一:各地区农村居民家庭人均纯收入与家庭人均消费支出的数
据(单位:元)地区 北京
天津 河北 山西 内蒙古 辽宁 吉林 黑龙江 上海 江苏 浙江 安徽 福建 江西 山东 河南 湖北 湖南 广东
广西
海南 重庆 四川 贵州 云南 西藏 陕西 甘肃 青海 宁夏
Y
9439.63 7010.06
4293.43 3665.66 3953.1 4773.43 4191.34 4132.29 10144.62 6561.01 8265.15 3556.27 5467.08 4044.7 4985.34 3851.6 3997.48 3904.2 5624.04 3224.05 3791.37 3509.29 3546.69 2373.99 2634.09 2788.2 2644.69 2328.92 2683.78 3180.84
X 6399.27 3538.31 2786.77 2682.57 3256.15 3368.16 3065.44 3117.44 8844.88 4786.15 6801.6 2754.04 4053.47 2994.49 3621.57 2676.41 3090 3377.38 4202.32 2747.47 2556.56 2526.7 2747.27 1913.71 2637.18 2217.62 2559.59 2017.21 2446.5 2528.76
新疆
3182.97 2350.58
二.参数估计:
Dependent Variable: X Method: Least Squares Date: 11/11/11 Time: 08:22 Sample: 1 31 Included observations: 31
CoefficienVariable t Std. Error t-Statistic
C 179.1916 221.5775 0.808709 Y 0.719500 0.045700 15.74411 R-squared 0.895260 Mean dependent var
Adjusted R-squared 0.891649 S.D. dependent var S.E. of regression 493.6240 Akaike info criterion Sum squared resid 7066274. Schwarz criterion Log likelihood -235.2084 F-statistic Durbin-Watson stat 1.461684 Prob(F-statistic)
Prob. 0.4253 0.0000 3376.309 1499.612 15.30377 15.39628 247.8769 0.000000
根据回归结果,则模型估计的结果为:
ˆi=179.1916 + 0.719500 Yi X(0.808709 ) (15.74411) R2=0.895260 F= 247.8769
三.检验模型的异方差: (一)图形法
1)绘制et2对Yt的散点图即E2对Yt的散点图:
300000025000002000000E2150000010000005000000200040006000Y80001000012000 2)判断:
由此散点图可知残差平方ei2对解释变量Y 的散点图主要分布在图像中的下三角部分,大致可以看出残差平方ei2随着Yi的变动成增大的趋势,因此,模型很可能存在异方差,但是否确实存在异方差还寻妖进一步的检验。
(二)Goldfeld-Quanadt检验
删除中间5个观测值后,余下的部分构成了两个样本区间:1—12和19—31,它们的样本个数均是13个。即n1=n2=13
1)将定义区间为1—12,则回归结果为:
Dependent Variable: Y Method: Least Squares Date: 11/26/11 Time: 09:36
Sample: 1 12
Included observations: 12
Variable
C
Coefficient
4618055
Std. Error 37464
t-Statistic 89195
Prob. 0.0218167
7
0.0078477
1111.2229409.57498342.71311235043505439
0.45190030.1364277803.31237756963690404
X
R-squared
Adjusted R-squared S.E. of regression Sum squared resid Log likelihood
21532061 0.52317020.4754872
58538097 S.D. dependent var 203.33230
6057605 Akaike info criterion 413440.26
6867037 Schwarz criterion -79.71143
24534216 F-statistic 2.0303963
Durbin-Watson stat
2223662 Prob(F-statistic) 966162
52263
83
2453.88583333333 280.755559499988 13.6185720755703 13.6993898505349 10.9718451630769 0.007847763690404
46
35034633 Mean dependent var
2)将定义区间为19-31则回归结果为:
Dependent Variable: Y Method: Least Squares Date: 11/26/11 Time: 09:41 Sample: 19 31
Included observations: 13
Variable C X
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood
Coefficient -567.8182 0.823005
Std. Error 604.7985 0.094621
t-Statistic -0.938855 8.697894
Prob. 0.3680 0.0000 4429.221 1831.121 16.02687 16.11379 75.65337
0.873057 Mean dependent var 0.861517 S.D. dependent var 681.4199 Akaike info criterion 5107665. Schwarz criterion -102.1747 F-statistic
Durbin-Watson stat
2.577712 Prob(F-statistic)
0.000003
3)根据上面两表求出F统计量。
由表1得到残差平方和为∑e1t=6867037,由表2得到的残差平方和为∑e2t=5107665,根据Goldfeld-Quanadt检验,F统计量为
F=
22ee22t2=
1t5107665=0.7438
6867037在=0.05下,上式中分子、分母的自由度均为10,查F分布表的临界值F0.05(10, 10)=2.98,因为F=0.7438 根据White检验构造的辅助函数为: t=0+1xt+2xt+vt 经估计出现的White估计结果如下: White Heteroskedasticity Test: F-statistic Obs*R-squared Test Equation: Dependent Variable: RESID^2 Method: Least Squares Date: 11/26/11 Time: 09:44 Sample: 1 31 Included observations: 31 Variable Coefficient Std. Error 7.194463 Probability 10.52295 Probability t-Statistic 0.003011 0.005188 Prob. 22 C X X^2 R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat 69872.27 -72.02221 0.020337 641389.0 248.7240 0.020627 0.108939 -0.289567 0.985972 0.9140 0.7743 0.3326 227944.3 592250.3 29.16732 29.30610 7.194463 0.003011 0.339450 Mean dependent var 0.292268 S.D. dependent var 498241.3 Akaike info criterion 6.95E+12 Schwarz criterion -449.0935 F-statistic 2.430258 Prob(F-statistic) 从上表可以看出,nR2=10.52295由White检验知,在=0.05下,查分布表,得临界值0.05(2)=5.9915,同时X和X2的t检验值也显著。比较计算的统计量与临界值,因为nR2=10.52295> 222 20.05(2)=5.9915,所以拒绝原假设,不拒绝备择假设,表明模型 存在异方差。 四.异方差性的修正 运用加权最小二乘(WLS),选用权数1t= 11Xt,2t = 1X22t,3t= X t得到的用权数的结果如下: Dependent Variable: Y Method: Least Squares Date: 11/24/11 Time: 22:00 Sample: 1 31 Included observations: 31 Weighting series: W2 Variable C X R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood Durbin-Watson stat R-squared Adjusted R-squared S.E. of regression Durbin-Watson stat Coefficient 787.2847 0.561472 Prob. 0.0001 0.0000 2743.600 1165.059 14.13528 14.22780 101.4992 0.000000 Std. Error 173.6964 0.055731 t-Statistic 4.532534 10.07468 Weighted Statistics 0.946060 Mean dependent var 0.944200 S.D. dependent var 275.2095 Akaike info criterion 2196468. Schwarz criterion -217.0969 F-statistic 2.482750 Prob(F-statistic) Unweighted Statistics 3376.309 1499.612 10254472 0.848003 Mean dependent var 0.842762 S.D. dependent var 594.6448 Sum squared resid 1.741955 上表的估计结果如下: Yt=787.2847 + 0.561472Xt (4.531534) (10.07468) R 2=0.946060,DW=2.482750,F=101.4992 括号中数据为t统计量值。 可以看出运用加权最小二乘法消除了异方差性后,参数的t检验均显著,F检验也显著,并说明每增加1元钱,平均来说将增加0.561元的消费支出,而不是起初模型设定中得到的增加0.7195元消费支 出。 因篇幅问题不能全部显示,请点此查看更多更全内容