Abstract This paper examines the robustness of various models of spatial autocorrelation through a series of Monte Carlo experiments in which each model takes a turn at the data generator. The generated data are then used to estimate all of the models. The estimated models are evaluated primarily on their predictive power.
