【正文】 ob(Fstatistic) 從上圖中可以看出，nR 2=，比較計算的nR 2=異方差。用以上兩種方法，可以檢驗模型是存在異方差的。c) 修正模型1）用加權二乘法修正異方差現象步驟如下：①當權數w1=1/x時，用軟件分析得：Dependent Variable: YMethod: Least SquaresDate: 12/11/14 Time: 13:22Sample: 1 34Included observations: 34Weighting series: W1Variable Coefficient Std. Error tStatisticX C Weighted StatisticsRsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Akaike info criterionSum squared resid Schwarz criterionLog likelihood HannanQuinn criter. Prob. 統(tǒng)計量的臨界值，（2）=，所以拒絕原假設，不拒絕備擇假設，表明模型存在 Prob. Fstatistic DurbinWatson stat Prob(Fstatistic) Unweighted Statistics Rsquared Mean dependent var Adjusted Rsquared . dependent var . of regression Sum squared resid 1489089.DurbinWatson stat 得方程模型為：Y=t=（）（）R 2= F= DW=對此模型進行White 檢驗如下：Heteroskedasticity Test: White Fstatistic Prob. F(2,31) Obs*Rsquared Prob. ChiSquare(2) Scaled explained SS Prob. ChiSquare(2) Test Equation:Dependent Variable: WGT_RESID^2Method: Least SquaresDate: 12/11/14 Time: 11:20Sample: 1 34Included observations: 34Collinear test regressors dropped from specification Variable Coefficient Std. Error tStatistic Prob. C WGT^2 X*WGT^2 Rsquared Mean dependent var Adjusted Rsquared . dependent var . of regression Akaike info criterion Sum squared resid 58053732 Schwarz criterion Log likelihood HannanQuinn criter. Fstatistic DurbinWatson stat Prob(Fstatistic) 2從上圖中可以看出，nR =，比較計算的統(tǒng)計量的臨界值，因為nR 2=響。（2）=，所以接受原假設，即該模型消除了異方差的影②當權數w2=1/x2時，用軟件分析得：Dependent Variable: YMethod: Least SquaresDate: 12/11/14 Time: 13:27Sample: 1 34Included observations: 34Weighting series: W2Variable Coefficient Std. Error tStatisticX C Weighted StatisticsRsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Akaike info criterionSum squared resid Schwarz criterionLog likelihood HannanQuinn criter.Fstatistic DurbinWatson statProb(Fstatistic) Unweighted StatisticsRsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Sum squared residDurbinWatson stat 得方程模型為：Y=+t=（）（）R 2= F= DW=用White 檢驗模型得：Heteroskedasticity Test: WhiteFstatistic Prob. F(3,30)Obs*Rsquared Prob. ChiSquare(3)Scaled explained SS Prob. ChiSquare(3)Prob. 2138654. Test Equation:Dependent Variable: WGT_RESID^2Method: Least SquaresDate: 12/11/14 Time: 11:19Sample: 1 34Included observations: 34Variable Coefficient Std. Error tStatisticC WGT^2 X^2*WGT^2 X*WGT^2 Rsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Akaike info criterionSum squared resid 3583851. Schwarz criterionLog likelihood HannanQuinn criter.Fstatistic DurbinWatson statProb(Fstatistic) 從上圖中可以看出，nR 2=，比較計算的nR 2= Prob. 統(tǒng)計量的臨界值，（2）=，所以拒絕原假設，不拒絕備擇假設，表明模型存在異方差。此模型并未消除異方差。③當權數w3=1/sqr(x)時，用軟件分析得：Dependent Variable: YMethod: Least SquaresDate: 12/11/14 Time: 13:21Sample: 1 34Included observations: 34Weighting series: W3Variable Coefficient Std. Error tStatistic Prob.X C Weighted StatisticsRsquared Mean dependent var Adjusted Rsquared . dependent var. of regression Akaike info criterionSum squared resid Schwarz criterionLog likelihood HannanQuinn criter.Fstatistic DurbinWatson statProb(Fstatistic) Unweighted StatisticsRsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Sum squared residDurbinWatson stat 得方程模型為：Y=+t=（）（）R 2= F= DW=對所得模型進行White 檢驗：Heteroskedasticity Test: WhiteFstatistic Prob. F(2,31)Obs*Rsquared Prob. ChiSquare(2)Scaled explained SS Prob. ChiSquare(2)Test Equation:Dependent Variable: WGT_RESID^2Method: Least SquaresDate: 12/10/14 Time: 13:23Sample: 1 34Included observations: 34Collinear test regressors dropped from specification Variable Coefficient Std. Error tStatisticC WGT^2 X^2*WGT^2 Rsquared Mean dependent varAdjusted Rsquared . dependent var. of regression Akaike info criterionSum squared resid +09 Schwarz criterionLog likelihood HannanQuinn criter. 1045123. Prob. Fstatistic DurbinWatson statProb(Fstatistic) 從上圖中可以看出，nR 2=，比較計算的nR 2=統(tǒng)計量的臨界值，（2）=，所以拒絕原假設，不拒絕備擇假設，表明模型存在異方差。此模型并未消除異方差。綜上所述，用加權二乘法w1的效果最好，所以模型為：得方程模型為：Y=t=（）（）R 2= F= DW=2）用對數模型法用軟件分析得：Dependent Variable: LNY Method: Least SquaresDate: 12/11/14 Time: 09:54Sample: 1 34Included observations: 34Variable Coefficient Std. Error tStatistic Prob.LNX C Rsquared Mean dependent var Adjusted Rsquared . dependent var . of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood HannanQuinn criter. Fstatistic DurbinWatson stat Prob(Fstatistic) 得到模型為：LnY= LNX+對此模型進行White 檢驗得：Heteroskedasticity Test: WhiteFstatistic Prob. F(2,31) Obs*Rsquared Prob. ChiSquare(2) Scaled explained SS Prob. ChiSquare(2) Test Equation:Dependent Variable: RESID^2Method: Least SquaresDate: 12/11/14 Time: 09:55Sample: 1 34Included observations: 34 Variable Coefficient Std. Error tStatistic Prob.C LNX LNX^2 Rsquared Mean dependent var Adjusted Rsquared . dependent var . of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood HannanQuinn criter. Fstatistic DurbinWatson stat Prob(Fstatistic) 從