A hybrid univariate dimension reduction method for statistical moments assessment

Abstract

This paper presents an index α to evaluate the type of response function, which can help judging the application of two existing form of univariate dimensional reduction (UDRM), say the additive and multiplicative UDRMs for statistical moments assessment. It can determine which one is more effective to decompose a multivariate response function. Then, a new hybrid univariate dimensional reduction (HUDRM) is proposed to compute the raw moments of the response function. The results show that the proposed HUDRM can significantly decrease the relative errors of statistical moments in cases where both the additive and multiplicative UDRMs are not able to provide satisfactory results.