空间回归模型选择的反思

空间计量经济学存在两种最基本的模型:空间滞后模型和空间误差模型,这里旨在重新思考和探讨这两种空间回归模型的选择,结论为:Moran’s I指数可以用来判断回归模型后的残差是否存在空间依赖性;在实证分析中,采用拉格朗日乘子检验判断两种模型优劣是最常见的做法。然而,该检验仅仅是基于统计推断而忽略了理论基础,因此,可能导致选择错误的模型;在实证分析中,空间误差模型经常被选择性遗忘,而该模型的适用性较空间滞后模型更为广泛;实证分析大多缺乏空间回归模型设定的探讨,Anselin提出三个统计量,并且,如果模型设定正确,应该遵从Wald统计量>Log likelihood统计量>LM统计量的排列顺序。

The Choice of Spatial Econometric Models Reconsidered in Empirical Studies

There are two kinds of basic spatial econometric models:spatial lag model and spatial error model.The aim of this paper is to reconsider the choice of spatial regression models.Four conclusions are drawn:firstly,Moran's I is used to test spatial dependence among residuals in the OLS model.Secondly,in empirical studies Lagrange Multiplier (LM)test is performed to point out which model is in favor of. However,LM test is designed to test for spatial dependence for residuals of regression models,but ignored the vary basic economic theory or theoretical grounds,which may lead to a wrong specification. Hence,the choice of spatial econometric models should be based first on the basic theoretical grounds and then on LM test results.Thirdly,spatial error model ignored in empirical studies is better than spatial lag model to some extent.Finally,empirical studies usually ignore to test model specifications.Anselin proposed three statistics.If specification is appropriate,it follows that Wald >Log likelihood>LM.