PERSPECTIVES ON SPATIAL ECONOMETRICS: LINEAR SMOOTHING WITH STRUCTURED MODELS
指出标准空间计量模型依赖参数结构，易受模型误设影响，而空间AR模型类似半参数平滑。通过蒙特卡洛实验证明，当邻接矩阵误设时，非参数预测值和边际效应估计比空间AR模型更准确。
ABSTRACT Though standard spatial econometric models may be useful for specification testing, they rely heavily on a parametric structure that is highly sensitive to model misspecification. The commonly used spatial AR model is a form of spatial smoothing with a structure that closely resembles a semiparametric model. Nonparametric and semiparametric models are generally a preferable approach for more descriptive spatial analysis. Estimated population density functions illustrate the differences between the spatial AR model and nonparametric approaches to data smoothing. A series of Monte Carlo experiments demonstrates that nonparametric predicted values and marginal effect estimates are much more accurate then spatial AR models when the contiguity matrix is misspecified.