Research Outlier Diagnostics Spatial Error Model And Generalized Spatial Model

Outliers exist frequently in spatial data. However, the existing of outliers may causedeviation during modeling the data set, and the explanation of the economic implication mayalso be affected. Worse still, we may draw a completely adverse conclusion. Therefore the issueabout detecting outliers in spatial econometric model is very significant. However, the issuesabout detecting outliers in spatial econometric model receives almost no attention and is still anunexplored topic in current literature. Thus this paper studies outlier detection in two spatialeconometric models—the spatial auto-regressive error model (SEM) and the general spatialmodel (SAC). Specific formulas and approximate distributions for detecting outliers are derivedunder mean-shift outlier model and variance-weighted model, respectively, for the two spatialeconometric models. Finally, analysis of stochastic simulation and a real example shows thatthe proposed test statistics are effective in identifying outliers in the spatial auto-regressive errormodel (SEM) and the general spatial model (SAC). At the mean time a method to modify thetwo models is also proposed. By comparison of the postulated models and the correctionmodels, the proposed test statistics are confirmed effective for detecting outliers in the twospatial econometric models. And the correction models have an improvement on the postulatedmodels and can incorporate the outliers.The main results we obtained are listing as follows:(1) Outliers detection is studied in the spatial auto-regressive error model (SEM) based onscore test statistics. The formula and corresponding approximate distribution for single outlierand multiple outliers are derived under mean-shift outlier model and variance-weighted modelrespectively.(2) Outliers detection is studied in the general spatial model (SAC) based on score teststatistics. The formula and corresponding approximate distribution for single outlier andmultiple outliers are derived under mean-shift outlier model and variance-weighted modelrespectively．(3) The proposed method is confirmed effective in identifying outliers in the spatial auto-regressive error model (SEM) and the general spatial model (SAC) through the analysis of stochastic simulation and a real example．And modified models to adjust the detected outliersare also proposedHowever, it is crucial to establish a proper model for the given dataset．The use ofinappropriate model may cause a misleading for drawing the conclusion．The issue aboutdetecting outliers is very important in the field of statistical diagnostics Thus the resultsobtained in this paper fill the blank of statistic diagnostics in the research of spatial econometricmodels, and have both theoretical and applied values．...