Abstract
This paper presents an optimization procedure that helps composite indicator developers achieve the most plausible choice of weights without being restricted as the complexity of synthetic models escalates. Given a predefined aggregation function, variance-based sensitivity analysis and Monte Carlo simulations are employed to establish non-parametric methods for measuring the importance of each input to the output uncertainty. Utilizing the computational power of these methods, the weights are calibrated by an optimization procedure to attain the best fit with the estimated measures of importance. The procedure has been tested in two artificially created examples and in one practical case of well-being measurement to confirm its accuracy and efficiency in building composite indicators.