Usefulness of the Information Contained in the Prediction Sample for the Spatial Error Model

A recent study proposed an estimation approach that uses data on the independent variables and location for the prediction sample, and suggested that it may improve estimation and prediction. This is an incomplete data approach following an iterative process along the lines of the EM algorithm. The present study compares this approach with a partial data approach that uses only data on the dependent and independent variables and location for the estimation sample. Our Monte Carlo experiments show that unless the estimation and prediction samples constitute the whole population and the data generating model is used as the data fitting model, the incomplete data approach is not guaranteed to be superior to the partial data approach