. Results of parametric regressions

The remainder of the discussion focuses on parametric durationanalysis, of which numerous specifications were estimated. The regres-sion results reported follow Eq. (3) to account for right-censored data,while model specification is examined through Akaike's InformationCriterion (AIC) and the Bayesian Information Criterion (BIC). We

A) Mirte Stove

B) Lakech Stove

Fig. 1. Income-based KM survival functions.

present theWeibull, exponential and Coxmodel results for comparisonpurposes. Results from the Weibull–Gamma mixture models are alsoreported to examine the sensitivity of theWeibull model to unobservedheterogeneity. The results for the Mirte stoves are available in Table 4,while the Lakech stove estimates are reported in Table 5.

As noted in Section 3, the exponential model assumes a constanthazard rate, while the Weibull model allows for monotonically in-creasing, monotonically decreasing or constant hazard rates. Givengreater model freedom, it is not surprising that both the AIC and theBIC prefer the Weibull assumptions to all the rest of the models, al-though there is some disagreement between the AIC and BIC Weibulland Weibull–Gamma models for the Mirte stove estimates. Further-more, shape estimates, denoted by P, support the hypothesis that, re-gardless of stove technology, the hazard rate is monotonicallyincreasing, which is not surprising. As technology becomes morewidespread, its use becomes increasingly common. More generally,the results reported across each of the specifications are qualitativelyvery similar, suggesting that the choice of specification does not havea significant impact on the results, at least within this subset of thefamily of proportional hazard models. Given the similarity of results,the remainder of the discussion does not distinguish betweenspecifications.

Economically, we find strong evidence that the Mirte stove is anormal good across our categorical measures of the income distribu-tion. However, although the price estimate is negative, we find eco-nomically small and statistically insignificant price effects. The lackof price effects for the Mirte stove is most likely an indicator ofprice manipulation by NGOs working to incentivize Mirte adoption.Some of the regional differences, discussed below, support that con-tention. The Lakech stove is also found to be a normal good, althoughthe income effect is significant only for those in the second quartile ofthe income distribution. More striking, though, is the sensitivity ofLakech stove adoption to its price. The higher the price, the less likelythe Lakech stove was adopted at any point in time, and these esti-mates are statistically significant.

These results accord with those contained in Barnes et al. (1994);middle-income families have adopted improved stoves far morequickly than poor families in most African countries. On the otherhand, these income results may also indicate that households willnot shift to other, better, sources of energy as their income increases,as postulated by the energy ladder hypothesis, unless we consider thevariant of the energy ladder hypothesis proposed by Barnes et al.(1994). Importantly, Masera et al. (2000) note that the original ener-gy ladder hypothesis does not appropriately account for other factorsthat are likely to affect household switches to modern energy ser-vices, such as: affordability, availability, and cultural preferences.Therefore, since the majority of households that depend on biomassare poor, the design and price of new and improved biomass cookstoves should consider poor household capacity to purchase thenew technology.

In addition to standard economic variables, like price and income,wealth, as measured by home ownership and separate kitchen facili-ties were included in the regression. As Shanko et al. (2009) note,Mirte stove installation and utilization require access to additional fa-cilities, partly due to the fact that it is larger in size thanmanymodernand improved biomass cook stoves. Therefore, it is not surprising thatseparate kitchen facilities and Mirte stove ownership are comple-mentary. Similarly, home ownership is associated with increasedMirte adoption rates, as home ownership signifies a willingness to in-vest in household technology. On the other hand, the Lakech stove issimple and easily mobile, and, therefore, does not require additionalspace. As a result, it is not surprising that home ownership and accessto a separate kitchen are not significant factors in the adoption ofLakech stoves.

The analysis also considered potential substitute technologies. Al-though the electric Mitad is a potential substitute for the Mirte stove,

Table 4Determinants of Mirte cook stove adoption.

Variable Weibull Weibull–Gamma

Exponential Cox PH

HH head sex −0.275⁎⁎ −0.318⁎⁎ −0.249⁎⁎ −0.262⁎⁎

(0.13) (0.15) (0.13) (0.13)HH head age 0.006 0.006 0.010⁎ 0.007

(0.01) (0.01) (0.01) (0.01)HH head read and write 0.578⁎⁎⁎ 0.634⁎⁎⁎ 0.557⁎⁎⁎ 0.570⁎⁎⁎

(0.20) (0.22) (0.20) (0.20)HH head grades 9‐12 1.130⁎⁎⁎ 1.239⁎⁎⁎ 1.020⁎⁎⁎ 1.097⁎⁎⁎

(0.22) (0.25) (0.22) (0.20)HH head high school 0.985⁎⁎⁎ 1.084⁎⁎⁎ 0.908⁎⁎⁎ 0.977⁎⁎⁎

(0.24) (0.27) (0.24) (0.24)Number of children −0.037 −0.042 −0.018 −0.034

(0.03) (0.04) (0.03) (0.03)Number of adults 0.076⁎⁎⁎ 0.087⁎⁎⁎ 0.064⁎⁎⁎ 0.074⁎⁎⁎

(0.03) (0.03) (0.03) (0.03)HH owns dwelling 0.243⁎ 0.297⁎ 0.255⁎ 0.244⁎

(0.13) (0.15) (0.13) (0.13)HH owns electric Mitad 0.119 0.205 0.103 0.132

(0.16) (0.20) (0.16) (0.16)HH owns separatekitchen

0.453⁎⁎⁎ 0.485⁎⁎⁎ 0.435⁎⁎⁎ 0.427⁎⁎⁎

(0.15) (0.17) (0.15) (0.15)Price of Mirte stove −0.032 −0.017 −0.022 −0.025

(0.06) (0.07) (0.06) (0.06)Income: ETB501–ETB1499

0.615⁎⁎⁎ 0.691⁎⁎⁎ 0.578⁎⁎⁎ 0.588⁎⁎⁎

(0.13) (0.15) (0.13) (0.13)Income: ETB1500–ETB2499

0.638⁎⁎⁎ 0.689⁎⁎⁎ 0.592⁎⁎⁎ 0.603⁎⁎⁎

(0.19) (0.22) (0.19) (0.19)Income: aboveETB2500

1.119⁎⁎⁎ 1.298⁎⁎⁎ 0.979⁎⁎⁎ 1.094⁎⁎⁎

(0.23) (0.30) (0.23) (0.23)Tigrai Region −

1.095⁎⁎⁎−1.183⁎⁎⁎ −0.970⁎⁎⁎ ‐1.049⁎⁎⁎

(0.20) (0.22) (0.20) (0.20)Amhara Region −

0.431⁎⁎⁎−0.466⁎⁎⁎ −0.363⁎⁎⁎ −

0.401⁎⁎⁎

(0.12) (0.14) (0.12) (0.12)P 1.192# 1.285#

(0.05) (0.07)AIC 1470.0 1469.1 1896.7 5427.3BIC 1566.2 1570.7 1987.6 5512.9

⁎ Significantly different from 0 at 10%.⁎⁎ Significantly different from 0 at 5%.

⁎⁎⁎ Significantly different from 0 at 1%.# Significantly different from 1 at 1%. Likelihood-ratio test of no heterogeneity: χ1

2=2.88, p=0.045.

Table 5Determinants of Lakech stove adoption.

Variable Weibull Weibull–Gamma

Exponential Cox PH

HH head sex 0.068 0.068 0.082 0.037(0.11) (0.11) (0.11) (0.11)

HH head age −0.034⁎⁎⁎ −0.034⁎⁎⁎ −0.017⁎⁎⁎ −0.033⁎⁎⁎

(0.01) (0.01) (0.00) (0.01)HH head read and write −0.090 −0.090 −0.046 −0.088

(0.14) (0.14) (0.14) (0.14)HH head grades 9–12 −0.095 −0.095 −0.088 −0.065

(0.17) (0.17) (0.17) (0.17)HH head high school 0.143 0.143 0.152 0.131

(0.18) (0.18) (0.18) (0.18)Number of children −0.007 −0.007 0.019 −0.005

(0.03) (0.03) (0.03) (0.03)Number of adults 0.080⁎⁎⁎ 0.080⁎⁎⁎ 0.073⁎⁎⁎ 0.075⁎⁎⁎

(0.03) (0.03) (0.02) (0.03)HH owns dwelling 0.074 0.074 0.064 0.102

(0.11) (0.11) (0.11) (0.11)HH owns metal stove −0.673⁎⁎⁎ −0.673⁎⁎⁎ −0.630⁎⁎⁎ −0.636⁎⁎⁎

(0.11) (0.11) (0.11) (0.11)HH owns separatekitchen

−0.111 −0.111 −0.044 −0.127(0.11) (0.11) (0.11) (0.11)

Price of Lakech stove −0.928⁎⁎⁎ −0.928⁎⁎⁎ −0.521⁎⁎⁎ −1.102⁎⁎⁎

(0.05) (0.05) (0.04) (0.05)Income: ETB501–ETB1499

0.235⁎⁎ 0.235⁎⁎ 0.235⁎⁎ 0.204⁎⁎

(0.11) (0.11) (0.11) (0.11)Income: ETB1500–ETB2499

0.250 0.250 0.236 0.237(0.17) (0.17) (0.16) (0.17)

Income: above ETB2500 0.324 0.324 0.338 0.220(0.22) (0.22) (0.21) (0.22)

Tigrai Region 3.548⁎⁎⁎ 3.548⁎⁎⁎ 2.038⁎⁎⁎ 4.197⁎⁎⁎

(0.24) (0.24) (0.23) (0.26)Amhara Region 1.656⁎⁎⁎ 1.656⁎⁎⁎ 0.984⁎⁎⁎ 1.900⁎⁎⁎

(0.13) (0.13) (0.13) (0.13)P 1.703# 1.703#

(0.04) (0.04)AIC 1088.3 1090.3 2133.1 7013.3BIC 1184.5 1191.9 2224.0 7098.8⁎

⁎ Significantly different from 0 at 10%.⁎⁎ Significantly different from 0 at 5%.

⁎⁎⁎ Significantly different from 0 at 1%.# Significantly different from 1 at 1%. Likelihood-ratio test of no heterogeneity: χ1

2=0.00, p=0.496, but model did not converge.

611A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

there is no empirical support that there is substitution betweenthose technologies, possibly, due to differences in relative costs.Gebreegzihabher et al. (2012), for example, provide evidence thatthe high cost was the main reason that Tigrai households did notadopt the electric Mitad stove, despite the fact that about 80% of sam-ple households in the region used electricity. On the other hand, thehypothesis that the metal charcoal stove is a potential substitute forthe Lakech charcoal stove is supported. Given the better performanceof the Lakech stove over the metal stove, reduced adoption rates forsubstitute stoves, although understandable, implies that additionalpolicies and programs may be needed to increase the rate of adoptionof the technically superior Lakech stove.

In addition to standard economic variables, a few additional re-sults are worth discussing. Education was hypothesized to be associ-ated with increased willingness to adopt newer technologies;however, that hypothesis was only supported in relation to theMirte biomass stove, not with the Lakech stove. Furthermore, we hy-pothesized that female-headed households with many childrenwould favor the adoption of these new cook stove technologies,since both women and children are assumed to be most affected byindoor air pollution. However, this hypothesis was not generallysupported for either stove. Although female-headed households aremore likely to adopt the Mirte biomass cook stove, the sex of thehousehold head is not a significant determinant of the Lakech stove

adoption. Similarly, there is no empirical support for the hypothesisthat the number of children in the household is associated with theadoption of either stove technology. The lack of support for thechild hypothesis could be due to the inability, within the data, to sep-arate very young children from older children.

Finally, as noted above, location variables were included in theanalysis to control for region-specific differences that could be relatedto improved stove adoption, and the results suggest large regionaldifferentiation. The speed of Mirte stove adoption is lower for house-holds in Amhara and Tigrai, compared to those residing in Oromiya,while the opposite is true for the Lakech stove. Since the former re-gions are associated with low levels of biomass, we had expected rel-atively greater adoption rates for both technologies in Amhara andTigrai. However, if there was significant NGO involvement in theOromiya region, and since we do know that Mirte stove purchaseswere subsidized, the negligible price effects estimated for the Mirtestoves could be masked by regional differences in NGO subsidizationprograms, which could have affected the observed regional pattern ofadoption.

5. Conclusions and implications

The heavy dependence and inefficient utilization of biomass re-sources have contributed to the depletion of the forest resources inEthiopia, and the ill health of Ethiopians. Traditional cooking technol-ogies, one source of inefficient utilization of biomass resources, as

612 A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

well as a source of indoor air pollution and ill health, have ledpolicymakers to seek the advancement of affordable alternativecooking technologies that use fewer resources and result in less in-door air pollution. In Ethiopia, two different alternative cook stoveoptions have received the most attention, the Mirte biomass cookstove and the Lakech charcoal stove. Although a number of studieshave shown that these stoves use less biomass, and can, thus, be as-sumed to result in less innocuous health effects, these technologieshave not been universally adopted in Ethiopia.

The purpose, therefore, of this study was to examine potential rea-sons for the lack of adoption of these generally superior technologies.Specifically, the preceding analysis has attempted to shed light on thetiming of biomass cook stove adoption decisions in select urban areasof Ethiopia, based on duration analysis applied to a cross-sectionalsurvey with recall data matched to price information collected fromthe ESA. The results of the analysis support Barnes et al.'s (1994)study implying that energy efficiency might be an intermediate stepalong the road to more modern energy services. Along these lines,both the Mirte and Lakech stove adoption is shown to increase withincome. The survival analysis also supports the contention that ratesof adoption tend to increase, as adoption becomes more widespread.The estimated models suggest that economic incentives, especiallyprices, also correlate, as expected, with adoption, although only sig-nificantly so for the Lakech charcoal stove. The insignificant correla-tion with price for the Mirte stove could be due to the nature of theprice data, which is based on self-reported price information, ratherthan actual market prices, and, therefore, may not fully reflect themarket for these stoves. Furthermore, there is some evidence thattechnology substitution matters. In the case of the Lakech stove,households in possession of a metal stove are much less likely tohave adopted the technologically superior Lakech stove. However,the availability of the electric Mitad alternative does not affect adop-tion rates for the Mirte stove, which could be due to the better perfor-mance of the Mirte stove in reducing the energy cost of preparing thestaple food, injera. Due to data limitations, our analysis could notspeak directly to the reasons as to why households did or did notadopt the various technologies; thus, further analysis is warranted,such that policy makers and/or energy planners can further assessthe potential impact of electric Mitad stoves and other improved bio-mass cook stoves on overall welfare and biomass use.

The adoption of improved stoves is important for many stake-holders, including governmental institutions and non-governmentalorganizations. For example, if richer households adopt more quicklythan poorer households, as shown here, or if prices strongly impactadoption decisions, then the design and dissemination of the stovesshould reflect the interests, preferences and affordability of thesestoves at the level of the household. If, on the other hand, the speedof adoption is affected by the lack of awareness of the potential ben-efits of these stoves – which could not be considered here – differentstrategies could be devised to introduce and disseminate the technol-ogies or educate the population about the benefits of these technolo-gies. Some possibilities include dissemination via demonstrations,posters, and radio or TV advertisements. Furthermore, the analysiscan provide information for stove producers and other stakeholdersregarding the pattern of demand for new stoves and, hence, can begood for production planning. Finally, as already noted, given the im-portance of reducing the current pressure on biomass resources, in-creasing land productivity and reducing the ill effects of indoor airpollution, understanding the determinants of adoption, as well asthe speed of adoption, can provide information that policymakerscan use to increase the speed of adoption, generally.

A few shortcomings associated with the analysis should also benoted. Firstly, given the inability of this study to control for culturalattitudes, differences in perceptions related to the benefits of im-proved cook stove technologies and the underlying prices of bio-mass fuels, future research must give more attention to collecting

information related to perceptions, biomass fuel prices, stove charac-teristics and attitudes on adoption decisions. Secondly, given the as-sumed importance of improved stoves in saving biomass resources,as well as Muneer and Mohamed's (2003) Sudanese evidence of re-bound effects, future research in this area should also address the po-tential for rebound effects, by collecting additional data on biomassfuel consumption across households with different types of cookstoves. Finally, this analysis was based on recall data. As such, manyof the variables used in the analysis were collected at the end of thetime horizon, making it difficult to draw conclusions regarding causaldeterminants of stove adoption. Therefore, future research devoted tothe collection of panel data, allowing for manymore time varying var-iables, possibly including changes in biomass stove incentive pro-grams, has the potential to aid in our understanding of the causalimpacts of policies and other variables on stove adoption.

References

Abou-Ali, H., 2003. Water and health in Egypt: An empirical analysis. Unpublished PhDDissertation, Department of Economics, Gothenburg University.

Amacher, G., Hyde, W., Joshee, B., 1992. The adoption of consumption technologiesunder uncertainty: a case of improved stoves in Nepal. J. Econ. Dev. 17 (2), 93–105.

Arif, T., Ashraf, A., Miller, G., Mobarak, A.M., Akter, N., Ali, A.R.M.M., Sarkar, M.A.Q.,Hildemann, L., Dey, N.C., Rahman, M., Dwivedi, P., Wise, P., 2011. Promotion of im-proved cookstove in Rural Bangladesh, Research and Evaluation Division, BRAC.Working Paper No. 22.

Barnes, D.F., Openshaw, K., Smith, K.R., van der Plas, R., 1994. What makes people cookwith improved biomass stoves? A comparative international review of stove pro-grams. World Bank Technical Report No. 242.

Bess, M., 1998. Mirte stoves in Ethiopia. Energy for sustainable development (ESD), UK.(Retrieved from the internet on February 10, 2007).

Bhattacharya, S.C., Abdul-Salam, P., 2002. Low greenhouse gas biomass options forcooking in the developing countries. Biomass Bioenergy 22, 305–317.

Brookmeyer, R., Corrada, M.M., Curriero, F.C., Kawas, C., 2002. Survival following a diag-nosis of Alzheimer disease. Arch. Neurol. 59, 1764–1767.

Bruce, N., Perez-Padilla, R., Albalak, R., 2002. The Health Effects of Indoor Air PollutionExposure in Developing Countries. World Health Organization, Geneva.

Burton, M., Rigby, D., Young, T., 2003. Modelling the adoption of organic horticulturaltechnology in the UK using duration analysis. Aust. J. Agric. Resour. Econ. 47 (1),29–54.

Cameron, A.C., Trivedi, P.K., 2005. Microeconometrics: Methods and Applications. Cam-bridge University Press, Cambridge, UK.

Cooke-St. Clair, P., Hyde, W.F., Köhlin, G., 2008. Fuelwood, forests and communitymanagement— evidence from household studies. Environ. Dev. Econ. 13, 103–135.

Cox, D.R., 1972. Regression models and life tables (with discussion). J. R. Stat. Soc. Ser. B24, 406–424.

Dadi, L., Burton, M., Ozanne, A., 2004. Duration analysis of technological adoption inEthiopian agriculture. J. Agric. Econ. 55 (3), 613–631.

Emmelin, A., Wall, S., 2007. Indoor air pollution: a poverty-related cause of mortalityamong the children of the world. Chest 132, 1615–1623.

Ethiopian Environmental Protection Agency (EPA), 2004. The Third National Report onthe Implementation of the UNCCD/NAP in Ethiopia.

Ethiopian Environmental Protection Agency (EPA), 2011. Ethiopia's Climate-ResilientGreen Economy Strategy. Prepared for RIO+20 United Nations Conference onSustainable Development, Rio de Janeiro, Brazil.

Fuglie, K.O., Kascak, C.A., 2001. Adoption and diffusion of natural-resource-conservingagricultural technology. Rev. Agric. Econ. 23 (2), 386–403.

Fullerton, D.G., Bruce, N., Gordon, S.B., 2008. Indoor air pollution from biomass fuelsmoke is a major health concern in the developing world. Trans. R. Soc. Trop.Med. Hyg. 102, 843–851.

Gebreegzihabher, Z., van Kooten, G.C., van Soest, D.P., 2005. Land degradation inEthiopia: what do stoves have to do with it? University of Victoria WorkingPaper 2005-16.

Gebreegzihabher, Z., Mekonnen, A., Kassie, M., Köhlin, G., 2012. Urban energy transi-tion and technology adoption: the case of Tigrai, Northern Ethiopia. Energy Econ.34, 410–418.

Geist, H., Lambin, E., 2003. Is poverty the cause of deforestation? Int. For. Rev. 5 (1),64–67.

Gutierrez, G.R., 2002. Parametric frailty and shared frailty survival models. Stata J. 2 (1),22–44.

Handa, S., Koch, S.F., Ng, S.W., 2010. Child mortality in eastern and southern Africa.Popul. Rev. 49 (1), 8–35.

Heltberg, R., 2005. Factors determining household fuel choice in Guatemala. Environ.Dev. Econ. 10, 337–361.

International Energy Agency (IEA), 2002. World Energy Outlook. OECD, Paris.Jaggia, S., 1991. Specification tests based on the heterogeneous generalized gamma

model of duration: with an application to Kennan's strike data. J. Appl. Econ. 6,169–180.

Jan, I., 2012. What makes people adopt improved cookstoves? Empirical evidence fromrural northwest Pakistan. Renewable Sustainable Energy Rev. 16 (5), 3200–3205.

613A.D. Beyene, S.F. Koch / Energy Economics 36 (2013) 605–613

Jan, I., Humayun, K., Hayat, S., 2012. Determinants of rural household energy choices:an example from Pakistan. Pol. J. Environ. Stud. 21 (3), 635–641.

Kennan, J., 1985. The duration of contract strikes in U.S. manufacturing. J. Econ. 28,5–28.

Kiefer, N.M., 1988. Economic duration data and hazard functions. J. Econ. Lit. 26,646–679.

Lavy, V., Strauss, J., Thomas, D., Vreyer, P., 1996. Quality of health care, survival andhealth outcomes in Ghana. J. Heal. Econ. 15, 333–357.

Lee, S.H., 2003. A duration analysis of the adoption of privatization policy: evidence fromthe telecommunications sector in the Asia-Pacific region. Int. Econ. J. 17 (2), 65–78.

Locatelli, F., Marcelli, D., Conte, F., D'amico, M., Del Vecchio, L., Limido, A., Malberti, F.,Spotti, D., 2001. Survival and development of cardiovascular disease by modalityof treatment in patients with end-stage renal disease. J. Am. Soc. Nephrol. 12,2411–2417.

Masera, O.R., Saatkamp, S.D., Kammen, D.M., 2000. From linear fuel switching to mul-tiple cooking strategies: a critique and alternative to the energy ladder model.World Dev. 28 (12), 2083–2103.

Masera, O.R., Edwards, R., Armendáriz, C., Berrueta, V., Johnson, M., Rojas, L., Riojas-Rodríguez, H., 2007. Impact of “Patsari” improved cookstoves on indoor air qualityin Michoacan, Mexico. Energy Sustainable Dev. 8 (3), 53–59.

Mekonnen, A., Köhlin, G., 2008. Biomass fuel consumption and dung use as manure:evidence from rural households in the Amhara Region of Ethiopia. Resources Forthe Future Discussion Paper dp-08-17-efd.

Mishra, A., 2008. Fuel for clean energy debate— a study of fuelwood collection and pur-chase in rural India. South Asian Network for Development and EnvironmentalEconomics. Policy Brief, Kathmandu.

Mobarak, A.M., Dwidevi, P., Bailis, R., Hildenman, L., Miller, G., 2012. Low demand fornontraditional cookstove technologies. Proc. Natl. Acad. Sci. 109 (27), 10815–10820.

Muneer, S.E.T., Mohamed, E.W.M., 2003. Adoption of biomass improved cook stoves ina patriarchal society: an example from Sudan. Sci. Total. Environ. 307, 259–266.

Nepal, M., Nepal, A., Grimsrude, K., 2010. Unbelievable but improved cookstoves arenot helpful in reducing firewood demand in Nepal. Environ. Dev. Econ. 16, 1–23.

Shanko, M., Abebe, T., Lakew, H., 2009. A report on Mirt biomass injera stove marketpenetration and sustainability study in Amhara, Oromiya and Tigrai NationalRegional States. GTZ Sun Energy.

Slaski, X., Thurber, M., 2009. Cookstoves and obstacles to technology adoption by thepoor. Program on Energy and Sustainable Development Stanford University WorkingPaper 89.

Smith, K.R., Mehta, S., 2003. The burden of disease from indoor air pollution in developingcountries: comparison of estimates. Int. J. Hyg. Environ. Health 206 (4–5), 279–289.

Smith, K.R., Mehta, S., Maeusezahl-Feuz, M., 2004. Indoor air pollution from solid fueluse. In: Ezzatti, M., Lopez, A.D., Rodgers, A., Murray, C.J.L. (Eds.), ComparativeQuantification of Health Risks: Global and Regional Burden of Disease Attributableto Selected Major Factors. World Health Organization, Geneva, pp. 1435–1493.

Sorrell, S., Dimitropoilos, J., Sommerville, M., 2009. Empirical estimates of the directrebound effect: a review. Energy Policy 37, 1356–1371.

Tasleem, A., Zahoor, U., Mir, H.K., Nazli, R., 2007. Chronic bronchitis in women usingsolid biomass fuel in rural Peshawar, Pakistan. Chest 06 (2529), 1472–1476.

Vance, C., Iovanna, R., 2006. Analyzing spatial hierarchies in remotely sensed data:insights from a multilevel model of tropical deforestation. Land Use Policy 23(3), 223–238.

World Health Organization (WHO), 2002. World Health Report: Reducing Risks,Promoting Health Life. World Health Organization, Geneva.

World Health Organization (WHO), 2006. Fuel for Life, Household Energy and Health.World Health Organization, Geneva.

World Health Organization (WHO), 2009. Global Health Risks: Mortality and Burden ofDisease Attributed to Selected Major Risks. World Health Organization, Geneva.

Yosef, A.F., 2007. Assessing environmental benefits of Mirt stoves with particular refer-ence to indoor air pollution (carbon monoxide and suspended particulate matter)and energy conservation. Maters of Science Thesis, Addis Ababa University.