Fuel stacking, housing quality, and health disparities in rural South Africa: A double machine learning approach

The relationship between fuel type, housing quality, and health outcomes is examined in rural South Africa using the 2019 General Household Survey dataset from Statistics South Africa (StatsSA). We apply double machine learning within a multivalued treatment effects framework to distinguish between clean, unclean, and mixed fuel types as treatment regimes, allowing us to infer health impacts from fuel-stacking behavior. Our findings show that using mixed fuels does not lead to health benefits from partially switching away from unclean fuels, which challenges common beliefs about “energy stacking” in low-income households. We also use Sorted Effects Analysis (SEA) to explore differences within the population and identify vulnerable groups most affected by the health risks of unclean fuel use. The SEA results reveal that the health effects of polluting fuels are not uniform, and there is some evidence that housing quality—such as roofing and walls—moderates the relationship between polluting fuels and health outcomes. These findings suggest that effective policies should consider the roles of fuel type, housing quality, and socioeconomic vulnerabilities to promote a fair and equitable transition to clean energy in rural South Africa. • Fuel stacking does not lower health risks compared to using polluting fuel. • Housing construction quality influences the fuel-health relationship. • Machine learning analysis shows that fuel impacts vary significantly across households. • Transitioning to clean fuel sources requires housing upgrades to maximize health benefits.