Data-driven weights and restrictions in the construction of composite indicators

Abstract

Abstract. Composite indicators are increasingly recognized as a useful tool in policy analysis and public communication. However, if poorly constructed, they can send misleading policy messages. Perhaps the most difficult aspect of constructing a composite indicator is choosing weights for the components. The categorization of Croatian territorial units for development policy is based on the value of the composite indicator called the development index. The main goal of this paper is to propose an empirical approach for weight selection. In order to generate the set of non-subjective weights, principal component analysis and linear programming methods have been applied. An application of data envelopment analysis to the field of composite indicators, known as the Benefit-of-the-Doubt approach, has been demonstrated subject to proportional subindicator share restrictions. Additionally, the Monte Carlo simulation of weights was conducted, and confidence intervals for the values of the development index were estimated. Owing to the fact that the examined weighting schemes have resulted in the different categorization of territorial units, use of unit-specific weights and incorporating uncertainty in the construction of a composite indicator looks promising for further work.

A Sustainability Index of Mining Countries

“…Figure 3 thus indicates that the SIMC relies both on a sound theoretical basis and provides stable statistical results. This result is also interesting because most of previous studies (see Desai 1991;Coste et al 2005;Zhou et al 2006;Brand et al 2007;OECD and JRC 2008;Cherchye et al 2008;Cherchye et al 2008;Nathan et al 2012;Chinn and Ito 2008;Tarantola and Vertesy 2012;Perišić 2015;Dialga and Le 2016) pointed out that the composite scores are unstable when both aggregation and weighting method change.…”

The index of economic freedom: methodological matters

“…The PCA and the BOD are two statistical methods increasingly used in the construction of composite indicators to overcome certain methodological limitations including in the steps of weighting and aggregation (Dialga and Hang Giang, 2016; Perišić, 2015; Zanella et al , 2015; Coste et al , 2005; Shwartz et al , 2010; OECD and JRC, 2008). Both approaches get their merits in that they allow the generation of weights following the data structure of the initial variables.…”

“…They show that the composite cores resulting from PCA and BOD techniques dramatically diverge. Studies of Perišić (2015); Vyas and Kumaranayake (2006); Zanella et al (2015) pose general frameworks of the application of both methods to weight and to aggregate composite indices.…”