Decomposition of high-order statistics

Abstract : ANOVA analysis is a very common numerical technique for computing a hierarchy of most important input parameters for a given output when variations are computed in terms of variance. This second central moment can not be retained as an universal criterion for ranking some variables, since a non-gaussian output could require higher order (more than second) statistics for a complete description and analysis. In this work, we illustrate how third and fourth-order statistic moments, \textit{i.e.} skewness and kurtosis, respectively, can be decomposed. It is shown that this decomposition is correlated to a polynomial chaos expansion, permitting to easily compute each term. Then, new sensitivity indices are proposed basing on the computation of the kurtosis. An analytical example is provided with the explicit computation of the variance and the skewness. Some test-cases are introduced showing the importance of ranking the kurtosis too.
Type de document :
[Research Report] RR-8193, INRIA. 2012
Liste complète des métadonnées
Contributeur : Pietro Marco Congedo <>
Soumis le : mercredi 19 décembre 2012 - 10:41:08
Dernière modification le : jeudi 11 janvier 2018 - 06:22:35
Document(s) archivé(s) le : mercredi 20 mars 2013 - 11:32:01


Fichiers produits par l'(les) auteur(s)


  • HAL Id : hal-00766853, version 1



Remi Abgrall, Pietro Marco Congedo, Gianluca Geraci, Gianluca Iaccarino. Decomposition of high-order statistics. [Research Report] RR-8193, INRIA. 2012. 〈hal-00766853〉



Consultations de la notice


Téléchargements de fichiers