Our main aims in this article are: (i) to model the means by which rainfall affects malaria incidence in the state of Pará, one of Brazil's largest states; and (ii) to check for similarities along the counties in the state. We use state of the art spatial–temporal models which can, we believe, anticipate various kinds of interactions and relations that might be present in the data.
We use the traditional Poisson–normal model where, at any given time, the incidences of malaria for any two counties are conditionally independent and Poisson distributed with log-mean explained by rainfall and random effects terms. Our methodological contribution is in allowing some of the random effects variances to evolve with time according to a dynamic model. Additionally, the change of support problem caused by combining malaria counts (per county) with rainfall (per station) is partially solved by interpolating the whole state through a Gaussian process.
Posterior inference and model comparison are computationally assessed via Markov chain Monte Carlo (MCMC) methods and deviance information criteria (DIC), respectively. Copyright © 2005 John Wiley & Sons, Ltd.