Nonparametric Mean Preservation in Censored Regression

The aim of this book is to estimate the conditional
mean of some functions depending on the response
variable Y (moments, distributions...) in regression
models where this response is possibly censored. In
parametric regression, polynomial and nonlinear
conditional means are estimated in a new way while,
in nonparametric regression, some new estimators are
provided to approximate general L-functionals
(conditional mean, trimmed mean, quantiles...). The
ideas developed in those methods lead to establish
more general results in nonparametric estimation of
the conditional mean of functions depending on Y and
other variables and where the response can follow
other schemes of incomplete data (not only censored
but also missing or length-biased data). For each
procedure, asymptotic properties are established
while finite sample behavior is studied via
simulations. Examples from a variety of areas
highlight the interest of using the proposed
methodologies in practice.

Autorentext

Cdric Heuchenne is Professor of Statistics at HEC Management School-University of Lige. He holds a Master''s degree in Applied Sciences and a Ph.D. in Statistics from the Catholic University of Louvain. His research interests focus on nonparametric statistical inference for complex data structures.

Klappentext

The aim of this book is to estimate the conditional mean of some functions depending on the response variable Y (moments, distributions...) in regression models where this response is possibly censored. In parametric regression, polynomial and nonlinear conditional means are estimated in a new way while, in nonparametric regression, some new estimators are provided to approximate general L-functionals (conditional mean, trimmed mean, quantiles...). The ideas developed in those methods lead to establish more general results in nonparametric estimation of the conditional mean of functions depending on Y and other variables and where the response can follow other schemes of incomplete data (not only censored but also missing or length-biased data). For each procedure, asymptotic properties are established while finite sample behavior is studied via simulations. Examples from a variety of areas highlight the interest of using the proposed methodologies in practice.