Skip to main content

Improving stove evaluation using survey data: who received which intervention matters

Valerie Mueller, Alexander Pfaff, John Peabody, Yaping Liu, Kirk R. Smith
Ecological Economics 93 (2013) 301–312

PDF link iconAs biomass fuel use in developing countries causes substantial harm to health and the environment, efficient stoves are candidates for subsidies to reduce emissions. In evaluating improved stoves’ relative benefits, little attention has been given to who received which stove intervention due to choices that are made by agencies and households. Using Chinese household data, we find that the owners of more efficient stoves (i.e., clean-fuel and improved-biomass stoves, as compared with traditional-biomass and coal stoves) live in less healthy counties and differ, across and within counties, in terms of household characteristics such as various assets. On net, that caused efficient stoves to look worse for health than they actually are.We control for counties and household characteristics in testing stove impacts. Unlike tests that lack controls, our preferred tests with controls suggest health benefits from clean-fuel versus traditional-biomass stoves. Also, they eliminate surprising estimates of health benefits from coal, found without using controls. Our results show the value, for learning, of tracking who gets which intervention.


Demonstrating bias and improved inference for stoves’ health benefits

Valerie Mueller, Alexander Pfaff, John Peabody, Yaping Liu, Kirk R. Smith
International Journal of Epidemiology (2011) 1–9 [doi:10.1093/ije/dyr150]

PDF link iconBACKGROUND: Many studies associate health risks with household air pollution from biomass fuels and stoves. Evaluations of stove improvements can suffer from bias because they rarely address health-relevant differences between the households who get improvements and those who do not. METHODS: We demonstrate both the potential for bias and an option for improved stove inference by applying to household air pollution a technique used elsewhere in epidemiology, propensity-score matching (PSM), based on a stoves-and-health survey for China (15 counties, 3500 households). RESULTS: Health-relevant factors (age, wealth, kitchen ventilation) do in fact differ considerably between the households with stove improvements and those without. We study the resulting bias in estimates of cleaner-stove impacts using a self-reported Physical Component Summary (PCS). Typical stoves-literature regressions with little control for non-stove factors suggest no benefits from a cleaner-fuel stove relative to a traditional biomass stove. Yet increasing controls raises the impact estimates. Our PSM estimates address the differences in health-relevant factors using ‘apples to apples’ comparisons between those with improved stoves and ‘similar’ households. This generates higher estimates of clean-stove benefits, which are on the order of one half the standard deviation of the PCS outcome. CONCLUSIONS: Our data demonstrate the potential importance of bias in household air pollution studies. This results from failure to address the possibility that those receiving improved stoves are themselves prone to better or worse health outcomes. It suggests the value of data collection and of study design for cookstove interventions and, more generally, for policy interventions within many health outcomes.