Abstract:
|
The Environmental Sustainability Index (ESI) is a composite index designed to estimate a country's progress towards achieving environmental sustainability. The index, covering 142 countries in 2002, is based on 68 variables, which are organized along five dimensions considered to sufficiently define environmental sustainability. The aggregation algorithm of the ESI is complicated by the occurrence of missing values. Ad-hoc methods such as case deletion, mean substitution, or single linear regression have been shown to result in potential selection bias, underestimates of the covariance matrix, and inflated regression coefficients. We propose the application of multiple imputation techniques to the ESI data in order to avoid the above-mentioned problems and to obtain estimates of the uncertainty inherent in the missing values. The presentation investigates which imputation model reflects best the ESI dataset with respect to the imputation model, missing data generating processes, prior distributions on model parameters, and the grouping of observations and variables. Building on the selected model, the variance of the index is estimated and confidence intervals calculated.
|