Nonparametric Maximum Penalized Likelihood Estimation of a Density from Arbitrarily Right-Censored Observations.

摘要

Based on arbitrarily right-censored observations from a probability density function f deg, the existence and uniqueness of the maximum penalized likelihood estimator (MPLE) of f deg is proven. In particular, the first MPLE of Good and Gaskins of a density defined on (0, infinity) is shown to exist and to be unique under arbitrary right-censorship. Furthermore, the MPLE is in the form of an exponential spline which knots at the observed censored and uncensored data points. (Author)