A0326
Title: Smoothing spline density estimation from doubly truncated data
Authors: Jacobo de Una-Alvarez - University of Vigo (Spain) [presenting]
Abstract: Smoothing splines can be introduced as the solution to a penalized maximum likelihood problem in a reproducing kernel Hilbert space. In the setting of density estimation, they provide a flexible, nonparametric approach to approximate the target, achieving a compromise between bias and variance in an automatic way. Doubly truncated data are often encountered in survival analysis, epidemiology, reliability engineering, economics or astronomy, among other fields. They appear in particular with interval sampling, when the sampled units are those for which the event of interest occurs between two particular dates, a sampling mechanism that is ubiquitous, for instance, in clinical research. Estimation from doubly truncated data requires proper corrections for the sampling bias. Smoothing splines are introduced and investigated for density estimation in the presence of double truncation. Through Monte Carlo experiments, the relative benefits of smoothing splines compared to other popular nonparametric approaches, such as kernel density estimation, are illustrated. Real data illustrations are provided.
