Inference for Over- and Under-dispersed Point Processes

Robert Wolpert and Katja Ickstadt

Duke University and Technische Universität Darmstadt

August 2000

Some point process models are only appropriate for over-dispersed point distributions more irregular than Poisson point processes, while many others are appropriate only for under-dispersed distributions. Markov point processes may be appropriate for either over- or under-dispersed point distributions, but the extraordinary difficulty of calculating their partition functions presents a formidable computational obstacle to making statistical inference. We present a model, Matérn-thinned Cox processes, that offers computationally tractable inference for both over- and under-dispersed patterns.

Keywords: Cox process, inference, Markov point process, Poisson point process, thinned processes.


The manuscript will be available very soon ... in postscript and pdf formats.