To make the 25 indicators comparable, each metric was converted to a proximity-to-target- measure with a range of 0 to 100.
Initially, we examined the distribution of each indicator to identify whether extreme values skew the aggregations of some indicators. Extreme outliers (greater than or equal to three standard deviations from the mean) are more likely to be the result of data processing (especially for modeled data) than actual performance. Accordingly, we adjusted outliers using a recognized statistical technique called winsorization – in this case trimming at the 95th percentile of the distribution. In a small number of cases even this level of winsorization left significant outliers, and in such cases we winsorized at a greater level based on a comparison of the two alternative values (see Methodology section).
A second decision concerned the treatment of countries that exceeded the long-term performance or sustainability target. To avoid rewarding “over-performance,” no indicator values above the long-term target were used. In the few cases where a country did better than the target, the value was reset so that it was equal to the target. Once those two adjustments were made, a simple arithmetic transformation was undertaken: the observed values were placed onto a zero to 100 scale where 100 corresponds to the target and zero to the worst observed value.
Aggregation is an area of inescapable methodological controversy. While the field of composite index construction has become a well-recognized subset of statistical analysis, there is no clear consensus on how best to construct composite indices. Various aggregation methods exist, and the choice of an appropriate method depends on the purpose of the composite indicator as well as the nature of the subject being measured.
To help identify appropriate groupings and weights for each indicator, we carried out a principal component analysis (PCA). Most categories did not have clear referents in the PCA results. Absent a PCA-derived basis for weighting the indicators, equal weights were used with some refinements determined by the EPI team with expert guidance.
The Environmental Health and Ecosystem Vitality subcategories each represent 50% of the total EPI score. This equal division of the EPI into issues related to (1) humans and (2) nature is not a matter of science but rather policy judgment. But this even weighting of the two overarching objectives of environmental policy reflects a widely-held intuition, and this choice (used in the 2006 Pilot EPI) has not been generally criticized. Indeed, for every “deep ecologist” who favors more weight being placed on Ecosystem Vitality, there is a “humans first” environmental policymaker who prefers that the tilt go the other way.
Within the Environmental Health Objective/Policy Category, the Environmental Burden of Disease (DALY) indicator is weighted 50% and accordingly contributes 25% of the overall EPI score, because it is widely regarded to be the most comprehensive and carefully-defined measof environmental health burdens. The effects of Water and Air Pollution on human health comprise the remainder of the Environmental Health subcategory and are each allocated a quarter of the total score for Environmental Health, reflecting a widespread policy consensus.
The two water-related Environmental Health indicators (Adequate Sanitation and Drinking Water) are equally weighted. In the Air Pollution sub-category, Urban Particulates and Indoor Air Pollution receive equal weights, and double the weight given to the effects of ground-level Ozone on human health. Urban particulates and indoor air pollution are widely acknowledged by the United Nations Environment Programme (UNEP), World Health Organization (WHO), and United Nations Children’s Fund (UNICEF) as important indicators of the burden of air pollution on human health. There is, however, a growing literature that suggests a link between ozone exposure and human health. Our human exposure to ozone metric assesses person-days of exposure per year to ground-level ozone exceeding 85 parts per billion (ppb). Because this indicator is experimental, we give it half the weight of those with known reliability.
Within the Ecosystem Vitality Objective, the Climate Change indicator carries 50% of the weight (i.e., 25% within the total EPI). This is owing to the increasing importance attached to climate change in policy discussions, and its potential to have far reaching impacts across all aspects of ecosystem vitality and natural resource management. The Air Pollution (effects on ecosystems) policy category is weighted at 5% of the Ecosystem Vitality Objective. This slightly lower weight when compared to water, biodiversity, and productive natural resources is owing to the fact that Air Pollution is already partially captured in the Environmental Health Objective. The remaining indicators: Water, Biodiversity, and Productive Natural Resources, are each evenly weighted to cover the remaining 22.5% of the Ecosystem Vitality Objective.
