Recent innovations in spatial statistics have focused on the problem of non-normal data: specifically spatial counts and spatial proportions. A number of approaches have been proposed to model spatial proportions, including spatial GLMMs and Bayesian hierarchical models. While these approaches have their advantages, there is a serious jump in complexity over traditional spatial models which limits their use outside of statistics.
This paper will present an alternative approach called beta-binomial kriging. Inspired by Poisson kriging (Monestiez, 2006), beta-binomial kriging models latent beta spatial random fields as a function of observed sample proportions. Beta-binomial kriging is relatively simple compared to other models, and provides accurate predictions under a wide variety of circumstances. The approach is ideally suited for problems in epidemiology and spatial ecology where underlying probabilities of a phenomenon are of primary interest.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.