As a programmer I have used the spdep package successfully for spatial filtering. But would appreciate it if someone could offer a description (preferably with supporting references) of how this concept works. Let's suppose I have a standard linear regression model with an integer response variable and 3 or more predictors comprising real values:
According to Chun and Griffith, a spatial filter would be constructed from missing predictors which are spatially correlated, and which will help to model the autocorrelation of the all observations. How is the autocorrelation among multivariate predictors taken into account?
How does this approach compare to the spatial Durbin model?
Does this concept have any relationship with the Perron–Frobenius theorem from linear algebra?