Developing Composite Indicators for Flood Vulnerability Assessment: Effect of Weight and Aggregation Techniques

Abstract

The assessment of flood vulnerability involves multidimensional and complex interactions between environment, social and economic dimensions. Indicator-based vulnerability assessment such as the Flood Vulnerability Index (FVI) is widely used in vulnerability studies to summarise complexity and multidimensionality issues to gauge the level of vulnerability. To assess the various factors of vulnerability, we employed a set of 21 environmental, social and economic indicators to quantitatively assess the three factors of vulnerability, namely exposure, susceptibility and resilience to flood at the subnational level. The construction of the vulnerability index involves sequential steps, including selecting the indicators, their normalisation, weightage and aggregation to a final index. In this study, we looked into which weighting and aggregation technique as the most suitable to develop the final composite indicator. The weighting techniques employed are equal weight, unequal weight and principal component analysis. Two different aggregation techniques, namely additive mean and geometric mean were used to aggregate the indicators to a single index. This study employed reliability and sensitivity analyses to evaluate the robustness of the FVI constructed using various techniques. This study shows the wider application of the equal weighting and additive mean techniques to develop composite indicator for flood vulnerability assessment.

Computational procedures for quality assessment of latent concepts

“…With this aim, a simulation‐based average ranking shift index (SiARS) has been defined, similarly to Liew et al (2019), that is comparing the original rankings for the whole set of countries and those resulting from the application of each method in each random sample.…”

“…The stability of the ranking across the results of the simulation analysis is indeed a measure of reliability of the methodology. The ranking shift index in Liew et al (2019) is extended for the simulation study. Specifically, the impact of the size and composition of data on the variability of the latent construct estimates and the resulting ranking has been assessed.…”