Andrew Lawson

Latent structure models can be developed for the mean level of a space-time count data observation process. The focus is on small area health outcomes observed in fixed spatial units and fixed time periods. We assume a Poison data level model with mean parameterized as a weighted mixture of temporal components. Each area has a distribution of weights assigning the area to a component. The model development is within the Bayesian paradigm, and we make a set of different choices for prior distributions for the weights and temporal components. Posterior sampling can be used to estimate both weights and components. Identification and label switching are considered in relation to single/double chain dynamics. Allocation paradigms are considered as we focus on spatial ‘clustering’ of temporal profiles. Goodness of fit measures suggest that from a relative risk viewpoint these mixtures can outperform conventional ST random effect models (such as proposed by Knorr-Held, 2000), while also providing latent component information. In this talk we focus on the incorporation of covariate predictor information in the model and how they can modulate the risk in different ways. An example of asthma ED visits within counties of Georgia with a PM 2.5 predictor is presented. Joint seminar with the University of Georgia Department of Epidemiology and Biostatistics