This paper addresses the sensitivity analysis of time-dependent computer models. Often, in practice, we partition the inputs into a subset of inputs relevant to the application studied, and a complement subset of nuisance inputs that are not of interest. We propose sensitivity measures for the relevant inputs of such dynamic computer models. The subset of nuisance inputs is used to create replication-type information to help quantify the uncertainty of sensitivity measures (or indices) for the relevant inputs. The method is first demonstrated on an analytical example. Then we use the proposed method in an application about the safety of restraint systems in light tactical vehicles. The method indicates that chest deflection curves are more sensitive to the addition of pretensioners and load limiters than to the type of seatbelt.
Adaptive importance sampling for extreme quantile estimation with stochastic black box computer models
