Measuring the importance of variables in composite indicators

Abstract

The importance of input variables to composite indicators is measured using the nonlinear Pearson correlation ratio, as a guide how the weights assigned actually affect the output. The correlation ratio is estimated using local-linear regression and penalised splines. The approach is demonstrated on a case study, the Resource Governance Index, in which it is concluded that the importance of input variables does not appear to reflect the intentions of the developers. Overall, the approach can be used to adjust the weights of composite indicators to more closely reflect the intentions of developers.