Estimation of the Parameters in a Spatial Regressive-~Autoregressive Model Using Ord's Eigenvalue Method
In this thesis, we study one of Ord's (1975) global spatial regression models.
Ord considered spatial regressive-autoregressive models to describe the interaction
between location and a response variable in the presence of several covariates. He also
developed a practical estimation method for the parameters of this regression model
using the eigenvalues of a weight matrix that captures the contiguity of locations.
We review the theoretical aspects of his estimation method and implement it in the
statistical package R.
We also implement Ord's methods on the Columbus, Ohio, crime data set from the
year 1980, which involves the crime rate of each neighborhood of the city as a response
variable and the average income and average house value of each neighborhood as
covariates. We use different weight matrices that capture different "nearest neighbor"
notions and compare the results.