Analysis of variance for functional data.

Abstract

In this dissertation we present an extension to the well known theory of multivariate analysis of variance. In various situations data are continuous stochastic functions of time or space. The speed of pollutants diffusing through a river, the real amplitude of a signal received from a broadcasting satellite, or the hydraulic conductivity rates at a given region are examples of such processes. After the mathematical background we develop tools for analyzing such data. Namely, we develop estimators, tests, and confidence sets for the parameters of interest. We extend these results, obtained under the normality assumption, and show that they are still valid if this assumption is relaxed. Some examples of applications of our techniques are given. We also outline how the latter can apply to random and mixed models for continuous data. In the appendix, we give some programs which we use to compute the distributions of some of our tests statistics.