Global Metrics, Local Estimation: Magnifying the Health Impact of Environmental Justice

Abstract

In the first chapter, Mapping development and health effects of cooking with solid fuels in low-income and middle-income countries, 2000–18: a geospatial modeling study, the prevalence of solid-fuel use for cooking is mapped at a 5 km à 5 km resolution in 98 LMICs based on 2.1 million household observations of the primary cooking fuel used from 663 population-based household surveys over the years 2000 to 2018. We use observed temporal patterns to forecast household air pollution in 2030 and to assess the probability of attaining the Sustainable Development Goal (SDG) target indicator for clean cooking. We aligned our estimates of household air pollution to geospatial estimates of ambient air pollution to establish the risk transition occurring in LMICs. Finally, we quantified the effect of residual primary solid-fuel use for cooking on child health by doing a counterfactual risk assessment to estimate the proportion of deaths from lower respiratory tract infections in children younger than 5 years that could be associated with household air pollution. We found that while reliance on solid-fuel use for cooking has declined globally, it remains widespread. 593 million people live in districts where the prevalence of solid-fuel use for cooking exceeds 95%. 66% of people in LMICs live in districts that are not on track to meet the SDG target for universal access to clean energy by 2030. Household air pollution continues to be a major contributor to particulate exposure in LMICs, and rising ambient air pollution is undermining potential gains from reductions in the prevalence of solid-fuel use for cooking in many countries. We estimated that, in 2018, 205 000 (95% uncertainty interval 147 000–257 000) children younger than 5 years died from lower respiratory tract infections that could be attributed to household air pollution. The second chapter, Scales of environmental justice: global sensitivity analysis of the Washington Environmental Health Disparities Map, is focused on understanding the factors that drive environmental health inequalities more locally, within the context of the Washington State Environmental Health Disparities (EHD) Map, a composite indicator of environmental justice that synthesizes 19 different environmental and population health indicators to generate cumulative impact rankings by census tract. We conducted a global sensitivity analysis of the EHD mapping methodology by permuting across alternative methods derived from the composite indicator construction literature, generating estimates of the uncertainty that results from analyst decisions in the development process. We estimated first, second and total order sensitivity statistics to quantify the relative influence of parameter choices on the tract rankings and on the accuracy of classifying communities in the top 20% of impact. On average, census tracts changed by more than one hundred ranks across these permutations. The observed deviations from the baseline EHD index were largest in the middle of the impact spectrum and smallest for tracts in the top 10% of impact. The formula used in aggregation and the method of data normalization were the most sensitive parameter decisions for both tract ranking and impact classification. We demonstrate that the EHD rankings were more robust in the highest impact tracts and relatively uncertain throughout the rest of the index, suggesting that this data is better suited for classifying hotspots than for estimating an ordinal impact. There are strong assumptions underlying the baseline EHD ranking methodology and these assumptions substantially drive the results In the third chapter, Validating the structure of an environmental justice index in Washington State, a multivariate case study, we explore the mechanisms through which the most sensitive parameter choices in the EHD methodology impact the final index. The statistical characteristics of the raw indicator data are analyzed in order to understand the influence of various ranking transformations on the data distributions, and case studies that exhibit large changes in ranking and impact classification between linear and nonlinear transformations are analyzed to assess the bias introduced by normalization methodology. Nonlinear transformations are observed to favor the high-impact classification of tracts with above average values for the majority of indicators while reducing the effect of outliers. A variance-weighted Principal Component Analysis (PCA) is employed to compare the results of inductive aggregation to the more theoretically derived baseline index. A PCA based index is observed to agree generally with the baseline index and classify the impacted tracts with high accuracy using only the first two principal components. The loading structure of individual indicators within these first two components suggests that twin gradients of urban environmental degradation and socioeconomic deprivation are driving the index rankings in the current formulation of the EHD map.