Correlated inputs

So far, all described measures assumed uncorrelated input parameters. Typically, this assumption is not met by practical applications as joint estimations of parameters tend to produce correlated estimates. This gives reason to expand sensitivity measures to a more general setting.

Today, several recent contributions deal with the extension of the Sobol’ sensitivity measures to the setup with correlated inputs. For instance, estimators for two complementary sets of measures are developed by [Kucherenko.2012] and [Mara.2015]. On the other hand, the only contribution to the computation of EE-based screening measures for correlated parameters is made by [Ge.2017]. Some authors, e.g. [Saltelli.2004] even negate the necessity for expanded Elementary Effects due to overcomplexity. Obviously, this can lead to false results.