Estimation non paramétrique du mode conditionnel dans un modèle de censure.

Abstract

In this thesis, we are interested in the nonparametric estimation of the conditional density and the conditional mode, in a twice censorship model. This means that the response variable Y is right censored by a variable R and that min(Y, R) is left censored. On the one hand, we build estimators, by the kernel method, for the conditional density and the conditional mode and establish a rate of the almost complete convergence for them. On the other hand, we give the rate of the mean square convergence of the conditional density estimator. Finally, a simulation study is conducted in order to illustrate, for a nite size, the performance of the proposed estimators.