ABSTRACT
The significant advance in the statistical analysis of spatial data is to acknowledge the fact that the configuration of observations carries important information about the relationship of data points. A case in point are nonlinear mixed model applications for clustered data. Two important sources of variation there are the changes in response as a function of covariates for each cluster and cluster-to-cluster heterogeneity. Non-separable, valid covariance functions for spatio-temporal data are typically more complicated than separable models but incorporate space-time interactions. Instead of integration in the frequency domain, nonseparable covariance functions can also be constructed by summation or integration in the spatio-temporal domain.
