W.J. Padgett, Distinguished Professor Emeritus

Department of Statistics

University of South Carolina

and

Visiting Professor, Mathematical Sciences

Clemson University

Bayes Bandwidth Selection in Kernel Density Estimation with Censored Data

Problems with right-censored data arise frequently in survival analysis and reliability applications, and estimation of the lifetime density function is often of interest. Two inherent problems in kernel density estimation for lifetime data are the "spillover" at the origin and the selection of the smoothing parameter (or bandwidth) values. To address these issues, we propose the use of asymmetric kernels with a Bayesian approach to bandwidth selection. In particular, the inverse Gaussian density function is used as the kernel. The (local) Bayes bandwidth is exact for any sample size, only depends on the prior parameters, and can be easily calculated from the censored data. Strong pointwise consistency of the density estimator is proven, and it is also shown that meaningful bandwidths with the same rates of convergence as for the classical asymptotically optimal bandwidths can be obtained for suitable choices of the prior parameters.
** (Joint work with K.B. Kulasekera, Clemson University)

Back to Colloquium Series