Traitement des données physiques : remarques sur la corrélation

Abstract

The interpretation of some latest very accurate physical data (like spectroscopical data) needs sophisticated models that often lead to correlation problems. The aim of this paper is to show that, when necessary, it is possible to work beyond correlation. The problems of numerical correlation with linear or nonlinear model or the problem of exact correlation with nonlinear models can be solved (applications are made in the real case of the red system of NbN). This latter situation leads to a new and interesting statistical point of view of the data treatment; however, computations are generally far longer than in the usual situation of weak correlation. The proposed point of view is compared with other recent works about the problems encountered in data reduction.