第五讲空间计量经济学基本模型的Geoda估计

《第五讲空间计量经济学基本模型的Geoda估计》由会员分享，可在线阅读，更多相关《第五讲空间计量经济学基本模型的Geoda估计（25页珍藏版）》请在装配图网上搜索。

1、第五讲空间针量從济曇模型的Geodai针一、模型数据的准备dbf数据库操作工具：WPS软件可以修改.dbf数据库方法：用WPS打开china.dbf，删掉无关的列，保存为china.dbfo注 意，不要更改文件名。用geoda打开china.dbf,根据需求增加变量，之后，在file菜单选择Save,保存china.dbf。再用WPS打开china.dbf,将数据粘贴导入china.dbf文件，点击保存，完成数据增加。开展后续的空间计量分析几点说明 .dbf文件在WPS里无法直接增加新列, 也就是无法直接增加新的变量。在wps里修改.dbf数据时，需要同时修 改数据类型，选择带小数位数的实数类

2、 型。db啲文件名，前后保持一致，不然, 无法与.shp地图文件匹配。解释变量的空间滞后处理工具：Geoda软件方法：生成新变量首先生成空间权重矩阵W 在d bf文件中增加新变量WX运用菜单工具table-variable calculation-spacial lag工具计算。二、基本模型的GeoDa估计A权重矩阵对GeoDa能力的约束GeoDa只能给出基于邻接关系的W/只能估计基于邻接关系的空间计量模型A可以估计的模型类型/OLS/SLM “SEMA软件操作步骤/I、打开.shp数据文件/2、创建W （若已有W,则省略该步骤）SLM只能使用对称的W, K最近距离W不能用/3、在菜单选Met

3、hods-Regression/4、选择变量，以及W“5、选择Models类型（OLS、SLM、SEM）/6、运行nmRegression ReportS3SUMMARY OF OUTPUT:ORDINARY LEAST SQUARES ESTIMATIONDat-a Eetchin-aDepen且Ent Variable:B9 1995NuiTLberofObservat ions:30Meam dependentva.r-1916.07Numbei?ofVa.ria.bles:S.D. He口endentvar1454 D3Deq-reesofF r cedom:28Rsqu-ared.0

4、 751526Fstmt iwt iu:84.68SAdj usted Rsquared0.742G52Prob(Fstat ist ic):5.S3133e-010Sum squa.i?ed r-esidua.1 : 157598e+007Logr 1 ikel ihood-240.14ES i ginasquare562850Akaike info uxiterion:484.29SE. of i?egr=c?ession750.233Schwa. i?z cr-i t er-ion:487.092Sigm-asqu-aare ML525327S . E of r-ecrr-ession

5、ML :724.794Va.i?ia.bleCoef ficientStd . Errol?tS ta. tt zlcPr-obaibi 丄 i t yCONSTANT-103.845225S.7258-0.40137170.69119B24_199512 . 530991.3616789 . 2026080.00000REGRESSION DIAGNOSTICSMULTICOLLINEARITV CONDITIOH NUMBER 3. 491332 TEST ON NORMALITY OF ERRORSTESTDFVALUEPEOEJ a:rquEEEta.25. EiECIZ0. D528

6、6DIAGNOSTICS FOR HETEEOEKEDASTICITY RANDOM COEFFICIENTSTESTDFVALUEPEOEEreuschPag-an test7.71890.00546KoenkerBa.sset t test4.13430.04202DIAGNOSTICS FOR SPATIAL DEPENDENCEFOR WEIGHT MATRIX :china gr-al(rowst andard i zedweights)TESTMI/DFVALUEPKOBMorans I (error)0 21112 16840 03013Lagr-amge Muitlplien?

7、(lag)15.04030.02476Robust LM ( la.qr)12 4 0 9 4匸I 12061Lagrange Mu 11 ipl ier-(error)12.S210.09304Kobust LM (error)10.1902LI . 6 6 2 7 7SUMHARY OF OUTPUT: SPATIAL LAG MODEL - MAXIMUM LIKELIHOODESTIMATIONData setchinaSpa tie. J. Weigh tchina .ga.1Dependent VariableB9_1995 Nuinber of Observatiotis30Me

8、an dependent var1916.07 Number of Variables3S D dependent var1454.03 Degrees of Freedom27Lag coef f.(Rhci)0.270554E-squared0.787439 log likelihood-238.062Sq Correiaticin一Akaike info criterion482.124Signa-square449401 Schwarz criterion486.3285.E of regression670.374VariableCoef f icientStd.Errorz-val

9、ueProbabilityU B9 19950.27055390.13950911.9393280.05246CONSTANT-493.214312.7952-1.5767950.11484B24 199511.643591.2622979.2241310.00000REGRESSION DIAGNOSTICSDIAGNOSTICS FOE HETEROSKEDASTICITYEANDOM COEFFICIENTSTESTDFVALUEPROBBreusch-Pagsm test14.66500.03077DIAGNOSTICS FOE SPATIAL DEPENDENCESPATIAL LA

10、G DEPENDENCE FOR WEIGHT MATRIX : china.galTESTDFVALUEPROBlikelihood Ratio Test14.16520.04126Regression ReportS34Number of Observations Nuinber of Variables Degrees of FreedomSUMMARY OF OUTPUT: SPATIAL ERROR MODEL - MAXIMUM LIKELIHOOD Data :setSpatiul WeightDependent VariableMean dependent varS D dep

11、etiden t varESTIMATION30228china china galB9_1995 1916.066667 1454.034219 0.366046R-squaredSq Correlation Sigma-squareS.E regression0.778967467312683.602R-squared (BUSE)Log likelihoodAkaike info criterionSchwarz uriterion-238.881459481.763484.565VariableCoef f icientStd.Errorz-valueProbabilityCONSTA

12、NT2.278373273.81120.0083209630.99336B24 199511.60751.2541459.2553110.00000LAMBDA0.36604630.21650481.6907080.09089La oneff (lajTLbdB)REGRESSION DIAGNOSTICSDIAGNOSTICS FOR HETEROSKEDASTICITY RANDOM COEFFICIENTSTESTEreusch-Pagan testDF1VALUE4.9624PROB0.02590DIAGNOSTICS FOR SPATIAI DEPENDENCESPATIAL ERR

13、OE DEPENDENCE FOE IJEIGHTMATRIX :china galTESTDFVALUEPROBLikelihood Ratio Test12.52670.11194A结果说明/模块一：模型的基本统计信息/模块二：回归结果的统计信息/模块三：回归系数及其显著性/模块四：模型结果的诊断（SLM、SEM）蓝色线条以上，并方差诊断，原假设为无异方差蓝色线条以下，空间相关性诊断，原假设为不存在空间相关性三、最优模型的确定Run SpatialError ModelRun SpatialLag Model选择标准及步骤/I、做一次OLS估计/ 2、对 比LM统计量，LM-Lag和LM-

14、Error/3、若均不显著，则无需进行空间计量分析/4、若只有一个显著，则设定为与显著统计量对应的 空间计量模型5、若均 显著，再对 比Robust LM-Lag和Robust LM- Error/6、选择显著（相对显著）的统计量对应的空间计量 模型Dimsncs fop sram depeudeuceFOE WIGHT iTRII: china.ga (roFstandardized weights)TESTKI/DFMEPKOBMoran1s I (error)Ulll2.1M0.03013Lagrange Kultiplier (lag)15.0403Eobust LM (lag)11

15、12.4094o.imiLagrange Kultiplier (error)1 12.8210o,恤Robust LB (error)11 10.19020.07Regression ReportS34SUMHARY OF OUTPUT: SPATIAL LAG MODEL - MAXIMUM LIKELIHOODESTIMATIONData setchinaSpa tie. J. Weigh tchina .ga.1Dependent VariableB9_1995 Nuinber of Observatiotis30Mean dependent var1916.07 Number of

16、Variables3S D dependent var1454.03 Degrees of Freedom27Lag coef f.(Rhci)0.270554E-squared0.787439 log likelihood-238.062Sq Correiaticin一Akaike info criterion482.124Signa-square449401 Schwarz criterion486.3285.E of regression670.374VariableCoef f icientStd.Errorz-valueProbabilityU B9 19950.27055390.1

17、3950911.9393280.05246CONSTANT-493.214312.7952-1.5767950.11484B24 199511.643591.2622979.2241310.00000REGRESSION DIAGNOSTICSDIAGNOSTICS FOE HETEROSKEDASTICITYEANDOM COEFFICIENTSTESTDFVALUEPROBBreusch-Pagsm test14.66500.03077DIAGNOSTICS FOE SPATIAL DEPENDENCESPATIAL LAG DEPENDENCE FOR WEIGHT MATRIX : c

18、hina.galTESTDFVALUEPROBlikelihood Ratio Test14.16520.04126Number of Observations Nuinber of Variables Degrees of FreedomSUMMARY OF OUTPUT: SPATIAL ERROR MODEL - MAXIMUM LIKELIHOOD Data :setSpatiul WeightDependent VariableMean dependent varS D depetiden t varESTIMATION30228china china galB9_1995 1916

19、.066667 1454.034219 0.366046R-squaredSq Correlation Sigma-squareS.E regression0.778967467312683.602R-squared (BUSE)Log likelihoodAkaike info criterionSchwarz uriterion-238.881459481.763484.565VariableCoef f icientStd.Errorz-valueProbabilityCONSTANT2.278373273.81120.0083209630.99336B24 199511.60751.2

20、541459.2553110.00000LAMBDA0.36604630.21650481.6907080.09089La oneff (lajTLbdB)REGRESSION DIAGNOSTICSDIAGNOSTICS FOR HETEROSKEDASTICITY RANDOM COEFFICIENTSTESTEreusch-Pagan testDF1VALUE4.9624PROB0.02590DIAGNOSTICS FOR SPATIAI DEPENDENCESPATIAL ERROE DEPENDENCE FOE IJEIGHTMATRIX :china galTESTDFVALUEPROBLikelihood Ratio Test12.52670.11194确立最优模型（难点）/I、确定OLS、SLM、SEM模型/2、对确定后的模型，展开诊断检验“3、如果各项诊断均通过检验，则确定该模型 为最优模型“4、如果有诊断未通过，一般通过调整W、调 整解释变量重新回归。重复步骤3、步骤4,直至确定合适的模型。练习问题:。考虑空间溢出效应的地区人均GDP影响因素分析数据文件: china.shp论文提纲。全局Moranl检验。局部Moran I检验回归分析。运用三类不同的w分别做出结果，选最好的