Multiple imputations of missing data in the environmental sustainability index - pain or gain?
by
Tanja Srebotnjak
United Nations Statistics Division, New York, USA
Coauthors: Center for International Earth Science Information Network (CIESIN), Columbia University, USA

The United Nations Conference on Environment and Development (UNCED, Rio de Janeiro, 1992) brought about a rapidly growing need for meaningful, aggregated measures of sustainable development. This demand is reflected in the Environmental Sustainability Index (ESI)[1], an index that aims at measuring a country's performance in sustaining a healthy, natural environment.

The talk focuses on the problem of missing data in the construction of the index. Ad-hoc solutions such as listwise or pairwise deletion, mean substitution, best guess imputation or regression methods were shown to result in the loss of valuable information, potential selection bias, as well as an underestimation of the uncertainty about predictions of the missing values (Schafer 1997).

We propose the application of multiple imputations (Rubin 1976) using MCMC as an alternative to these methods. Multiple imputations have been successfully used in a wide range of missing data situations, avoiding the above-cited problems and under certain model assumptions yielding consistent and more efficient results. Based on the simulations, we compare the index and some of its properties with the original ESI 2001 and derive an estimate of the precision of the index.

[1] The ESI is an initiative of the Global Leaders for Tomorrow Environment Task Force, The World Economic Forum, in collaboration with the Yale Center for Environmental Law and Policy (YCELP), Yale University, and the Center for International Earth Science Information Network (CIESIN), Columbia University. More information can be found on: http://www.ciesin.columbia.edu/indicators/ESI/

United Nations Statistics Division

Date received: August 15, 2001