This is our 12th blog post for our Job Market Paper Series blog for 2024-2025.
T. V. Ninan is a PhD candidate in Economics at the University of Washington, Seattle. His interests span environment, development and education. You can find his JMP here.
Around 2.6 billion people worldwide rely on solid fuels like wood, dung and charcoal for cooking, leading to severe indoor air pollution (IAP). IAP is one of the biggest health hazards in low- and middle-income countries, accounting for more than 3.2 million deaths in 2020 alone. But beyond the immediate health risks, there’s another, less visible cost to cooking with solid fuels: it can hinder children’s learning, as I show in my job market paper. To do this, I evaluate the Pradhan Mantri Ujjwala Yojana (PMUY), a program launched in 2016 in India that dramatically increased access to clean cooking technology. By 2021, PMUY had provided over 100 million subsidized Liquified Petroleum Gas (LPG) connections to socioeconomically marginalized households across India, making it the largest clean cooking transition initiative in the world.
How does cooking technology impact learning?
Solid fuel emission from homes is among the leading contributors to mortality caused by ambient pollution in poor countries (Yun et al., 2020). More worryingly, during cooking, PM2.5 exposure skyrockets to levels more than twice the concurrent ambient concentrations, far exceeding even the least stringent air quality target set by the WHO (Somanathan et al., 2022; Berkouwer and Dean, 2024). Coupled with recent evidence linking ambient air pollution to reduced learning outcomes (Balakrishnan and Tsaneva, 2021), these facts suggest a negative relationship between solid fuel use and learning. This may stem both from reduced cognitive abilities in children, or lower school attendance and time spent on other activities that boost learning (such self-studying).
Additionally, switching to LPG can lead to time savings for the primary cook which could be reallocated towards activities that support children’s learning (Krishnapriya et al., 2021), such as helping with schoolwork or dropping children to school. However, if school is viewed as childcare, then more free time for the mother could reduce school attendance, negatively impacting learning. Thus, the effect stemming from a change in primary cook’s time use could be positive or negative based on which channel dominates. Additionally, if children help with chores, LPG adoption may lead to time savings for them too, likely benefitting their learning. Lastly, clean fuels are more expensive than solid fuels, and if the cost of fuel diverts household resources away from education, this would hinder learning. Ultimately, the overall impact on learning will depend on which of these effects dominate.
Cooking Fuel in India and PMUY
In 2015, India had the highest number of people primarily dependent on solid fuels for cooking. To address this, in May 2016, the government launched PMUY nationally. This initiative waived off the security deposit, installation costs and equipment expenses for new LPG connections issued to eligible households. Additionally, these households were provided interest free loans for the first refill and hotplate, with the loan to be repaid using subsidies received for subsequent refills from the government.
The intensity of the program in a district was determined by several factors, like the number of below poverty line households without an LPG connection, the proximity of LPG distributors, awareness of the program, and the ability to provide necessary documentation. Overall, the program was highly successful in expanding LPG access across India. In 2015, before the program, LPG coverage stood at around 56%. By 2022, coverage had increased to over 99%.
Higher Access to LPG boosts learning outcomes
There was considerable variation in the intensity of the PMUY program at the district level by 2021, with the proportion of LPG connections per 100 households ranging from 0.6 to 78 with a median of 29. I use this variation in treatment intensity in a difference-in-differences framework with district and year fixed effects and control for district levels of educational outcomes interacted with a time trend and demographic composition of the district to evaluate the program’s impact. Learning outcomes were obtained from the Annual Status of Education Report which provides district-representative information on foundational learning in India. Using 2011-2014 as the pre-intervention period, I find that a 10-percentage point (pp) increase in program coverage leads to an improvement of 0.032 and 0.027 standard deviations in reading and math indices respectively (Figure 1). The average district had a PMUY coverage of 31%, which implies an effect size of about 0.1 standard deviations. This is particularly noteworthy as 0.1 standard deviations is the median effect for education interventions in a large meta-analysis of studies in developing countries (Evans and Yuan, 2022).
I also find significant improvements in each component of the indices (Figure 2). All skills improved, with larger effects as the skills get more challenging. A commonly cited metric for learning is the proportion of class 5 students that can perform class 2 level skills. The class 5 reading and math skill showed an improvement of 2.6pp each for a 10pp improvement in PMUY coverage. This translates to an impressive 6% improvement in reading and a 10% improvement in the ability to do division.
I also find that the positive impact of the program is notably higher for girls and for households in the middle of asset distribution, both of which align with the program’s design. In a patriarchal society like India, girls tend to spend more time indoors than boys and are therefore more likely to benefit from a reduction in IAP. Furthermore, while the program targeted marginalized households, the effect size would be the largest on those households that are poor enough to be targeted yet have enough means to afford regular LPG refills, which is exactly what I find.
What can we say about the main mechanisms?
Given the positive program effects, we can rule out mechanisms that hinder learning as dominant channels. In addition, I provide three reasons for why the dominant mechanism is likely an increased participation in school, either because children are healthier or due changes in the primary cook’s time use.
First, my analysis focuses on children who are enrolled in schools and are aged 10 or under. They are less likely to be helping with cooking and related activities like firewood collection, minimizing the impact of children’s time use on learning outcomes. Second, I find an increase in children’s school attendance because of the program, which indicates that primary cooks do not see school merely as childcare. Third, I find no impact on the learning outcomes of unenrolled children, ruling out mechanisms that are not related to school enrollment, such as cognitive improvements or mothers homeschooling kids.
Reliability of Results and Future Research
To identify unbiased effects of PMUY, we must assume that high intensity and low intensity districts would follow similar trends in the absence of PMUY. I show that the trends were parallel for all pretreatment years except a couple of early years where a large education reform happened in India. I argue that this reform happened in the early years and affected treatment and control districts differently initially but had no differential impact thereafter. I conduct several checks to support my identifying strategy, including analyzing a subset of states for which trends were parallel throughout pre-intervention, a synthetic difference-in-differences analysis and performing sensitivity analyses for potential violations from the parallel-trends assumption. All these approaches yield qualitatively consistent results with the main specification. Ideally one would also want to control for time varying factors that caused differential learning trends post the education reforms but evidence on what these factors are is limited. Future research could examine this aspect. In addition, future projects could collect data on time use and health outcomes, to completely disentangle the mechanisms driving the learning improvements. The suggestive evidence I provide points to the direction such research must take.
Policy Relevance
My JMP suggests that clean transition policies have benefits beyond health—they can also improve educational outcomes. There is broad consensus that education is essential for sustained economic development. However, despite a significant increase in primary school enrollment over the past 60 years, evidence shows that learning outcomes in many low- and middle-income countries remain alarmingly low. These countries also have some of the highest numbers of solid fuel users, causing a dual challenge. This paper shows that tackling one issue can help address the other. By ensuring that every household has access to clean cooking, we can create healthier homes as well as improve children’s learning.
Feature image taken by T. V. Ninan during his field visits.
