As household surveys are conducted irregularly and reported heterogeneously, the WHO Global Household Energy Model (developed in collaboration with the University of Exeter, UK and maintained in collaboration with the University of Glasgow, UK) is employed to estimate trends in household use of six fuel types:
- unprocessed biomass (e.g., wood)
- charcoal
- coal
- kerosene
- gaseous fuels (e.g., LPG)
- electricity
Trends in the proportion of the population using each fuel type are estimated using a Bayesian hierarchical model, with urban and rural disaggregation, drawing on country survey data. Smooth functions of time were the only covariate. Estimates for overall ‘polluting’ fuels (unprocessed biomass, charcoal, coal, and kerosene) and ‘clean’ fuels (gaseous fuels, electricity, as well as an aggregation of any other clean fuels like alcohol) are produced by aggregating estimates of relevant fuel types. Estimates produced by the model automatically respect the constraint that the total fuel use equals 100 percent.
GHEM is implemented using the R programming language and the NIMBLE software package for Bayesian modelling with Markov chain Monte Carlo (MCMC). Summaries can be obtained to provide both point estimates (e.g., means) and measures of uncertainty (e.g., 95 percent credible and 95 percent prediction intervals). The GHEM is applied to the WHO household energy database to produce a comprehensive set of estimates, together with associated measures of uncertainty, of the use of four specific polluting fuels and two specific clean fuels for cooking, by country, for each year from 1990 to 2020. Further details on the modelling methodology and validation can be found in Stoner et and others 2020, and more detailed analysis of individual fuel use can be found in Stoner and others 2021. The complete set of estimates can be downloaded from the WHO Global Health Observatory website.
Only surveys with less than 15 percent of the population reporting “missing”, “no cooking” and “other fuels” were included in the analysis. Surveys were also discarded if the sum of all mutually exclusive categories reported was not within 98-102%. Fuel use values were uniformly scaled (divided) by the sum of all mutually exclusive categories excluding “missing”, “no cooking” and “other fuels”. Countries classified by the World Bank as high income in the 2023 fiscal year were assumed to have transitioned to clean household energy. They are therefore reported as 100 percent access to clean fuel and technologies; no fuel-specific estimates were reported for high-income countries. In addition, no estimates were reported for low- and middle-income countries without data suitable for modelling (Bulgaria, Lebanon, and Libya). Modelled specific-fuel estimates were reported for 128 low- and middle-income countries plus three countries with no World Bank income classification (República Bolivariana de Venezuela, Niue, and Cook Islands), and estimates of overall clean fuel use were reported for 195 countries.