Uncertainty Analysis for a Social Vulnerability Index

Abstract

Indexes have gained favor over the past decade as a tool to measure social vulnerability to hazards. Numerous index designs have been put forward, yet we still know very little about their reliability. This research investigates the methods of social vulnerability index construction, examining decisions related to indicator selection, scale of analysis, measurement error, data transformation, normalization, and weighting. Each of these stages is imbued with uncertainty due to choices made by the index developer. The study applies Monte Carlo–based uncertainty analysis to assess and visualize uncertainty for a hierarchical social vulnerability index. Confidence limits are computed for the index rankings, leading to a finding of a high magnitude of uncertainty. The performance of the index compared to alternative configurations is strong in some places but statistically biased in about a third of the census tracts. The variability of index rankings is also assessed, indicating that index precision decreases with increasing vulnerability. Uncertainty analysis provides a useful, yet largely unapplied stage of index production that highlights places where the model is most reliable. If applied to the creation of social vulnerability indexes, output metrics can be produced with a greater degree of precision, transparency, and credibility.