Areal prediction of survey data using Bayesian spatial generalised linear models

The conditional autoregressive approach is popular to analyse data with geocoded boundary. However, spatial prediction is often challenging when observed data are sparse. It becomes more challenging in predicting areal units with different areal boundaries. Hence, this paper develops a spatial generalised linear model for spatial predictions using data from spatially misaligned sparse locations. A spatial basis function associated with the conditional autoregressive models and the kriging method is considered. The proposed model demonstrates its better predictive performance through a simulation study and then is applied to understand the spatial pattern of undecided voting preferences in Australia.