MPS-RR 2002-2
February 2002

Perfect simulation and inference for spatial point processes

Jesper Møller

Kasper K. Berthelsen

Abstract

The advantages and limitations of using perfect simulation for simulation- based inference for pairwise interaction point processes are discussed. Various aspects of likelihood inference for the Strauss process with unknown range of interaction are studied. A large part of the paper concerns non-parametric Bayesian inference for the interaction function. Markov chain Monte Carlo methods, particularly path sampling, play an important role. Several empirical results and various datasets are considered.

Availability: [ zipped ps-file ] [ pdf-file ]

[ Help on down-loading/viewing/printing ]