Maximum likelihood estimation of a mixing distribution

Abstract

Given a random sample from a mixed density where the form of the component densities is know, subroutine NPMLG computes the nonparametric maximum likelihood estimate of the mixing distribution. Because of the nature of the mixture density, the estimation procedure is an iterative one. the algorithm used here is a syntheses of the EM approach (Dempster, Laird, and Rubin, 1977), and a stepwise inclusion procedure based on the characterization theorem (Simar, 1976). This synthesis combines the computational simplicity of the EM procedure with the certainty that the solution is the global maximum.