Some Fast Methods for Fitting Some One-parameter Spatial Models

Abstract

It is common in geographic modelling to use a one-parameter spatial model to specify the inverse covariance matrix in terms of I-βW, for some known matrix W. Exact Gaussian maximum likelihood estimation of β requires evaluation of the determinant of the covariance matrix. For large data sets, this evaluation of the determinant can be slow and good approximations can be useful. Seventy regional configurations are used to consider some approximations to the determinant of I-βW that are fast to evaluate, and their usefulness is compared.