The narrowly defined composite indexes described here represent the best way of summarising discrete, qualitative information. “Composite indexes are much easier to interpret than trying to find a common trend in many separate indicators” (Nardo et al., 2005[1]). However, their development and use can be controversial. These indexes are easily and often misinterpreted by users due to a lack of transparency about how they are generated, which makes it difficult to truly unpack what they are actually measuring.
The OECD has taken several steps to avoid or address common problems associated with composite indexes. The composites presented in this publication were developed using the steps identified in the Handbook on Constructing Composite Indicators (OECD/European Union/EC-JRC, 2008[2]) that are necessary for the meaningful construction of composite or synthetic indexes.
Each composite index is based on a theoretical framework representing an agreed concept in the area it covers. The variables comprising the indexes are chosen based on their relevance to the concept. Each index is constructed in close collaboration with the relevant OECD expert groups, which advised on the variables and the weighting schemes to use for the composite.
A number of statistical analyses were also conducted to ensure the validity and reliability of the composite indexes. The survey questions used to create the indexes are the same across countries, to ensure indexes are comparable. In order to eliminate scale effects, all indicators and variables were normalised between “0” and “1” for comparability. To build the composites, all indicators were aggregated using a linear method. The index scores were determined by adding together the weighted scores of each indicator. Statistical tools (i.e. Cronbach’s alpha) were also employed to establish the degree of correlation among a set of variables comprised in each index and to check their internal reliability. This implies that all of the variables comprised in each index have intrinsic value but are also interlinked and capture the same underlying concept. Finally, sensitivity analysis using Monte Carlo simulations was carried out to establish the robustness of the index scores to different weighting options.