NONPARAMETRIC VARIANCE-BASED METHODS OF ASSESSING UNCERTAINTY IMPORTANCE

This paper examines the feasibility and value of using nonparametric v ariance-based methods to supplement parametric regression methods for uncertainty analysis of computer models. It shows from theoretical con siderations how the usual linear regression methods are a particular c ase within the general framework of variance-based methods. Examples o f strengths and weaknesses of the methods are demonstrated analyticall y and numerically in an example. The paper shows that relaxation of li nearity assumptions in nonparametric variance-based methods comes at t he cost of additional computer runs. (C) 1997 Elsevier Science Limited .