Spatial interaction between neighbouring counties: cancer mortality data in Valencia Spain

The statistical analysis of geographical mortality data has usually been approached via regression models that include appropriate covariates. These models assume stochastic independence of mortality counts for neighbouring sites, a questionable assumption that spatial automodels (Besag, 1974, Journal of the Royal Statistical Society, Series B 36, 192-236) make unnecessary. This paper presents the use of the autopoisson distribution in order to detect spatial interaction between neighbouring sites. If this interaction results in being nonsignificant, the auto-Poisson distribution reduces to a usual Poisson regression model, a particular case of generalized linear models (McCullagh and Nelder, 1989, Generalized Linear Models, 2nd edition. London: Chapman and Hall) which can be analyzed with the GLIM package.