Estimating Global Household Air Pollution: A Multivariate Hierarchical Model for Cooking Fuel Prevalence
share
Globally, an estimated 3.8 million deaths per year can be attributed to household air pollution. Information on the proportion of people relying primarily on different polluting fuels for cooking, which acts as a proxy for pollution exposure, is available in the form of nationally-representative household surveys. However, the absence of a modelling framework for comprehensively estimating the use of individual fuels inhibits fuel-specific policy interventions. To address this, we develop a multivariate hierarchical model (GHHEM) for data from the World Health Organization Household Energy Database, spanning the period 1990-2016. Based on Generalized-Dirichlet-Multinomial distributions, the model jointly estimates trends in the use of eight individual fuels, whilst addressing a number of challenges involved in modelling the data. These include: missing values arising from incomplete surveys; missing values in the number of survey respondents; and sampling bias in the proportion of urban and rural respondents. The model also includes regional structures to improve prediction in countries with limited data. We assess model fit using within-sample predictive analysis and conduct an out-of-sample prediction experiment to evaluate the model's forecasting performance. Overall, this work substantially contributes to expanding the evidence base for household air pollution, which is crucial for developing policy and planning interventions.
READ FULL TEXT
