A sensitivity analysis of the PAWN sensitivity index
The PAWN index is gaining traction among the modelling community as a moment-independent method to conduct global sensitivity analysis. However, it has been used so far without knowing how robust it is to its main design parameters, which need to be defined ab initio by the analyst: the size (N) and sampling (ε) of the unconditional model output, the number of conditioning intervals (n) or the summary statistic (θ). Here we fill this gap by running a sensitivity analysis of a PAWN-based sensitivity analysis. We show that PAWN is highly sensible to the setting of (N,n,ε, θ), and that such uncertainty creates non-negligible chances of PAWN producing non-robust results in a factor prioritization or factor screening contexts. Increasing the precision of PAWN is a complex affair due to the existence of important interactions between (N,n,ε, θ), which we found significant up to the third-order. Even in an ideal setting in which the optimum choice for (N,n,ε, θ) is known in advance, PAWN might not allow to distinguish a truly influential, non-additive model input from a truly non-influential model input.
READ FULL TEXT