Exploration of implicit weights in composite indicators : The case of resilience assessment of countries’ electricity supply

Sammanfattning: Composite indicators, also called indices, are widely used synthetic measures for ranking and benchmarking alternatives across complex concepts. The aim of constructing a composite indicator is, among other things, to simplify and condense the information of a plurality of underlying indicators. However, to avoid misleading results, it is important to ensure that the construction is performed in a transparent and representative manner. To this end, this thesis aims to aid the construction of the Electricity Supply Resilience Index (ESRI) – which is a novel energy index, developed within the Future Resilient Systems (FRS) programme at the Singapore-ETH Centre (SEC) – by looking at the complementary and fundamental component of index aggregation, namely the weighting of the indicators. Normally, weights are assigned to reflect the relative importance of each indicator, based on stakeholders’ or decision-makers’ preferences. Consequently, the weights are often perceived to be importance coefficients, independent from the dataset under analysis. However, it has recently been shown that the structure of the dataset and correlations between the indicators often have a decisive effect on each indicator’s importance in the index. In fact, their importance rarely coincides with the assigned weights. This phenomenon is sometimes referred to as implicit weights. The aim of this thesis is to assess the implicit weights in the aggregation of ESRI.  For this purpose, a six-step analytical framework, based on a novel variance-based sensitivity analysis approach, is presented and applied to ESRI. The resulting analysis shows that statistical dependencies between ESRI’s underlying indicators have direct implications on the outcome values – the equal weights assigned a-priori do not correspond to an equal influence from each indicator. Furthermore, when attempting to optimise the weights to balance the contribution of each indicator, it is found that this would require a highly unbalanced set of weights and come at the expense of representing the indicators in an effective manner. Thereby, it can be concluded that there are significant dependencies between the indicators and that their correlations need to be accounted for to achieve a balanced and representative index construction. Guided by these findings, this thesis provides three recommendations for improving the statistical representation and conceptual coherence of ESRI. These include: (1) avoid aggregating a negatively correlated indicator – keep it aside, (2) remove a conceptually problematic indicator – revise its construction or conceptual contribution, and (3) aggregate three collinear and conceptually intersecting indicators into a sub-index, prior to aggregation – limit their overrepresentation. By revising the index according to these three recommendations, it is found that ESRI showcases a greater conceptual and statistical coherence. It can thus be concluded that the analytical framework, proposed in this thesis, can aid the development of representative indices.