Firewood, forests, and fringe populations: Exploring the inequitable socioeconomic dimensions of Liquified Petroleum Gas (LPG) adoption in India

Abstract

Liquified petroleum gas (LPG) is an important clean fuel alternative for households that rely on burning biomass for daily cooking needs. In India, Pradhan Mantri Ujjwala Yojana (PMUY) has provided poor households with LPG connections since 2016. We investigate cooking fuel use in households to determine the impact of the policy in the Central Indian Highlands Landscape (CIHL). The CIHL has a large population of marginalized social groups, including Indigenous, Scheduled Tribe, Schedule Caste, and Other Backward Caste people. We utilize survey data from 4,994 households within 500 villages living in forested regions collected in 2018 and a satellite-derived measure of forest availability to investigate the household and ecological determinants of LPG adoption and the timing of this adoption (pre- or post-2016). In addition, we document patterns of firewood collection and evaluate the extent to which households acquiring LPG change these activities. The probability of cooking with LPG was lowest for marginalized social groups. We observe that households recently adopting LPG, likely through PMUY, are poorer, more socially marginalized, less educated, and have more forest available nearby than their early-adopter counterparts. While 90% of LPG-using households continue to use firewood, households that have owned LPG for more years report spending less time collecting firewood, indicating a waning reliance on firewood over time. Policies targeting communities with marginalized social groups living near forests can further accelerate LPG adoption and displace firewood use. Despite overall growth in LPG use, disparities in access to clean cooking fuels remain between socioeconomic groups in India.

1. Introduction

Globally, 2.8 billion people, often the world’s poorest and most marginalized, burn biomass to meet their daily household energy needs [1]. Inefficiently burning traditional solid biomass – firewood, coal, agricultural residue, and dung –for cooking and heating has substantial negative impacts on public health and the environment. Exposure to household air pollution (HAP) from the incomplete combustion of biomass is one of the greatest global environmental health risks, estimated to account for 2.3 million premature deaths each year [2]. The extraction of biomass can also hinder forests’ ability to provide a healthy ecosystem for people by contributing to forest degradation, deforestation, and climate change around the world [3–5]. In recent decades, clean cooking fuels such as Liquified Petroleum Gas (LPG) have been an important tool for programs and policies aiming to deliver its multiple benefits, including: improved air quality, climate change mitigation, and reduced biomass demand. For example, Sustainable Development Goal 7 aims for affordable and sustainable energy availability, which includes accelerating the access to clean and safe cooking fuels. However, achieving sustainable development will require an understanding of who has access to clean cooking and how that access changes the use of traditional solid biomass.

Traditional biomass-based cooking is widespread across India. In 2011, about half of India’s households used firewood as their primary cooking fuel and 12% used it as a secondary fuel, totaling 150 million households [6]. However, the burden of biomass use in India is unequal across gender, social groups, and regions. Recognizing specific groups of stakeholders with unequal access to clean cooking fuel – a key tenet of energy justice [7] – is necessary to address the equity in promoting clean cooking and sustainable development.

Attaining the multiple benefits of fuel transitions requires that clean cooking fuels significantly displace traditional biomass use. However, studies from around the world and in India show that households rarely cease to use their traditional cooking practices when they adopt cleaner cooking technologies [8–13]. There are multiple reasons households may continue to use biomass after acquiring a cleaner fuel, a practice termed fuel stacking, including: household economics, individual preferences, and specific energy end uses. Historically, high costs and low availability of clean cooking fuels have limited the penetration into regions with significant household reliance on biomass [14–17], largely excluding poor and marginalized households. In contrast, the availability of biomass as a monetary-cost-free alternative cooking fuel is often considered a driver of continued traditional cooking practices [18].

Quantitative and qualitative evidence suggests that biomass availability can affect fuel collection time and effort, thereby influencing household fuel choices, including the decision to adopt cleaner fuels [19–21]. In previous quantitative studies seeking to understand the association between biomass availability and cooking fuel choice, biomass availability has been defined in several different ways, including: proxies for assessing geographic variabilities in fuel choice that might be due to biomass availability [22,23], distance to the nearest forest [24–26], time spent on fuel collection [27,28], forest area per person, perceived convenience and reliability of biomass fuel supply [29], and satellite-derived measures of biomass availability or forest cover [30,31]. While not as well characterized as individual and household determinants of fuel choice, characterizing the supply-side determinants of fuel choice can inform the motivations for continued biomass use after clean cooking fuel adoption and use. Given that even limited traditional biomass-based cooking can lead to high health risks and continued environmental and climate impacts [32], displacing household biomass use with clean fuels can havae substantial implications for health, environmental, and climate burdens. Efforts to understand the extent to which LPG use and biomass availability modifies biomass collection patterns can help us achieve this goal by identifying strategies to curb continued household biomass combustion.

1.1. Disparities in India’s cooking fuel

India’s rural households (71% of the country) are more dependent on firewood and have limited access to LPG as compared to urban households (62% vs. 21% in 2011, respectively) [6]. Among rural households, wealth and formal educational attainment are strongly positively associated with using LPG [33–37]. Recent evidence also indicates that stable, salaried incomes as compared to more seasonal agricultural or day labor are associated with LPG ownership [38]. Furthermore, clean cooking adoption, much like cooking itself, is gendered. Women are primarily responsible for cooking and biomass collection, disproportionately facing the negative health and well-being burdens of biomass cooking. And yet, men often control finances. There is evidence that when women are involved in decision-making then a household is more likely to have LPG in rural India [39].

Further, tribe and caste status have been an important determinant of LPG access in communities. India’s Scheduled Tribe (ST) or Scheduled Caste (SC) communities, terms in the Indian Constitution that describe a diverse group of historically marginalized Indigenous and religious communities, are highly reliant on biomass and are socioeconomically disadvantaged. For example, the human development index and human poverty index, composite measures of life expectancy, education level, and standard of living, is lower in ST communities than the rest of India [40]. ST and lower caste households have low rates of clean cooking fuel adoption [41]. An analysis of the National Sample Survey data (2011–2012) found that ST and SC households were 9% less likely to own LPG as compared to non-ST or non-SC households [42].

There are a high number of ST and other non-general caste households in the forested regions of the Central Indian Highlands Landscape (CIHL), a region which spans across three Indian states of Madhya Pradesh, Maharashtra, and Chhattisgarh. In the CIHL, households have traditionally met their subsistence and livelihood needs with forest resources. For example, in rural households in Madhya Pradesh located within a distance of two kilometers of the forest, more than half of households were ST or SC and they derived 49% of their income from forest products [43].

Households in the CIHL are heavily reliant on biomass for meeting their household energy needs (Figure 1). In 2011, 86% of rural households in the CIHL used firewood as their primary cooking fuel, compared to 63% of all rural Indian households, while only 5% relied on LPG (Table S1) [6]. Households living near forests can collect firewood at no monetary cost, which may be a barrier to investing in an alternative, more costly cooking fuel [44,45]. Still, while not monetarily costly, these households devote effort and time to collect firewood.

Figure 1.

Map of 500 survey villages, indicated as colored circles, cities with populations greater than 88,000 people, protected areas, and tree cover in the Central Indian Highlands Landscapes (CIHL). Tree cover data is from Hansen et al. (2013). Fig. 1A. The color of each village indicates the proportion of surveyed households which use Liquified Petroleum Gas (LPG) for cooking, where darker shades of blue represent a higher proportion. There are four categories of proportion of households cooking with LPG, classified according to quantile. Fig. 1B. The color of each village indicates the first year in which LPG was used for cooking by households within that village, where darker shades of red represent more recent years. Except in the 35 villages where LPG was not used by any household, LPG was available in all villages in 2017.

1.2. LPG expansion and fuel stacking

India has pioneered several ambitious clean cooking fuel programs to address the high burden of biomass cooking in rural households in recent years. Notably, the Government of India, through Pradhan Mantri Ujjwala Yojana (PMUY), has provided about 80 million LPG connections to below poverty line households since 2016 [46]. PMUY beneficiaries – exclusively women – have their LPG cylinder deposit and regulator and installation charges covered by the program (1,600 Indian Rupees (INR) in total; 23 United States Dollar (USD)¹). Still, households are required to purchase a double-burner LPG stove (approximately 1,000 INR; 14 USD) and their first LPG refill (500 INR; 7 USD), with optional loan assistance.

The Government of India now estimates that 95% of Indian households have access to LPG, thanks in large part to PMUY [47]. An analysis of panel survey data collected in 2015 and then in 2018 (ACCESS) in six energy-poor north Indian states shows that access to LPG has increased for marginalized populations. The proportion of SC and ST households using LPG increased by 43% and 30%, respectively [48]. ACCESS data collected in Madhya Pradesh – located in the CIHL – shows that 59% of households acquiring LPG between 2015 and 2018 did so via PMUY [49]. Although access to LPG increased for SC households, LPG adoption has lagged among ST households [50].

While PMUY has helped to overcome the initial hurdles of LPG stove and connection access and affordability, the program does not address LPG use after adoption. Recent evidence shows that PMUY beneficiaries use LPG less than general customers across multiple contexts [38,47,51,52]. LPG cylinder refill costs remain barriers to sustained LPG use and may be exacerbated by the seasonality of income, community or cultural norms, or biomass availability [53,54]. Still, there is some evidence to suggest the longer a household has LPG, the larger a role it has in the household [54,55].

1.3. Study objectives

This study combines household-level and remotely-sensed satellite data to understand socioeconomic and environmental drivers of cooking fuel choice and firewood collection in rural Indian households living near forests in the CIHL. The region remains highly forested and there are a high number of ST, SC and OBC households that have long relied on forest resources for consumption and livelihoods. The diffusion of LPG after PMUY and patterns of fuel stacking in communities within forested regions remains unknown. This study population is of particular interest for jointly evaluating the socioeconomic and environmental drivers of fuel choices and firewood collection in traditionally disadvantaged populations, which is an important consideration to implement socially inclusive clean cooking fuel policy.

Our study contributes to understanding LPG adoption and clean cooking transitions by collecting and analyzing cooking fuel data from marginalized, forest-dependent populations where households cook with both LPG and firewood. The study addresses the following objectives: 1) to examine the socioeconomic and environmental drivers of the use of LPG for cooking before and after PMUY was implemented and 2) to assess the influence of LPG ownership over time on seasonal household firewood collection patterns. We address these objectives through analysis of household surveys from approximately 5,000 households living near forests in the CIHL.

2. Study Area

This study was carried out in the CIHL across 32 administrative units, known as districts (Figure 1). The total population within the study area is about 51 million (Table S1). While 35% of the CIHL are ST or SC, these groups comprise 8% and 16% of the India’s total population, respectively [6]. Communities in the CIHL are largely rural and face higher levels of poverty, measured as a combination of health, education, and living standards, compared to the rest of India [56]. In addition to major urban centers – Nagpur, Jabalpur, and Bhopal – there are numerous rural villages throughout the region where residents primarily engage in agriculture and livestock rearing as economic activity.

Forest-based economic and subsistence activities contribute substantially to households in the CIHL. The CIHL has some of India’s largest remaining forest patches [57], which can be found inside and outside the 16 protected areas² (PAs). Activities in forests include livestock grazing and collection of firewood and non-timber forest products, which are important for nutritional security, medicinal purposes, and for sales as raw material or finished products to local and urban markets, provide a critical cash income source [58].

The role of forest products in securing basic needs and providing a source of income can be particularly important for Indigenous and marginalized households [59]. Our study was limited to rural communities in the CIHL that live in proximity to forest. Understanding drivers of cooking fuel use in the ST and lower caste populations living in the forested areas of the CIHL is an important step towards achieving a just cooking fuel transition.

3. Methodology

This study leverages cross-sectional structured surveys administered to rural households living within 8 km of forests in the CIHL from February to March 2018. We assess the association between household characteristics, cooking fuel use, and firewood collection patterns. We also use published data on vegetation to incorporate availability of forest in our understanding of cooking fuel patterns.

3.1. Household Sampling

We selected 500 study villages in Madhya Pradesh, Maharashtra, and Chhattisgarh according to multi-stage criteria that resulted in a representative sample. The first criteria was selection of study villages that were not in PAs but were within eight kilometers of a forested region, as defined by Hansen et al. (2013) [60]. Next, we employed a stratified sampling scheme for village selection based on the distance of that village to a town and the distance of that village to a road. Towns were identified in the 2011 Census of India as a place with a municipality, a minimum population of 5,000, population density greater than 400 people per km², and at least 75% of the male population employed outside the agricultural sector. Village distance to a road was calculated using the Digital Chart of the World road maps (downloaded from http://www.diva-gis.org/gdata) [61].

Villages were split into two groups based on whether they were above or below the median distance to nearest town. These two village groups were each further split into two groups based on whether the distance to nearest road was above or below the median of that initial grouping. This process resulted in four village groups, each farthest and closest to a road and a town, from which 125 study villages were randomly selected from each group.

Ten households in each of the 500 study villages were surveyed. Study villages consisted of multiple hamlets, or tolas. Tolas were identified by asking the village head the number of tolas and how many people and households were in each. Within each tola, households were randomly selected with the number of sampled households per tola matching the tola’s relative size in the village. Households were surveyed first by selecting a random start point and direction in a tola, and then sampling every four to five households.

Surveys were administered to 5,000 households across 500 villages in Madhya Pradesh (N=3239), Maharashtra (N=946), and Chhattisgarh (N=809) (Table 1). Households in Madhya Pradesh represented 65% of survey households, as compared to households in Maharashtra and Chhattisgarh which were sampled at 19% and 16%, respectively, because Madhya Pradesh comprises the geographic majority of the CIHL. Six households missing variables used in analyses were dropped in the present study. These households did not differ significantly from households included in analysis on available socio-economic or household energy use variables.

Table 1.

Key characteristics of study sampling

	Full sample	Madhya Pradesh	Maharashtra	Chhattisgarh
Number of districts	32	21	6	5
Number of villages	500	324	95	81
Number of households	4994	3239 (65%)	946 (19%)	809 (16%)
Households with woman household head	1355 (27%)	1105 (82%)	83 (6%)	167 (12%)

3.2. Survey Instrument

The survey instrument was primarily designed to assess the social structure and economic activities of forest-dependent communities in the CIHL. The structured survey included questions related to household demographics, socio-economic status, natural resource use, household energy uses, and perceptions of forest status. The survey was piloted twice in 2017 within three districts of the CIHL (Balaghat, Seoni, and Mandla). A trained field team hired through MORSEL India, a social research company with experience in household questionnaires in rural India, implemented the survey across the study area between February 2^nd and March 28^th, 2018. Surveys were conducted in the local language, Hindi, and lasted approximately 45 minutes per household.

3.3. Satellite-Derived Measure of Forest Availability

We used gridded forest cover data (percent tree cover at 30-meter resolution) from the Global 2010 Tree Cover product to estimate village-level forest availability [60]. We obtained boundaries identifying the borders of each study village from ML Infomap Pvt. Ltd. (https://www.mlinfomap.com/Main/indiamaps.html). We estimated the percent tree cover within 2.74 kilometers of study village boundary edges in addition to forest inside village boundaries. We excluded forest cover within PAs because of restricted access to this forest (see Section 4.4 for more details on restricted access). We specified a 2.74-kilometer buffer because this was the mean distance reported by households to travel on average to collect firewood across all seasons (summer, monsoon, post-monsoon, winter) (Table S2). In doing so, we expect to capture the majority of trips commonly taken to collect firewood. We tested additional buffer distances (1 km, 2 km, 3 km, 5 km, 8 km, and 10 km), including the median reported distance traveled of 2.0 km, to evaluate for potential threshold distances at which forest cover does not affect collection patterns. However, 2.74 kilometers was selected because it explained the most variance in firewood collection across all seasons (along with 3 km). The 2.0 and 2.74 km buffer resulted in a lower Akaike Information Criterion than 3 km in logistic regressions where LPG ownership was the outcome variable (Table S3).

3.4. Outcome Variables

The present study evaluated three outcomes central to patterns of cooking fuel use and collection: 1) the use of LPG for cooking, 2) when LPG was acquired (before or after 2016), and 3) the time spent collecting firewood. These outcomes were used to examine recent LPG adoptions and identify fuel stacking patterns in households that use LPG and firewood for cooking.

Use of LPG for cooking.

Households were asked “Does your household use LPG for cooking?” Responses were used as a binary outcome variable in a multilevel logistic regression to determine the household and ecological characteristics associated with the use of LPG for cooking. Responses about firewood collection (see below) indicate that households are not exclusively using LPG.

LPG ownership after 2016.

Households who used LPG for cooking (N=2276) were asked, “When did you start using it?” These responses were categorized: 1) Before 2013, 2) 2013, 3) 2014, 4) 2015, 5) 2016, and 6) 2017. Responses were further grouped based on LPG adoption before (pre-2016) or after (2016 or 2017) PMUY. In Madhya Pradesh, PMUY was launched on July 4^th, 2016 [62]. PMUY was launched in Maharashtra on October 7^th, 2016 [63] and Chhattisgarh on August 13^th, 2016 [64]. This pre- or post-PMUY binary variable was used as an outcome in a multilevel logistic regression to assess variations in the determinants of LPG adoption before or after PMUY. In a sensitivity analysis, we re-specify the post-PMUY period to only include 2017. We observe no meaningful deviation in the associations between covariates and the outcome (Figure S1). We assume that households that adopted LPG after 2016 received LPG as a direct result of the policy, although LPG adoption could be influenced by other factors.

Firewood Collection.

Participants, including respondents who used LPG for cooking, were asked about their firewood collection patterns during each season of the year to assess the intensity of firewood collection and its variability in time. Seasons were defined as: summer (April – June), monsoon (July – September), post-monsoon (October – November), and winter (December – March). Specifically, participants were asked for each season: “In a typical week, how many days did you or a person in the household visit the forest to collect firewood?” Participants reporting firewood collection trips were then asked, “On average, how many hours did you or a person in the household spend collecting firewood on one day?” These two variables were multiplied to compute the outcome variable hours of firewood collection per week. Seasonal patterns in firewood collection required stratified analysis for the monsoon season, when much less firewood collection was reported, to determine associations with LPG ownership (see Section 4.3).

3.5. Statistical Approach

First, we assess the association between household characteristics, including income and socioeconomic status, education, and forest availability, and the use of LPG for cooking (Equation 1). Among LPG owners, we then aimed to understand the differences between households that adopted LPG before or after PMUY (Equation 2). In our third model, we assess the association between year of LPG adoption and changes in time spent collecting firewood, controlling for other covariates. Equations are described below:

$$\log \left(\frac{P\left(LPG\hspace{0.33em}for\hspace{0.33em}cookin{g}_{ij}\right)}{P\left(1-LPG\hspace{0.33em}for\hspace{0.33em}cookin{g}_{ij}\right)}\right)={\beta }_0+{\beta }_1\cdot X_i+{\epsilon }_i$$ (1)

$$\log \left(\frac{P\left(LPG\hspace{0.33em}ownership\hspace{0.33em}in\hspace{0.33em}2016\hspace{0.33em}or\hspace{0.33em}{2017}_{ij}\right)}{P\left(1-LPG\hspace{0.33em}ownership\hspace{0.33em}in\hspace{0.33em}2016\hspace{0.33em}or\hspace{0.33em}{2017}_{ij}\right)}\right)={\beta }_0+{\beta }_1\cdot X_i+{\epsilon }_i$$ (2)

$$Firewood\hspace{0.33em}Collection\hspace{0.33em}hrs/wee{k}_{ij}={\beta }_0+{\beta }_1\cdot LP{G}_{Yea{r}_i}+{\beta }_2\cdot Y_i+{\epsilon }_{ij}$$ (3)

where X_i is a matrix of covariates identified from reviews of the clean cooking fuel adoption and use literature [16,17,36,37], as well as evidence of correlations with both the outcome and explanatory variables of interest in the study data (Figure S2). Y_i, the matrix of covariates used in Equation 3, only includes X_i covariates that were statistically significantly (P < 0.05) associated with LPG ownership in Equation 1. The covariates are described in Table 2. We report models with district-level fixed effects³ (District FEs) to account for potential residual spatial confounding, as carried out elsewhere [65,66]. Models with District FEs additionally explained more variance in the outcome variables than those without District FEs (see Table S4). Additionally, we present results from the predicted probabilities of LPG ownership for continuous covariates modeled in Equations 1 and 2 (Figure S3). All analyses were carried out in R version 3.5.0 (R Core Team, 2018) using the MASS (Venables & Ripley, 2002), lme4 [67], and margins [68] packages.

Table 2.

Description of covariates used in statistical models

Covariate	Variable type	Description
Year of LPG ownership (2013 or earlier, 2014–2015, or 2016–2017)?	Binary	Responses to the question “When did you start using LPG?” were grouped into four categories based on similarities in household characteristics: 1) No LPG; 2) Acquired LPG in 2013 or earlier; 3) Acquired LPG in 2014–2015; and 4) Acquired LPG in 2016–2017 (Table 4). No LPG used for cooking was a baseline category. This variable was included exclusively in Equation 3.
Monthly expenditure (INR)	Continuous	Wealth has been positively associated with cleaner cooking uptake around the world [29,69]. However, consistent incomes are rare in many poor and rural communities in India and globally [70]. Therefore, we utilized monthly household expenditures, which is a reliable predictor of wealth used in previous studies [37,71]. This covariate was log transformed and standardized in analyses (Mean = 0, Standard deviation (SD) = 1).
Has money in a bank account?	Binary	In this study sample, having money in a bank account is an additional measure of wealth and capital. Baseline category was not having money in a bank account in the past year4.
Has saved money?	Binary	Having money to save is another measure of wealth and capital. Baseline category was not having money to save in the past year.
Woman as household head?	Binary	Because of the gendered nature of cooking and decision-making in rural Indian households, households headed by woman may be more likely to adopt cleaner cooking technologies [20,29,39,72–74]. In addition, women are the primary collectors of firewood and the beneficiaries targeted by PMUY. Baseline category is having a man as a household head5.
Caste (ST, SC, or OBC)?	Binary	We use general (or forward) caste as the baseline category. Other categories include Scheduled Tribe, Scheduled Caste, and Other Backward Class. Caste has been associated with cooking fuel choice in other case studies in India [34,35,39,42].
Education of the survey respondent (primary/secondary, high school, or intermediate and above)?	Binary	Education of the household head has been strongly positively associated with clean cookstove ownership in India previously [34,39,69,75]. Baseline category is no formal education, with additional categories being completed (i) Primary/Secondary school, (ii) High School, and (iii) Intermediate and above.
Increased difficulty in firewood collection?	Binary	We assessed changes in perceived difficulty of firewood collection. Participants were asked “Over the last five years, has it become easier or harder to collect firewood?” Responses were coded into five categories: 1) much easier, 2) somewhat easier, 3) stayed constant, 4) somewhat harder, and 5) much harder. A majority (88%) of respondents reported firewood collection as getting somewhat harder or much harder thus we recoded responses to be used as a binary variable where the baseline category was “stayed constant,” “somewhat easier,” or “much easier.” The correlation coefficient of a binary variable created from a collapsed Likert scale and the original scale is between 0.8 to 0.9, and such a variable transformation is appropriate where nuances of the response are not critical for interpretation [76].
Forest availability (% tree cover)	Continuous	Percent tree cover within a buffer distance of each study village and outside a PA was log transformed and standardized to use as a covariate in analyses (see Section 2.4 for more details).
Distance to road (km)	Continuous	Distance to nearest road (km) at the time of the survey was calculated at the village level using OpenStreetMap (OSM) road data. We consider distance to the nearest road as an indicator of access to LPG cylinder refills. To obtain historical road data from our study region, we used the Overpass API tool [77] using a bounding box of 17.7° to 26.4° N; 74.9° to 84.1° E to allow for a 1° buffer around the study region. Second, the Osmium tool [78] was used to extract historical OSM road layers last updated February 28th, 2018. This covariate was log transformed and standardized in analyses.

3.6. Qualitative Data

We provide context to firewood collection using responses from the open-ended question “Why do you think it has become easier/harder to collect firewood?” Responses were transcribed in the local language by field staff at the time of data collection and then hand coded according to emergent themes. Quotes and themes were then translated to English by a bilingual member of the research team.

4. Results

Of the 500 study villages, there were 35 villages (7%) where no households used LPG for cooking and only four villages where all households used LPG for cooking. There was substantial variation in LPG use across study villages, as well as year of LPG uptake (Figure 1).

Nearly half of households (46%) reported the use of LPG for cooking at the time of the survey in early 2018. Three-quarters of households with LPG reported to have acquired the stove and connection in 2016 or 2017. Households that reported acquiring LPG more recently were less wealthy, more likely to be ST, near more forest, and had lower levels of formal education than those that adopted LPG in 2013 or before (Table 3). Households that reported acquiring LPG between 2016–2017 were similar to households that reported to not cook with LPG. Of households that owned LPG, 90% also reported cooking with firewood and 68% collected firewood in at least one season through the year. More than half of study households (57%) were ST, over a quarter were OBC (27%), and 12% were SC; only 4% of households were general caste.

Table 3.

Summary statistics of households, by year of LPG adoption.

	Full sample	2013 or before	2014–2015	2016–2017	No LPG
Sample Size, N (%)	4994 (100%)	270 (5%)	277 (6%)	1729 (35%)	2718 (54%)
Age of Respondent, Mean (SD)a	42.0 (13.3)	43.2 (14.4)	41.6 (13.4)	41.9 (13.1)	41.9 (13.31
Woman as household head, N (%)a	1355 (27%)	56 (21%)	46 (17%)	468 (27%)	785 (29%)
Man Chief Wage Earner, N (%)a	4476 (90%)	250 (93%)	261 (94%)	1551 (90%)	1104 (89%)
Respondent education, N (%)a
High School	564 (11%)	41 (15%)	56 (20%)	212 (12%)	255 (9%)
Intermediate or Greater	534 (11%)	104 (39%)	54 (19%)	162 (9%)	214 (8%)
No Formal Education	2021 (40%)	48 (18%)	59 (21%)	656 (38%)	1258 (46%)
Primary/Secondary	1875 (38%)	77 (29%)	108 (39%)	699 (40%)	991 (36%)
Household Caste, N (%)a
General	198 (4%)	42 (16%)	20 (7%)	59 (3%)	77 (3%)
Other Backward Caste	1338 (27%)	105 (39%)	109 (39%)	465 (27%)	659 (24%)
Schedule Caste	608 (12%)	43 (16%)	39 (14%)	219 (13%)	307 (11%)
Scheduled Tribe	2850 (57%)	80 (30%)	109 (39%)	986 (57%)	1675 (62%)
Monthly Expenditure (INR), Mean (SD)a	3785 (2846)	6206 (4417)	4843 (3511)	3745 (2317)	3462 (2736)
Monthly Expenditure (USD), Mean (SD)	54 (41)	89 (63)	69 (50)	54 (33)	49 (39)
Tree cover (%), Mean (SD)a	5.64 (7.21)	3.48 (4.93)	4.10 (6.04)	5.42 (6.78)	6.28 (7.56)
Distance to Road (km), Mean (SD)	1.69 (2.35)	1.28 (2.03)	1.58 (2.23)	1.72 (2.44)	1.73 (2.33)
Has Saved Money?, N (%)a	1457 (29%)	131 (49%)	89 (32%)	522 (30%)	715 (26%)
Has Money in a Bank Account?, N (%)a	1822 (36%)	153 (3%)	107 (2%)	669 (13%)	893 (18%)

^aIndicates that there was a statistically significant difference between the households depending on year of LPG adoption at P < 0.05 in ANOVA. SD is standard deviation.

4.1. Determinants of using LPG for cooking

Households that used LPG were wealthier, better educated, and had higher odds of being general caste than those without LPG (Figure 2). Controlling for other covariates, the probability of cooking with LPG was higher by 15 percentage points (95% CI: 10 – 20 percentage points), 12 percentage points (95% CI: 7.6 – 17 percentage points), and 6.2 percentage points (95% CI: 3.1 – 9.3 percentage points) if the household head was educated at the intermediate or above, high school, or primary/secondary level, respectively, as compared to a household headed by a person with no formal education. In addition, the probability of using LPG was significantly positively associated with higher monthly expenditure (Figure S3).

Figure 2.

Coefficient plot for logistic regression with District fixed effects assessing the household and ecological characteristics that are associated with LPG ownership. Points represent coefficients of average marginal effects (percentage point change in the probability of LPG ownership) and whiskers show 95% confidence intervals.

Households belonging to the Scheduled Tribe (ST) caste, which comprised almost 60% of the study sample, had the lowest odds of having LPG at the time of the survey, as compared to the other castes (Scheduled Caste (SC), Other Backward Class (OBC), and general caste). Accounting for other household characteristics and covariates, the probability of using LPG was lower by 14 percentage points (95% CI: 6.5 – 21 percentage points) if a household was ST as compared to a household belonging to the general caste. Similarly, the probability of using LPG was 9.3 percentage points (95% CI: 2.0 – 17 percentage points) and 6.6 percentage points (95% CI: 1.3 – 15 percentage points) lower for OBC and SC households, respectively, as compared to belonging to the general caste.

In addition to household characteristics, some contextual environmental variables were associated with using LPG. The probability of using LPG was significantly negatively associated with higher tree cover (Figure S3). For every additional percent of tree cover nearby a village, the probability of using LPG decreased by 4.1 percentage points (95% CI: 2.5 – 5.7 percentage points). Participants stating that firewood collection had increased in difficulty in the past five had a 4.4 percentage point (95% CI: 0.0 – 8.6 percentage point) lower probability of using LPG as compared to those that did not perceive firewood collection to have become more difficult.

4.2. Explaining the timing of LPG adoption

Of all LPG users in the study sample, households that adopted LPG in 2016 or after were poorer, less educated, and had higher odds of being from a non-general caste than those that adopted LPG before PMUY (Figure 3). Controlling for other covariates, households that belonged to the Scheduled Tribe caste had an 18 percentage point (95% CI: 11 – 25 percentage point) higher probability of acquiring LPG after PMUY as compared to a household in the general caste. Similarly, households belonging to the Scheduled Caste or Other Backward Class had higher probabilities of acquiring LPG after PMUY, though somewhat lower than those in the ST caste.

Figure 3.

Coefficient plot for logistic regression with District Fixed Effects assessing the household characteristics that are associated with adopting LPG in 2016 or 2017. Points represent coefficients of average marginal effects (percentage point change in the probability of adopting LPG in 2016 or 2017) and whiskers show 95% confidence intervals.

The level of formal education of household members and monthly expenditures were both positively associated with having adopted LPG prior to 2016 (those that are considered general consumers as opposed to likely being PMUY beneficiaries). For example, a household headed by a person that had completed an education at the intermediate level or above had 15 percentage points (95% CI: 10 – 20 percentage points) higher probability of acquiring LPG prior to PMUY as compared to a household headed by a person with no formal education. Similarly, the probability of LPG ownership prior to PMUY was significantly positively associated to monthly expenditure (Figure S3).

Households acquiring LPG after the beginning of PMUY had greater village-level forest cover than those that acquired LPG before PMUY. The use of LPG for cooking after 2016 was greater for households with high forest availability, whereas we had observed a significant negative association between nearby tree cover and the use of LPG across all years (Figure S3). LPG ownership after 2016 was also higher in households that perceived increased difficulty in firewood collection over the last 5 years.

4.3. Characteristics of firewood collection

Almost all households (95%) in the study sample reported that they use firewood for cooking at some point during the year. Nearly 70% of households reported weekly firewood collection during the summer, post-monsoon, and winter seasons, but only 33% of households reported weekly firewood collection during the monsoon season (Table 4). Households reporting to collect firewood during the summer, post-monsoon, or winter generally collected firewood during all three seasons. Almost all households (93%) reporting to collect firewood during the monsoon season collected firewood throughout the entire year.

Table 4.

Summary of household firewood collection patterns by season.

	Summer	Post-monsoon	Winter	Monsoon
Households reporting weekly firewood collection, N (%)	3465 (69%)	3442 (69%)	3384 (68%)	1661 (33%)
Purpose of firewood collection, N (%)
Cooking	3440 (99%)	3419 (99%)	3252 (96%)	1645 (99%)
Selling	22 (1%)	18 (1%)	18 (1%)	13 (1%)
Heating	3 (<1%)	5 (<1%)	112 (3%)	3 (<1%)
Number of days per week
Mean (SD)	3.52 (1.91)	3.22 (1.83)	3.73 (1.89)	2.50 (1.47)
Median (IQR)	3 (2, 5)	3 (2, 4)	3 (2, 5)	2 (1, 3)
Number of hours per day
Mean (SD)	4.51 (1.76)	4.46 (1.83)	4.54 (1.82)	4.06 (1.79)
Median (IQR)	4 (3, 5)	4 (3, 5)	4 (3, 5)	4 (3, 5)
Number of hours per week
Mean (SD)	16.48 (12.44)	14.78 (11.38)	17.82 (13.17)	10.20 (8.26)
Median (IQR)	12 (8, 21)	12 (6, 20)	15 (8, 24)	8 (5, 12)

SD is standard deviation and IQR is interquartile range.

The average distance traveled for firewood was 2.74 km (Standard Deviation (SD): 2.02) and the median distance was 2 km (interquartile range: 1.75 – 3.00 km) (Table S2). Distance traveled for firewood did not differ significantly across seasons of the year, suggesting that households might acquire wood from the same locations throughout the year.

On average, households that collected firewood at some point during the year reported to spend 15 hours (SD: 11, Median = 12) per week collecting firewood during the summer, post-monsoon, and winter seasons, with the greatest amount of firewood collection time during the winter and the least amount of time during the monsoon season (Table 4). Only a relatively small number of households (2.2% across all seasons and 3.0% in winter) reported to explicitly collect firewood for space heating purposes.

4.4. Understanding the perception that firewood collection has become more difficult

Qualitative results indicate that restricted access to firewood and a lack of forest were the top reasons for perceived increased difficulty in firewood collection. Yet, the average amount of forest outside of PAs within 2.74 km of villages where households reported increased difficulty in firewood collection over the last 5 years was significantly higher than for households who reported that firewood collection had gotten easier or not changed (Table S6). To account for this non-intuitive relationship, we included both forest availability and increased difficulty collecting firewood as covariates as we believe both encompass important aspects of decision-making related to cooking fuel use. The perceived change in difficulty to collect firewood variable captures perceived shifts in environmental conditions rather than a more objective measure of firewood availability.

Although access to PAs varies spatially, 52% of survey respondents stated that restricted access to firewood was one of the top reasons for increased difficulty in firewood collection. Those who discussed restricted specifically mentioned the “forest department,” “forest guards,” “forest officer,” “government,” or “village committee” as enforcing these restrictions. For example, “forest department do not allow us to take the firewood from the forest,” “government started to protect forest areas,” and “village committee not allowing us to go into forest.” Therefore, we excluded forest cover within PAs in our measure of forest available for firewood collection. Only 4% of villages contained a PA within 2.74 km of their boundary. The lack of forest was also discussed in 28% of responses as a driver of the increased difficulty in collecting firewood. For example, “There is no firewood in the forest these days” and “much less firewood in the forest and we are not allowed to enter into the forest.”

4.5. Determinants of firewood collection patterns

In response to the distinct seasonal pattern in firewood collection, we conducted analyses that considered firewood collection per week as the outcome for the monsoon season and the average time across summer, post-monsoon, and winter seasons separately. The use of LPG for cooking was significantly negatively associated with the number of hours per week spent collecting firewood, accounting for other covariates (Figure 4). In the summer, post-monsoon, and winter seasons, households without LPG spent 17 hours (SD: 11, Median = 14) per week collecting firewood compared to households with LPG that spent 13 hours (SD: 9.5, Median = 9.8) per week (Table 5).

Figure 4.

Coefficient plots for OLS regressions with District Fixed Effects. Regressions assessed the association between LPG ownership and time spent collecting firewood averaged across three seasons (summer, post-monsoon, and winter) and in monsoon. Points represent exponentiated coefficient estimates (percent change in time spent collecting firewood) and whiskers show 95% confidence intervals.

Table 5.

Hours spent collecting firewood per week by season and timing of LPG adoption.

	2013 or before	2014–2015	2016–2017	No LPG
N	270	277	1729	2718
Summer, post-monsoon, and winter
Mean (SD)	12.79 (10.54)	10.75 (6.92)	15.74 (11.05)	16.50 (11.18)
Median (IQR)	8.17 (6.00, 16.00)	8.83 (6.00, 13.92)	12.33 (7.75, 20.67)	14.00 (8.00, 21.50)
Monsoon
Mean (SD)	9.55 (7.51)	9.07 (6.63)	10.50 (8.03)	10.12 (8.50)
Median (IQR)	7.00 (4.25, 12.00)	8.00 (4.00, 12.00)	8.00 (5.00, 14.00)	8.00 (5.00, 12.00)

SD is standard deviation and IQR is interquartile range.

The more years a household owned LPG, the stronger the negative association of LPG use was with reported time collecting firewood. For example, households that adopted LPG in 2013 or before reported spending 53% fewer (95% CI: 47 – 59%) hours per week collecting firewood than households without LPG in the summer, post-monsoon, and winter seasons. Similarly, households adopting LPG in 2014 – 2015 and those in 2016 – 2017 spent 46% less (95% CI: 39 – 53%) and 14% less (95% CI: 8.6 – 19%) time collecting firewood than households without LPG, respectively. While the percent change in time spent collecting firewood was not meaningfully different between the “adopted LPG in 2013” and the “adopted LPG in 2014 – 2015” categories, both of these groups reported spending significantly less time collecting firewood than households adopting LPG in “2016 – 2017.” These households (2016 – 2017 adopters), in turn, spent significantly less time collecting firewood than households that did not use LPG. We observe consistent results when carrying out analyses for time spent collecting firewood during the monsoon season, but with attenuated effect sizes, perhaps owing to comparatively less overall firewood collection during this season.

Monthly expenditure, level of formal education, and household tribe and caste status were associated with time spent collecting firewood. The reported time spent collecting firewood each week was shorter by 4.7% (95% CI: 1.9 – 7.4%) with every 1% increase in monthly expenditure. Compared to a household with no formal education, when the household head obtained a primary or secondary, high school, or intermediate and above education, the household spent 6.5% (95% CI: 0.1 – 12%), 13% (95% CI: 4.2 – 21%), and 33% (95% CI: 25 – 39%) less time collecting firewood, respectively. Scheduled Tribe households spent 51% more (95% CI: 29 – 76%) hours collecting firewood than those in the general caste. In addition, SC and OBC households spent 41% (95% CI: 19 – 67) and 18% (95% CI: 1.5 – 38) more hours collecting firewood than those in the general caste, respectively.

Village-level tree cover and the perception that firewood collection had gotten more difficult over the last five years were associated with time spent collecting firewood. Accounting for other covariates, 1% greater forest cover was associated with 26% (95% CI: 22 – 30%) more reported firewood collection time. Households reporting increased difficulty in firewood collection reported to spend 67% more (95% CI: 53 – 82%) time collecting firewood as compared to households that did not perceive increased difficulty.

5. Discussion

This study examines LPG use and firewood collection in marginalized populations living in forested regions within the Central Indian Highlands Landscape. Education, monthly household expenditures, and tribe and caste status were strongly associated with the use of LPG. In particular, households belonging to the Scheduled Tribe designation had the lowest probability of using LPG, had adopted LPG most recently, and reported to spend the most time collecting firewood, even after controlling for other covariates. While almost all LPG users continue to collect and cook with firewood, more years cooking with LPG was associated with less firewood collection, suggesting a waning reliance on firewood for cooking. Finally, households near higher tree cover had lower odds of using LPG, adopted LPG more recently, and spent more time collecting firewood.

While India’s energy policies have focused more on expanding clean cooking than equitable access, our study finds LPG ownership increased in marginalized, less-formally educated, and poor households after PMUY. Of all households that use LPG for cooking, a ST household was 18 percentage points more likely to have acquired LPG after PMUY as compared to a general caste household. Nonetheless, the probability of using LPG overall for cooking was 7 – 14 percentage points lower among SC, OBC, and ST households as compared to general cate households. We contribute to growing evidence that, despite overall growth in LPG use owing to PMUY, disparities in access to cleaner cooking remain between social groups and across wealth gaps in India [50,79,80].

Consistent with case studies around the world that show the persistent role of biomass for cooking after the introduction of a clean fuel [12,17], households in the CIHL continue to rely on firewood for cooking despite the recent penetration of LPG. And yet, approaching the near-complete cessation of biomass use is a top priority to achieve cleaner indoor air [32,81]. However, the associations we find between length of time a household owns clean cooking fuel and traditional firewood collection are encouraging for future clean energy adoption and use in India. Still, ST, SC, and OBC households spent 18% – 51% more time collecting firewood than general caste households in the CIHL, even after controlling for years of LPG ownership and economic and demographic characteristics.

This study reached a population that lives in villages within forested regions, where households generally rely strongly on nearby forest products. Our results suggest that in the CIHL, biomass availability promotes firewood collection and hinders LPG use, and there should be further research on these associations. There is evidence from other regions of India that indicates that replacing firewood as a cooking fuel can generate positive environmental outcomes. For example, in South India, forest biomass was greater around communities where households cook with biogas [82]. Including communities living near forest in LPG expansion policies can expand the use of LPG in households that traditionally rely on biomass and may otherwise be unlikely to fully substitute firewood with LPG. In the CIHL, households near higher forest cover were more likely to own LPG after PMUY than other households with LPG. Expanding LPG ownership in households within forested regions should continue and be prioritized as a selection criterion for future LPG promotion policies.

Rural India is comprised of diverse communities where further attention on equity could help achieve energy justice. One potential strategy for increased LPG use would be targeted LPG subsidies or enhanced availability of LPG cylinder refills for specific groups and regions. For example, the amount of LPG subsidy might be linked to the highest education level attained by the household head, caste status, or monthly expenditures. While our study was restricted to largely ST and other lower caste communities in CIHL, our results have broader implications by motivating additional place-based analyses of barriers to clean cooking fuel adoption in recognition of the importance of household-level socioeconomic characteristics on cooking fuel choice.

The ability to afford clean cooking fuel is affected by income, which for the marginalized, rural populations in CIHL has traditionally depended on the extraction of forest goods. Increasing employment opportunities for this population can increase their capacity to use LPG and alleviate the burden of collecting firewood, which requires substantial time and effort. Women and children, household members who are generally responsible for firewood collection, in particular could experience further benefits along with a decreased burden of biomass collection. As argued elsewhere [38], clean cooking policies should consider the role of broader rural economic development and efforts to enhance education and women’s empowerment. A multifaceted approach to increase the use of clean cooking fuels that includes generating employment and providing education opportunities will have widespread benefits beyond clean energy access such as human capital development and gender equality. In the CIHL, the ST and other lower caste communities who face disparities in education and poverty would particularly benefit from a comprehensive rural development and clean cooking program.

5.1. Limitations and future areas of research

This study has a few limitations worth noting. First, we do not have multiple measures of cooking fuel use or firewood collection patterns over time. The cross-sectional nature of our study is limiting in two main ways: (1) we are limited in assessing self-reported historical patterns so we cannot capture the changing trade-offs between LPG and firewood use in the years since LPG adoption (e.g., waning reliance on firewood) and (2) we do not capture the precise seasonal patterns of cooking fuel use (e.g., LPG used more in the rainy season) and the corresponding seasonal determinants of fuel use (e.g., variable fuel availability, time-varying incomes). Panel surveys that visit households more than once over the course of many years and studies employing high-frequency surveys across a full year can offer valuable insights into the trajectories of fuel consumption patterns and their determinants [52,54,83,84].

While self-reported measures of cooking fuel use are at risk of survey bias because LPG is socially desirable across India, we do not use continuous measures of LPG use, such as cylinder refills per year, that may be at risk of over-reporting [85]. It is unlikely that participants differentially reported firewood collection based on ownership of LPG particularly because we calculate time spent collecting firewood using two questions (number of days visiting the forest and hours per trip). Nonetheless, while our comparisons across groups are not likely to be systematically biased, estimates of firewood collection intensity may contain errors and should be interpreted with caution.

Additionally, in our focus on forest-fringe communities, we specifically asked participants if they collected firewood from the forest at least weekly. However, about 20% of households did not report to collect firewood from the forest on a weekly basis during any season of the year. While we do not know how these households acquired firewood for their energy needs, there are a few possibilities: (1) they collected firewood less frequently than once per week; (2) they did collect firewood but not from the forest; (3) they received firewood from friends or family free of cost; or (4) they purchased firewood. Reported collection of firewood from the forest for sale was very rare in our sample (~1%), however, it is possible that firewood collection for sale was underreported due to the illicit nature of that activity. Nonetheless, future studies should investigate the possibility of less frequent firewood collection, firewood collection from non-forest sources, and the potential for a rural firewood market.

Additional limitations are that precise locations for firewood collection was not recorded and tree cover instead of biomass was used as a proxy for firewood availability. Greater specificity on firewood collection location and measures of biomass within those locations could enable even more precise estimates of the effects of biomass availability on household energy choices. We also find a mismatch between our satellite-derived measure of forest availability and perceptions of firewood availability. Households reporting increased difficulty in firewood collection in the last five years also lived near greater tree cover as compared to households who reported no change or increased ease in firewood collection. Change in perceptions of the ease of firewood collection does not directly represent the burden of firewood collection. Highly forested areas may be more likely to become less forested – and therefore firewood collection more difficult – than areas with already reduced forest availability. In addition, the ease of firewood collection may be determined by more factors than availability of forest, such as forest management systems that occur outside of PAs or interactions with neighboring communities. Finally, perceptions of environmental change, such difficulty in firewood collection, may be constructed from socio-cultural practices or cognitive biases and influenced by survey questions [86,87].

Additional research on drivers of perceptions of firewood availability and how households make cooking fuel decisions based on these perceptions along with availability of biomass may provide clarity into strategies for reducing the use of biomass-based fuels in forest-fringe communities that stack fuels. Furthermore, future studies could employ temporally-resolved forest cover measures and multiple surveys to enable panel models capable of capturing assessing forest cover dynamics and within-household shifts in fuel choices, firewood collection patterns, and perceived changes in firewood availability. The continued collection and demand for firewood, socioeconomic drivers of LPG ownership, and differences between measured and perceived biomass availability indicate complexities in behavioral transitions and energy access that impact cooking fuel use and require further exploration.

6. Conclusion

Our research contributes to a broader understanding of fuel stacking and incorporates tenets of energy justice into clean cooking fuel access within India. In households that use LPG and firewood to cook, the time spent to collect firewood was lowest among households that owned LPG the longest. Even households that adopted LPG most recently (2016 – 2017) spent significantly less time collecting firewood than households without LPG. While PMUY increased access to LPG in Indian households overall, and in our study sample, disparities in LPG access for ST and OBC populations remain. In addition to the disparities in LPG access that rural Indian households face by social group, we find that education level, income, and proximity to forest impact the use of LPG. Our findings suggest that incorporating an explicit motive to address inequitable access for marginalized stakeholders in PMUY may further expand LPG access, displace firewood use, and ultimately improve livelihoods in the CIHL. Similar approaches that examine barriers and inequalities in social groups can inform targeted clean cooking fuel expansion policies around the world.