Local Influence Analysis Of Spatial Regression Model Mix Since

With the development of GIS,GPS and RS, more and more people concernedabout the data geospatial information. Statistical analysis based on the spatial data iscalled spatial statistics. Spatial data has spatial autoregressive which is called thedependence among the data of different area. The spatial autoregressive(SAR) modelis a method of processing spatial correlation, which not only included the correlationbetween the independence and the variable of the own area, but also portrayed thecorrelation between the independence and the independence of the neighbor area. Thepresence of impact points or outliers has great influence on the estimation ofparameters and the inference of statistical, which is also an important problem in theSAR model. The literature of statistical diagnostic about the spatial model is rare.The local influence analysis proposed by Cook(1986) put forward the normalcurvature of the influence diagram to measure the impact of the data, and research thecombined effect through perturbing some parts of the model instead of deletingindividual data points. Because of the diversity of the disturbance scheme, localinfluence analysis has been widely used and developed. Lawrance(1988) pointed outthat the local influence analysis can identify the masking effect, but when there is astrong correlation among the data, local influence analysis can not identify all theinfluential points effectively. Through improving the local influence analysis, Shi andHuang(2011) proposed stepwise local influence analysis, which has been identifiedthat can identify masking effect among the data. In this paper, we apply the twomethods mentioned above to the SAR model to judge the influential observations andoutliers of the SAR model. We use two sets of numerical examples to test and verifythe effectiveness of the local influence analysis and the stepwise local influence analysis.In this paper we use three modes which are variance disturbance, meandisturbance and independent variables disturbance to diagnose the models. From themethod which mentioned before, we get the diagnostic statistics of the SAR modeland the first-order spatial AR (FAR)model that in the disturbance of variancedisturbance, mean disturbance and independent variables disturbance. In the localinfluence analysis, we use the largest eigenvector method to judge the influencepoints or outliers. In the stepwise local influence analysis, we use single pointremoved iterative method, and use average basis to determine the influential pointsand outliers. For the reason that the FAR model is a special form of the SAR model,in this paper, we just use two sets of numerical examples to test and verify theeffectiveness of the local influence analysis and the stepwise local influence analysisused in the SAR model. From the result, we know the diagnosis is effective. Fromthe effect of diagnosis, the stepwise local influence analysis can not only detect theinfluential points and outliers, but also can detect whether the point is stronginfluential point on the basis of the q-value.