ABSTRACT
This chapter describes the class of generalised linear models (GLM), with a focus on Gaussian, Binomial and Poisson models, and shows how to carry out exploratory spatial analysis to look for evidence of residual spatial correlation. It introduces the class of generalised linear mixed models to account for over-dispersion of nominally Binomial or Poisson data. The chapter examines how to use diagnostic procedures based on the empirical variogram to test for spatial independence. The class of GLMs extends the linear regression modelling framework to responses that are not normally distributed. A geostatistical model relaxes the assumption of stochastic independence between the observations by allowing spatially correlated residuals. The chapter utilizes the linear regression model for an exploratory analysis of data on malnutrition in Ghana. Analysis of the residual from an ordinary least squares regression can also help to inform the formulation of an appropriate geostatistical model.
