|
Abstract:
|
In Public Health we strive to predict disease incidence or mortality for a particular area, such as state or county. At an aggregate level, we can predict spatially correlated count data using a Generalized Linear Mixed Models (GLMM) framework with a Poisson outcome variable with an auxiliary variable in a bivariate relationship. A cokriging structure is applied in a GLMM setting which includes a Poisson outcome variable and an auxiliary variable with a named distribution, where the outcome variable and auxiliary variable are spatially correlated and correlated with each other. This methodology is examined in a real data setting with applications in public health, predicting West Nile virus incidence, and cancer incidence and mortality at the county level. Environmental auxiliary variables considered in the West Nile virus prediction are counts of infected mosquitos, birds, and percent irrigated farmland. In cancer incidence and mortality prediction we consider environmental variables, and socio-demographic variables such as racial distribution, education, and income as auxiliary variables. To evaluate the prediction performance, cross-validation is employed.
|
