Functional ANOVA in Computer Models With Time Series Output.

Computer models are important tools used in scientific investigations of natural phenomena. A computer model may be viewed as an input → output function. In order to assess the influence of each input on the output, formal sensitivity analyses are typically carried out. A specific type of sensitivity analysis addressed in this article is the functional ANOVA, in which an output variance decomposition is used to quantify each input's importance. Here we propose functional ANOVA for computer models with time series output. The argument given in this article originates in the concept of conditional expectation and variance with respect to time. We also establish a relationship between the proposed sensitivity indices and the global regular sensitivity indices existent in the literature. An application from the automotive industry is presented to illustrate the methods. [ABSTRACT FROM AUTHOR]