1. Introduction
In developing countries, biomass remains the most widely used fuel type. Firewood comprises the biggest share of it. Burning biomass for household energy is mainly related to indoor air pollution that causes numerous health problems, emission of greenhouse gases, brown clouds, and black carbon (Edwards et al., Reference Edwards, Smith, Zhang and Ma2004; Chengappa et al., Reference Chengappa, Edwards, Bajpai, Shields and Smith2007; Pant, Reference Pant2008; Malla, Reference Malla2009; Gustafsson et al., Reference Gustafsson, Krusa, Zencak, Sheesley, Granat, Engstrom, Praveen, Rao, Leck and Rodhe2009). Deforestation and climate change are interlinked where deforestation combined with natural decaying of biomass is contributing more than 17 per cent of the CO2 emissions globally (IPCC, 2007). While the debate on the link between fuelwood use and deforestation is far from over (e.g., Arnold et al., Reference Arnold, Kohling and Persson2006), scholars have attributed one major reason for deforestation to the extraction of wood for fuelwood, charcoal, polewood, and the commercial harvesting of forest products (e.g., Geist and Lambin, Reference Geist and Lambin2002).
Two successive Nepal Living Standards Surveys (NLSS I and II) indicate that the use and collection of firewood seems to be increasing over the years in Nepal. In 1995–1996, about 77 per cent of households reported that they were using firewood for cooking and 84 per cent of those firewood users collected it (CBS, 1996). However, in 2003–2004, about 84 per cent of households use firewood for cooking, and 88 per cent of these firewood user households collect it (CBS, 2004). Such a widespread and increasing trend in firewood collection and use may have two potential impacts: it may well threaten the sustainability of Nepal's forests resources and also cause negative health impacts due to indoor air pollution (Pant, Reference Pant2008; Malla, Reference Malla2009) given that a very small number of households are using the improved cookstoves (ICSs).
Historically, deforestation in Nepal is due mainly to expansion in agriculture, illegal timber extraction, and firewood collection (Bajracharya, Reference Bajracharya1983). To avoid massive deforestation, the Nepal government started transferring the user rights of the government managed forests to the local communities to manage locally under the widely celebrated community forestry program. While transferring the forest to the local communities during the 1980s, the Nepal government also tried to distribute ICSs on a limited scale, hoping that these cookstoves would be able to reduce firewood demand. In principle, alternative energy sources, such as electricity, solar, bio-gas, and so on, can be used in place of dirty energy, such as firewood. However, replacing the current use of firewood by such alternative energy sources may not be feasible given the current state of electricity generation and its coverage,Footnote 1 and unavailability of relatively cleaner and more affordable energy sources in Nepali villages.Footnote 2 One possible solution would be to adopt cooking stoves that would consume less firewood. The expectation is that such energy efficient cooking stoves might help to reduce firewood consumption and reduce the emissions at the household level. The ICS technology was first introduced in Nepal in the 1950s, and its use has been on the rise since the latter part of the 1990s. The NLSS data indicate that the ICS program has country-wide coverage that households in 31 out of 75 districts have reported the use of ICS in the 2003–2004 survey.
This paper investigates the impact of different types of cookstoves on firewood consumption in Nepal. More specifically, we want to investigate whether households with improved stoves use less firewood than households with the traditional mud stove or the open-fire stove. Many types of cookstoves are in use throughout Nepal. If an improved stove consumes less firewood than a mud or open-fire stove while meeting the household's energy need, then the adoption of such a less firewood consuming cookstove might help to maintain the sustainable use of firewood. This might be helpful in reducing indoor air pollution and lowering the pressures on the existing forests resources.
We use data from the Nepal Living Standards Survey 2003–2004 for empirical analysis. This is the most recent and the comprehensive household survey in Nepal. It provides socioeconomic and demographic information about the households, such as income, consumption, firewood collection, stove types, health, education, and so on. For this analysis, household level firewood consumption is used as the outcome variable. After controlling for different variables, such as average time to collect one bhari Footnote 3 of firewood, household income, household size, number of animals (cows and buffalos), we find that the type of cookstove significantly affects the amount of firewood demand. More specifically, we find that the improved stoves are not really improved in terms of firewood saving. Households with traditional mud stoves seem to consume less firewood than households that use the so-called ‘improved’ stoves. The results are not significantly different across different functional forms and different estimation methods after correcting for endogeneity in stoves adoption.
The paper is organized as follows. In the next section, we provide a short review of the related literature. Section 3 presents a brief history of ICS in Nepal. A basic theory, the econometric model, and the hypotheses of the study are presented in section 4. Data and the variables used in this paper are discussed in section 5. Regression results are presented and discussed in section 6. The final section concludes.
2. A brief overview of existing studies
Despite the widespread use of firewood in developing countries and its potential impacts on indoor air pollution and deforestation, the literature on the contribution of improved stoves to firewood saving is very limited. Cooke et al. (Reference Cooke, Kohlin and Hyde2008) provide a review of empirical literature from three developing countries on the adoption of ICSs. They document that the empirical evidence is inconclusive regarding the role of ICSs on fuelwood demand. Zein-Elabdin (Reference Zein-Elabdin1997) gives two reasons for the scarcity of the research on the impact of improved stoves on fuelwood demand in developing countries: lack of databases and lack of understanding of the market for traditional fuels, such as firewood.
Using time-series data, Zein-Elabdin (Reference Zein-Elabdin1997) estimates the demand and supply elasticities for charcoal in Khartoum (Sudan) in order to analyze the rebound effect of new technology adoption. The rebound effect is the behavioral response that offsets the beneficial effect of new technology. For example, if a new technology improves stove efficiency by 20 per cent and fuelwood consumption drops by 15 per cent, then the 5 per cent differential between the fuel efficiency and the decrease in fuelwood demand may be due to an increase in consumption of fuelwood as a result of efficient stoves. The author indicates that such a ‘rebound effect’ gets larger if a household's budget share on fuelwood is large, the income elasticity of fuelwood demand is high, and the supply elasticity is low. Zein-Elabdin's (Reference Zein-Elabdin1997) study is based on the market data in an urban environment. The author finds increased fuelwood consumption (charcoal) with efficient stoves. This is because the income elasticity of fuelwood demand is very high in the case study where efficient stoves reduce the demand for fuelwood in the first place. Lower demand, however, reduces the price of charcoal. As a result, households adopt cheaper fuel in alternative uses leading to overall higher demand for fuelwood.
Other studies have shown that the use of fuel-efficient cooking stoves could reduce the demand for fuelwood. Studies from China (Edwards et al., Reference Edwards, Smith, Zhang and Ma2004), Guatemala (McCracken et al., Reference McCracken and Smith1998), India (Chengappa et al., Reference Chengappa, Edwards, Bajpai, Shields and Smith2007), Madagascar (Bazile, Reference Bazile2001), Mexico (Masera et al., Reference Masera, Edwards, Arnez, Berrueta, Johnson, Bracho, Rioja-Rodriguez and Smith2007), and Tanzania (Makame, Reference Makame2007) show mixed impacts of the improved stoves on fuelwood saving. However, the diffusion of the new technology is not smooth in most of these countries except in China. The main reason for slow or no diffusion of cookstoves is attributed to the stove designs that sometimes do not meet the cooking requirements of the households.
Sinton et al. (Reference Sinton, Smith, Peabody, Yaping, Xiliang, Edwards and Quan2004) provide a detailed account of how the diffusion of the ICS became successful in China. On the other hand, Barnes et al. (Reference Barnes, Openshaw, Smith and van der Plas1993) provide an excellent review of diffusion of the improved stoves and provide an explanation of why households remain reluctant to adopt the new technology in many developing countries. This review highlights the importance of responding to the specificity of the local conditions when introducing a new technology. Basically, the authors claim that while the scarcity and higher price of fuelwood may help the dissemination of the improved stoves, their adoption is not guaranteed. In a related review paper, Barnes et al. (Reference Barnes, Openshaw, Smith and van der Plas1994, p. v) summarize the conditions for the success or failure of the ICSs adoption as, ‘no matter how efficient or cheap the stove, individual households have proved reluctant to adopt it if it is difficult to install and maintain or less convenient and less adaptable to local preferences than its traditional counterpart’. The emphasis is on the fact that the improved stove should not be totally different from what households have been using; it should be improved but not totally new to the users. Using survey data from two rural villages in Nepal, Amacher et al. (Reference Amacher, Hyde and Joshee1992) find that household characteristics, income level, and the availability of firewood determine the adoption of the improved stoves.
A related area of research is the contribution of improved stoves on reducing indoor air pollution and time saving. McCracken and Smith (Reference McCracken and Smith1998) conducted a study in the Guatemalan highlands, where they compared the thermal efficiency, emission content, and the length of cooking time between the traditional three-stone firestove and an improved Plancha fire stove. Their experiment revealed that while the improved stoves emitted less PM2.5 (fine air pollutant particles that are 2.5 μm in diameter or smaller), they took more time to prepare food as compared to the open-fire stoves.
In a study conducted in China, Edwards et al. (Reference Edwards, Smith, Zhang and Ma2004) find that the thermal efficiency that reduces fuel demand mainly leads to combustion inefficiency causing increased greenhouse pollutants and health damaging particulates. Chengappa et al. (Reference Chengappa, Edwards, Bajpai, Shields and Smith2007) conducted a study in India similar to what Edwards et al. (Reference Edwards, Smith, Zhang and Ma2004) did in China. They find that the improved stove reduces the PM2.5 concentration and CO, improving the indoor air quality. Masera et al. (Reference Masera, Edwards, Arnez, Berrueta, Johnson, Bracho, Rioja-Rodriguez and Smith2007) find the same type of results in the case of the Mexican version of the improved stove. Pant (Reference Pant2008) and Malla (Reference Malla2009) study the health impact of indoor air pollution using small surveys in a few Nepali villages. Both of these studies find that the improved stoves help to reduce the respiratory diseases significantly. Malla's study that covers five villages from one district (400 households) indicates that the improved stoves can help reduce the use of firewood. Pant's study that covers six villages from two districts (600 households), however, does not analyze the fuelwood efficiency of the cookstoves.
Results from these studies show that the evidence on the impact of improved stoves on firewood saving, indoor air pollution, and time saving remains inconclusive. However, most of these studies focus mainly on the technical aspects of the stove design, using small sample sizes with limited geographical coverage. Our study analyzes the impact of the stove type on household level firewood consumption with a nationally representative household survey.
3. Improved cookstoves in Nepal
The ICS program in Nepal was introduced in the mid 1950s, but its coverage remained very low during the next three decades. The main objectives of the program were to reduce the rate of deforestation along with reducing indoor air pollution and increasing the efficiency of household energy use. During the 1980s, the Nepal government tried to tie in the ICS program with the community forestry program in the hope of containing massive deforestation (Clemens et al., Reference Clemens, Rijal and Takada2010). In 1999, the Nepal government introduced the National ICS Program under the Energy Sector Assistance Program funded by the Danish International Development Agency. At the national level, the Alternative Energy Promotion Center has been administering the program with the help of several organizations, such as the Center for Rural Technology, Nepal (CRT/N), the Department of Women's Development, and other nongovernmental and community-based organizations.
The ICS program in Nepal is mostly supply driven. Interested organizations have been introducing several types of ICS technologies with different shapes and sizes, such as mud–brick ICS and metallic ICS (Practical Action, 2009). The most popular one is the mud–brick ICS. It can have one to three pot-holes, depending on a household's requirement. By 2004, the number of ICS in the country was estimated to be 150,000, where CRT/N alone distributed 100,000 mud–brick ICS (Clemens et al., Reference Clemens, Rijal and Takada2010). The NLSS data suggest that while the use of ICS covers all geographical areas, its rate of adoption is very low, with only 2 per cent of households in 31 districts (out of 75) using ICS during 2003–2004. Though not identified in the NLSS data set, secondary information indicates that a very few user households adopt the metallic cookstoves with a damper to regulate the air (Practical Action, 2009).
4. Basic theory, econometric model, and hypotheses
Assume that a representative household derives utility from the consumption of energy (both for heating and cooking), consumption of all other goods, and leisure time. The utility level also depends on household characteristics. A household can get energy from firewood or from alternative sources, such as petroleum products (e.g., kerosene and LPG). The firewood market is very thin in Nepal. Very few households actually buy firewood in the market; most households collect it rather than buying from the market. Therefore, a market price for firewood is nonexistent for most of the households. However, the time spent collecting firewood has an opportunity cost. The opportunity cost could be the returns on household labor. But such information is not available for many households. Thus, we use the collection time as a proxy for firewood price.
Maximizing a household's utility function subject to the budget constraint yields the demand for firewood (see Köhlin and Parks, Reference Köhlin and Parks2001; Palmer and MacGregor, Reference Palmer and MacGregor2009; Baland et al., Reference Baland, Bardhan, Das, Mookherjee and Sarkar2010 for more discussions on analytical models). For household i at community j, the firewood demand (FWij) is a function of collection time (CTij), household characteristics (HCij), and community characteristics (CCj). The firewood demand also depends on the efficiency of the stove that the household uses. Stove efficiency depends on its type (STij). Based on this premise, the reduced form equation for firewood demand is given by the following expression:
There are five types of cookstoves reported in the survey: open-fire stove (i.e., the traditional tripods type), mud stove, improved stove, kerosene stove, and other. Since the ‘other’ category is not specified and has very few observations, we use the four types of stove for the analysis.Footnote 4 Out of these four stove types, we use the traditional open-fire stove for comparison. Since the open-fire stove requires more firewood, researchers consider it as the most inefficient of all stove types (see Edmonds, Reference Edmonds2002). The open-fire stove has poor heat transmission because the fire is open in all directions. In comparison, the mud stove is more enclosed, which can transfer more heat energy for cooking. Since we expect other varieties of stoves to be more efficient (i.e., require less fuel) than the open-fire stove, we expect that households would use less firewood if they adopt mud, or improved stoves. Statistically, we expect negative coefficients of these stove types (βr < 0). Additionally, we expect that households with improved stoves will consume less firewood than households with mud stove. So, the coefficient of the mud stove is expected to be smaller in absolute terms than the coefficient of the improved stove.
If firewood is a normal good, then an increase in household income should increase the demand for firewood. However, firewood is also the most dirty fuel type, generating smoke that causes indoor air pollution. Therefore, with an increase in income a household may want to replace the dirty fuel with cleaner ones. Since the collection time (CTij) is a proxy for the price of firewood, we expect β4 < 0.
Another group of control variables is a vector of household characteristics (HCij), which includes household size (HHSIZE), gender of the household head (MALE), and the presence of a child below six years (CHILD) of age. The rationale for using these control variables is that household size controls for the scale factor, i.e., we expect a positive relationship between household size and firewood demand. As women are mostly responsible for collecting firewood (see Amacher et al., Reference Amacher, Hyde and Joshee1993; Khare et al., Reference Khare, Sarin, Saxena, Palit, Bathla, Vania and Satyanarayana2000), the presence of a small child (below six years) in the house may affect the amount of time they can spend on firewood collection.
We also control for the number of big-head animals (cows and buffalos) as households generally use these animals for milk production (female) and for plowing agricultural fields (males). Most of the households feed these animals cooked or warm feed (i.e., a mixture of water, salt, leftover food, rice husk) at least once in a day. So, we expect a positive effect of number of such animals on firewood demand. We also control for sources of firewood – community forest, government forest, and private forest. Since the Government of Nepal initiated the community forestry program with the objective of supplying more forest products (such as firewood and fodder) while conserving the forest, we expect households with access to a community forest to collect and consume more firewood compared to access to a government-controlled forest or a private forest.
The literature on the adoption of improved stoves (e.g., Amacher et al., Reference Amacher, Hyde and Joshee1992; Makame, Reference Makame2007) indicates that adoption rates mostly depend on the availability of firewood. It may also depend on knowledge about the harmful effects of indoor air pollution or the benefits of the improved stoves, and the location of households. In that sense, the type of stove a household uses could well be endogenously and jointly determined with the amount of firewood a household would demand. Therefore, we instrument the stove-type variable using the ordered logit model as we have four different types of stoves in our sample.
Stove type may well depend on cultural factors (such as cooking habits), location (whether hilly or mountain area vs. southern plain area), and the level of education along with the availability of firewood. We, therefore, use the following four variables as a set of instruments: Brahmin/Chhetri (culture), mountain/hills (location), remittance received, and education.
Here, the first three variables are the indicator variables, and the last one is household head's years of schooling. The variable related to the cultural factor controls for the cultural traditions of the Brahmin and Chhetri, which are the two dominant high castes in Nepali Hindu society. These two castes generally use the mud stove for cooking their main food items (mostly lunch and dinner), which is generally not the case for other castes/ethnic groups. The reason for including the remittance dummy as an instrument is that when a household member goes away for work and sends money back home, such money may come with attached information. What this means is that the person sending the money may also provide information regarding the negative impact of indoor air pollution to the family back home if the household is using firewood for household energy.
Finally, the choice of stove also depends on the location of the households. In Nepal, if households are located in the hills or in the mountain region, they mostly use the open-fire stove to keep themselves warm during winter as well as to dry food grains and other items, such as chilies, during both the rainy summers and the cold winters. During these seasons, the days are either shorter or the sunshine is rare and not enough to dry food grains or other items. The open-fire stove is convenient for these purposes as this type of stove is mainly placed in the middle of the living room so that household members can sit around it to avoid the cold. Above the open-fire stove, households fix a layer underneath the ceiling with enough space to put food grains or other items for drying.
The mud stove, on the other hand, is mostly placed in a corner of the living room, which prevents household members from sitting around it to avoid the cold, and this kind of stove is not very useful for drying food grains or other items as the heat transmits mostly toward the walls rather than directly toward the ceiling.Footnote 5 Given these unique features regarding the locational choice for placing different stove types, we use the mountain/hill dummy as an additional instrument. We also perform a statistical test in order to suggest that our instruments are not directly affecting the firewood demand.
5. Data and variables
We use the Nepal Living Standards Survey 2003–2004 (also called NLSS-II) for the empirical analysis. This survey follows the World Bank's Living Standard Measurement Survey (LSMS) methodology. This is the most recent and comprehensive household survey in Nepal that provides socioeconomic and demographic information about the households (such as income, consumption, labor market, firewood demand, collection time, education). This multitopic survey consists of a nationally representative sample of 3912 households from 326 primary sampling units (PSUs). These PSUs are selected from six strata using probability proportional to size (PPS) sampling, where size is measured based on the number of households. From each PSU, 12 households are selected systematically (CBS, 2004).
As the main objective of this paper is to investigate the impact of stove types on household-level firewood consumption, we include only a subsample of households who collect and use firewood for household energy. We define these households as firewood self-sufficient. In other words, we exclude those households from our analysis who buy firewood, or who collect firewood for selling purposes, and keep only those households who collect and use firewood. In the sample, these self-sufficient households are in the majority, and we use firewood collection time as a proxy measure of unit price (as in Amacher et al., Reference Amacher, Hyde and Joshee1993). Additionally, 13 households who reported collecting unusually high amounts (more than 30 bharis) of firewood per month are also excluded. With such adjustments, our sample size came down to 2606 observations. Figure 1 presents a box plot of the distribution of firewood demand by stove type.
Figure 1. Box plot of firewood use by stove type.
Table 1 presents the descriptive statistics of the variables used in this paper. FIREWOOD indicates the average monthly firewood collection by households, measured in the local unit bhari. On average a household collects about seven bharis of firewood per month. The average time to collect one bhari of firewood (COLLECTIONTIME) is less than 4 h. The average per capita household income for the subsample of firewood self-sufficient households is Rs 13,650.
Table 1. Variable definition and descriptive statistics
Note: bhari is a local unit of firewood measurement, where 1 bhari = 30 kg (average); no. of observations = 2606. This subsample includes households who collect and use firewood. We have excluded a few households who collected more than 1 bhari of firewood (average) per day as outliers.
Source: Nationally representative household survey data collected by the Central Bureau of Statistics, Nepal, popularly known as the Nepal Living Standards Survey, 2004.
The next four variables are the types of stove used by the households. In our sample, more than 38 per cent of households use the most traditional OPENFIRESTOVE, 56 per cent of households use MUDSTOVE, and 2 per cent of households use the IMPROVEDSTOVE. About 4 per cent of households use the KEROSENESTOVE.
Other variables used for the analysis are household size (HHSIZE), the gender of household head (MALE), the number of cows (COWS) and buffalos (BUFFS) the household owns, the place where the households collect the firewood (community, government, or private forest), and an indicator of the presence of a child below six years of age (CHILD). The last four variables are used as the instruments. In our sample, about one third of households receive remittances, and the same fraction of households belongs to the upper caste (Brahmin or Chhetri). About 61 per cent of households are from the hill/mountain region and the average years of schooling of the household head is less than three years, meaning that the majority of household heads are either illiterate or have very low education.
Table 2 presents the distribution of the same variables by stove types, which is our main variable of interest at the household level. This table indicates that households with open fire or improved stoves collect more firewood compared to households with the mud stoves.
An alternative to firewood collection time as the unit price of firewood could be the opportunity cost of firewood collection time. When the wage rate of hired farm workers increases, the opportunity cost of collecting firewood also goes up for the rural agricultural economy, thus drawing labor away from firewood collection. For the firewood self-sufficient households, the increased productivity of agricultural labor (in terms of the higher wage rate) leads to reduce firewood collection (Palmer and MacGregor, Reference Palmer and MacGregor2009). In our data set, a subsample of 756 households hired female farm workers, paying cash or in-kind wages. We use the wage rate of these hired female farm workers multiplied by the collection time as the opportunity cost of firewood collection. Table 3 presents summary statistics of this subsample by stove type.
Table 3. Distribution of variables by stove type for the households hiring female farm workers (N = 756)
Source: See table 1.
6. Results and discussion
In this section, we present and discuss the regression results. In our model, we use the dependent variable, FIREWOOD, in two different ways: in level (table 4) and in logs (tables 5–7). Similarly, we use COLLECTIONTIME and PCINCOME in level as well as in logs in order to see how sensitive the results are to the choice of the functional forms.
Table 4. Ordinary least-squared regression results
Notes: Standard errors in parentheses. ap < 0.01, bp < 0.1, cp < 0.05.
Table 5. Two-stage least-squared regression results
Notes: Standard errors in parentheses. ap < 0.1, bp < 0.01, cp < 0.05,.
Table 6. OLS and 2SLS regression results (Dep Var: Log(FIREWOOD) where COLLECTIONTIME is replaced with FEMALEWAGE)
Notes: Standard errors in parentheses. ap < 0.01, bp < 0.05, cp < 0.1.
Table 7. Fixed effect regression (Dep Var: Log(FIREWOOD))
Notes: Standard errors in parentheses. ap < 0.01, bp < 0.05, cp < 0.1.
We estimate ordinary least squared (OLS), two-stage least squared (2SLS), and fixed effect (FE) models in order to see whether the results are sensitive to a particular estimation method. Firewood collection time could be endogenous (Palmer and MacGregor, Reference Palmer and MacGregor2009). The best way to confirm such endogeneity is to perform the Hausman specification test. However, the NLSS data are collected using a two-stage stratified sampling method where our observations are clustered within the PSU. These PSUs are selected based on the PPS (CBS, 2004). While the Hausman specification test requires one of the estimators to be efficient, such clustered and p-weighted (or PPS) sample observations violate the requirements for the Hausman specification test (StataCorp, 2009).
Another issue is the difficulties in finding convincing exclusion restrictions (i.e., good instruments) in the data set for the COLLECTIONTIME that are not correlated with FIREWOOD demand. Therefore, we estimate alternative models using the wage rate of hired female farm workers multiplied by the collection time as a proxy for the opportunity cost (OPPORTUNITYCOST) of collection time. The wage rate of hired farm workers is exogenous to the households who hire such labor. This alternative measure of the opportunity cost comes, however, with a significant cost in terms of the smaller sample size of 756 observations, but it is worth doing to find the sensitivity of our results. Earlier, with the COLLECTIONTIME as a proxy for the unit price of firewood, the sample size was 2606 observations.
6.1 Ordinary least-squared results
Table 4 displays the regression results where the dependent variable, FIREWOOD, is in level. The first two columns (Models I and II) display the results from the OLS regression where the signs of the coefficients of all stove categories (mud, improved, and kerosene) are negative. While estimating these models, we also take into account the clustering and weighting issues of the NLSS data. The results indicate that households that use the mud stove, the improved stove, or the kerosene stove demand less firewood as compared to the traditional open-fire stove users. The results also indicate that the IMPROVEDSTOVE is comparatively less effective in terms of reducing firewood demand as the coefficient of the improved stove is smaller in absolute value than the coefficients of other stove types. Additionally, the coefficient of the improved stove is insignificant indicating that households with the improved stoves or open-fire stoves may use statistically comparable amounts of firewood per month, ceteris paribus.
Results in table 4 indicate that a household with the MUDSTOVE could consume 0.88 bhari less firewood per month, or over 10 bharis less firewood in a given year, compared to a household with the traditional open-fire stove. In the case of the KEROSENESTOVE, the monthly firewood saving by a household is about 3.6 bharis per month or about 43 bharis per year on an average. An interesting result here is the sign of the coefficient of firewood COLLECTIONTIME. As a proxy of firewood price, its coefficient is expected to be negative. Though not significant, it is positive. This result, however, is consistent with some other studies (e.g., Malla, Reference Malla2009). Such a positive price effect may indicate that firewood could be a Giffen good where the demand curve slopes upward. The positive sign of the coefficient of firewood collection time indicates that, for the self-sufficient firewood user households from rural Nepal, no cheaper substitutes are available for firewood that they can use when the collection time gets higher.Footnote 6 The per capita household income has no significant effect on firewood demand.
As expected, household size has a positive and significant effect on firewood consumption. Other variables with positive and significant coefficients are the number of cows and buffalos that households have. The positive coefficient of community forest indicates that households collect more firewood from the community forests as compared to what they collect from the private or the other forests. This is expected, as one of the main objectives of the community forestry in Nepal is to increase the availability of the forest products, such as firewood, to the local communities (Kanel, Reference Kanel2004).
For the robustness check, we added CHILD, an indicator variable for the presence of a child below six years old in the family. Since firewood collection in Nepal is done mostly by women (Amacher et al., Reference Amacher, Hyde and Joshee1993), the presence of young children may limit their ability to go out and collect more firewood. As expected, the sign of the coefficient of CHILD is negative. The signs and the significance of the coefficients of other variables in Models-I and II do not change when this new variable, CHILD, is added.
While the coefficient of COLLECTIONTIME is positive, we expect firewood collection time to act as a brake on firewood collection after a certain point. In other words, we expect a nonlinear relationship between firewood collection time and firewood demand. One way of addressing this issue is to get log transformation of these variables. So, we replace COLLECTIONTIME with its log transformation in Models III and IV. After such changes, the results are comparable with the first two models except that the sign of the coefficient of the IMPROVEDSTOVE is now positive, but still insignificant. These results indicate that on average the MUDSTOVE could save about 14.3 per cent of firewood while the KEROSENESTOVE could save up to 44.2 per cent of firewood compared to the traditional open-fire stove.Footnote 7 However, households with the IMPROVEDSTOVE and open-fire stoves could consume comparable amounts of firewood as there is no significant difference in firewood demand between mud-stove user and improved-stove user households.
6.2 Two-stage least-squared results
As discussed earlier, the type of stove households use may be endogenously determined depending upon the availability of firewood. If firewood is abundant, households may use an open-fire stove, and if no or less firewood is available, then a family may use some other alternatives, such as improved stoves or kerosene stoves. We use a two-stage (instrumental variable) estimation approach to address the issue of endogeneity.
As discussed earlier, the choice of the instruments is based on the premise that stove adoption may depend on cultural factors (Brahmin/Chhetri), location of households (mountain/hills), the knowledge of available options, the benefits of using alternative stove types, and the harmful effects of indoor air pollution (education) and the link to the urban centers and beyond in terms of remittance flow. As we have four different types of stoves, we use ordered logit for the first-stage regression.
We test the exclusion restriction for the proposed set of instruments as follows. First, we run regression of FIREWOOD on all explanatory variables plus the set of instruments by stove types (open-fire stove and mud stove, two widely used stove types). Second, we test the joint hypothesis that the set of instruments has no joint significant effect on FIREWOOD demand. As we estimate four 2SLS models (see table 5), we perform the joint significant tests for all four cases. The F-statistic ranges from 0.55 to 1.77 with p-value 0.17 to 0.69, indicating that there is not enough statistical evidence to reject the null hypothesis that the set of instruments has no significant effect on household level firewood demand. The rationale of using such a test for exclusion restriction is that in our sample only 2 per cent of households are using the improved stoves, presumably because not all households in the sample had a choice of adopting the improved stoves. If the instruments affect the firewood demand directly, this should show up in these regressions.
Results from the 2SLS regressions are presented in table 5. While comparing these results with the OLS results from table 4, we can see that after instrumenting the types of stove, the coefficient of the MUDSTOVE becomes bigger with negative sign, while the coefficient of the IMPROVEDSTOVE also becomes bigger but with positive sign. These results indicate that, on average, households with the IMPROVEDSTOVE are using more firewood than households with the open-fire or the mud stove. This result may seem counterintuitive, but it actually confirms what some of the researchers have been reporting on the inefficiency of the improved stoves with regard to fuel–wood consumption and cooking time (see McCracken and Smith, Reference McCracken and Smith1998).
Other results in table 5 are mostly consistent with the results from the OLS estimates (see table 4). The signs of the coefficients of the COLLECTIONTIME (Models V and VI) and the log (COLLECTIONTIME) are stable. As expected, the sign of the COLLECTIONTIME squared is negative, indicating that when collection time goes beyond 12 hours (0.329/0.026) for one bhari of firewood, households may start switching to other fuels. Given that the average collection time for firewood self-sufficient households is less than 4 h in our sample, the turning point seems to be quite far, meaning that those households would continue to collect firewood for household energy conditional on other control factors.
The effect of per capita income on firewood demand is negative but not significant, which is comparable to Baland et al. (Reference Baland, Bardhan, Das, Mookherjee and Sarkar2010). As before, the coefficient of community forest is positive and significant. After correcting for endogeneity, the effect of the presence of children below six years of age is still stable but not strong enough to draw any definite inference. The results are mostly stable with alternative model specifications.
6.3 Alternative measure of opportunity cost
So far, we have used the time taken to collect one unit of firewood as a proxy measure for the unit price of firewood. However, a better measure of a unit price of firewood would be the opportunity cost of collection time. This opportunity cost could be the wage rate of the person in alternative activity who collects firewood. The NLSS data do not provide information on who actually collects firewood, but Amacher et al. (Reference Amacher, Hyde and Joshee1993) and Khare et al. (Reference Khare, Sarin, Saxena, Palit, Bathla, Vania and Satyanarayana2000) find that firewood scarcity results in an increased labor burden on women. This indicates that women may be the ones who mostly collect firewood.
In the NLSS data, a small subsample of households hired female farm workers for agricultural activities. We utilize this information as a proxy measure of the opportunity cost of firewood collection time. As an individual household has no control over the wage rate of the hired female farm workers, we treat this variable as exogenous.
Table 6 reports the OLS and 2SLS results where collection time is replaced by the wage rate of the hired female farm workers multiplied by the collection time (OPPORTUNITYCOST). Again, these results are mostly consistent with earlier ones. From table 6, we can see that the MUDSTOVE user households could consume 17–32 per cent less firewood as compared to the open-fire stove users, while the IMPROVEDSTOVE user households may consume a comparable amount of firewood to households that use open-fire stoves. The sign of the coefficient of the OPPORTUNITYCOST is positive and insignificant. Other results are mostly comparable to what we have discussed earlier.
6.4 Fixed effect estimates
As an alternative to the simple OLS and the 2SLS estimation methods, we also use the fixed effect estimator. Our interest here is to account for the village level unobservable heterogeneity and examine how sensitive our results are in terms of alternative estimations. Table 7 presents the results from the fixed effect estimator. In terms of the right-hand side variables, these results are comparable to Models III and IV in table 4 and Models VII and VIII in table 5. In all respects, these results are mostly consistent with what we have discussed earlier with one exception: the coefficient of MUDSTOVE is now positive but not significant across all models, while the coefficient of IMPROVEDSTOVE switches sign from positive to negative based on the models as before but statistically insignificant across the models. In practical terms, we can see that households that use improved stoves or mud stoves could eventually consume a comparable amount of firewood to the open-fire stove user households, indicating that improved stoves are not really helping to reduce firewood consumption in Nepal.
7. Conclusion
This paper analyzes the effect of stove type on firewood consumption at the household level in Nepal. Using nationally representative household survey data from Nepal, we investigate the demand for firewood based on type of stoves households use. For this analysis, we use a subsample of firewood user households from the NLSS survey that collect and use firewood for household energy. As discussed above, our results are somewhat unexpected. More specifically, contrary to the common belief regarding the efficiency of improved stoves, we find that households with the improved stoves may use the same amount or even more firewood than households with the traditional mud stove or the open-fire stove. This issue, however, needs further investigation to arrive at a definite conclusion since only 2 per cent of households were using improved stoves during the survey year of our data set.
One possible explanation of why the improved stove user households may consume more or not less firewood than the traditional open-fire stove or the mud stove users could be the rebound effect as in Zein-Elabdin (Reference Zein-Elabdin1997).Footnote 8 For example, when the improved stoves reduce firewood demand in the first place, it would lower the shadow price of firewood. A lower shadow price could in turn prompt households to consume more firewood. An alternative explanation would be that the improved stoves mostly come with attached chimneys that help reduce the amount of smoke in the house. Traditionally, the chimney is not a part of the open-fire or the traditional mud stove in Nepal. As the smoke level declines due to the chimney (Malla, Reference Malla2009), household members may feel better in terms of health benefits and comfortable when the stove keeps running. Consequently, they may either keep their stove running for longer hours to keep the house warm or cook more frequently requiring more firewood.
Our results indicate that in the presence of the rebound effect, if the existing improved stove technology continues to exist and we want to reduce the demand for firewood, the short-term solution to reduce firewood consumption may be to replace open-fire stoves with mud stoves. This switching from open-fire stove to mud stove would be more acceptable as about 56 per cent of the households are already using mud stoves. Such replacement of the open-fire traditional stove with the mud stove does not require a heavy investment since it could be done with simple and locally available technology. While switching from an open-fire stove to a mud stove, adding a smoke hood or chimney would help to address the indoor air pollution problem (Malla, Reference Malla2009), but it might generate the rebound effect, and our fixed effect estimates indicate that households with all stove types seem to demand a comparable amount of firewood. Our results indicate the need to reexamine the improved stove technology and its dissemination. In the medium to long run, making cleaner fuel or gasifiers more accessible to rural households would be a better option that could reduce firewood demand as well as indoor air pollution.
