在申请出国读学位的16名学生中有如下GRE数量与词汇分数。 学生编号 1 2 3 4 5 6 7 8 数量成绩Q 760 600 720 710 530 650 800 650 词汇成绩V 550 350 320 630 430 570 500 680 是否准入Y(1=准,0=不准) 1 0 0 1 1 0 1 1 学生编号 9 10 11 12 13 14 15 16
数量成绩Q 520 800 670 670 780 520 680 500 词汇成绩V 660 250 480 520 710 450 590 380
是否准入Y(1=准,0=不准) 1 0 0 1 1 0 1 0 解:根据Eview软件得如下表:
Dependent Variable: Y
Method: ML - Binary Logit (Quadratic hill climbing) Date: 05/22/11 Time: 22:19 Sample: 1 16
Included observations: 16
Convergence achieved after 5 iterations
Covariance matrix computed using second derivatives
Variable C Q V
McFadden R-squared S.D. dependent var Akaike info criterion Schwarz criterion Hannan-Quinn criter. LR statistic Prob(LR statistic) Obs with Dep=0 Obs with Dep=1
Coefficient -11.10741 0.003968 0.017696
Std. Error 6.124290 0.008008 0.008752
z-Statistic -1.813665 0.495515 2.021914
Prob. 0.0697 0.6202 0.0432 0.562500 0.382391 1.900896 -5.827681 -10.96503 -0.364230
16
0.468521 Mean dependent var 0.512348 S.E. of regression 1.103460 Sum squared resid 1.248321 Log likelihood 1.110878 Restr. log likelihood 10.27469 Avg. log likelihood 0.005873
7 Total obs 9
于是,我们可得到Logit模型为:
ˆ11.1070.004Q0.0177V Yi (-1.81) (0.49) (2.02)
2 RMCF0.4685 , LR(2)=10.27
如果在Binary estination这一栏中选择Probit估计方法,可得到如下表:
Dependent Variable: Y
Method: ML - Binary Probit (Quadratic hill climbing) Date: 05/22/11 Time: 22:25 Sample: 1 16
Included observations: 16
Convergence achieved after 5 iterations
Covariance matrix computed using second derivatives
Variable C Q V
McFadden R-squared S.D. dependent var Akaike info criterion Schwarz criterion Hannan-Quinn criter. LR statistic Prob(LR statistic) Obs with Dep=0 Obs with Dep=1
Coefficient -6.634542 0.002403 0.010532
Std. Error 3.396882 0.004585 0.004693
z-Statistic -1.953127 0.524121 2.244299
Prob. 0.0508 0.6002 0.0248 0.562500 0.381655 1.893588 -5.742687 -10.96503 -0.358918
16
0.476272 Mean dependent var 0.512348 S.E. of regression 1.092836 Sum squared resid 1.237696 Log likelihood 1.100254 Restr. log likelihood 10.44468 Avg. log likelihood 0.005395
7 Total obs 9
于是,我们可得到Probit模型为:
ˆ6.6350.0024Q0.0105V Yi (-1.95) (0.52) (2.24)
2 RMCF0.4763 , LR(2)=10.44
第7章练习6
下表列出了美国、加拿大、英国在1980~1999年的失业率Y以及对制造业的补偿X的相关数据资料。 美 国 加拿大 英 国 年份 补助X 失业率Y 补助X 失业率Y 补助X 失业率Y (美元/小时) (%) (美元/小时) (%) (美元/小时) (%) 1980 1981 1982 1983 1984 1985 1986 55.6 61.1 67.0 68.8 71.2 75.1 78.5 7.1 7.6 9.7 9.6 7.5 7.2 7.0 49 54.1 59.6 63.9 64.3 63.5 63.3 7.2 7.3 10.6 11.5 10.9 10.2 9.2 43.7 44.1 42.2 39.0 37.2 39.0 47.8 7.0 10.5 11.3 11.8 11.7 11.2 11.2 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 80.7 64.0 86.6 90.8 95.6 100.0 102.7 105.6 107.9 109.3 111.4 117.3 123.2 6.2 5.5 5.3 5.6 6.8 7.5 6.9 6.1 5.6 5.4 4.9 4.5 4.9 68.0 76.0 84.1 91.5 100.1 100.0 95.5 91.7 93.3 93.1 94.4 90.6 91.9 8.4 7.3 7.0 7.7 9.8 10.6 10.7 9.4 8.5 8.7 8.2 7.5 5.7 60.2 68.3 67.7 81.7 90.5 100.0 88.7 92.3 95.9 95.6 103.3 109.8 112.2 10.3 8.6 7.2 6.9 8.8 10.1 10.5 9.7 8.7 8.2 7.0 6.3 6.1 解:(1)根据Eview 软件操作得如下表: 美国(US): Dependent Variable: Y Method: Least Squares Date: 05/22/11 Time: 22:38 Sample: 1980 1999 Included observations: 20
Variable C X
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
Coefficient 10.56858 -0.045403
Std. Error 1.138982 0.012538
t-Statistic 9.278972 -3.621189
Prob. 0.0000 0.0020 6.545000 1.432875 3.158696 3.258269 3.178133 0.797022
0.421464 Mean dependent var 0.389323 S.D. dependent var 1.119732 Akaike info criterion 22.56840 Schwarz criterion -29.58696 Hannan-Quinn criter. 13.11301 Durbin-Watson stat 0.001953
ˆ10.56860.0454X 根据上表可得对美国的OLS估计结果为:Ytt (9.28) (-3.62) R0.4215, R0.3893, D.W.=0.797, RSS=22.57
加拿大(CA):
Dependent Variable: Y Method: Least Squares Date: 05/22/11 Time: 22:43 Sample: 1980 1999 Included observations: 20
22 Variable C X
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
Coefficient 9.342452 -0.006580
Std. Error 1.810701 0.022333
t-Statistic 5.159579 -0.294648
Prob. 0.0001 0.7716 8.820000 1.600855 3.922848 4.022421 3.942286 0.578517
0.004800 Mean dependent var -0.050489 S.D. dependent var 1.640770 Akaike info criterion 48.45828 Schwarz criterion -37.22848 Hannan-Quinn criter. 0.086817 Durbin-Watson stat 0.771634
同样,根据上表可得对加拿大(CA)的OLS估计结果为:
ˆ9.34250.0066X Ytt (5.16) (-0.29)
R0.0048, R0.05, D.W.=0.579, RSS=48.46
英国(UK): Dependent Variable: Y Method: Least Squares Date: 05/22/11 Time: 22:48 Sample: 1980 1999 Included observations: 20
Variable C X
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
Coefficient 12.55426 -0.046591
Std. Error 0.990234 0.012777
t-Statistic 12.67808 -3.646353
Prob. 0.0000 0.0018 9.155000 1.916542 3.734513 3.834087 3.753951 0.698064
22 0.424845 Mean dependent var 0.392891 S.D. dependent var 1.493315 Akaike info criterion 40.13981 Schwarz criterion -35.34513 Hannan-Quinn criter. 13.29589 Durbin-Watson stat 0.001847
同样,根据上表可得对英国(UK)的OLS估计结果为:
ˆ12.55430.0466X Ytt (12.68) (-3.65)
R0.3036, R3.929, D.W.=0.6981, RSS=40.14
(2)将三个国家的数据合并成一个样本(共60个样本点),根据Eview软件得:OLS估计结果如下:
22Dependent Variable: Y Method: Least Squares Date: 05/22/11 Time: 22:58 Sample: 1980 2039 Included observations: 60
Variable C X
R-squared Adjusted R-squared S.E. of regression Sum squared resid Log likelihood F-statistic Prob(F-statistic)
Coefficient 12.14946 -0.049500
Std. Error 0.820266 0.009844
t-Statistic 14.81161 -5.028729
Prob. 0.0000 0.0000 8.173333 2.009120 3.921268 3.991079 3.948575 0.492398
0.303622 Mean dependent var 0.291616 S.D. dependent var 1.690988 Akaike info criterion 165.8475 Schwarz criterion -115.6380 Hannan-Quinn criter. 25.28811 Durbin-Watson stat 0.000005
ˆ12.1490.0495X 根据上表得估计方程为:Ytt (14.81) (-5.03)
R0.3036, R0.2916, D.W.=0.49, RSS=165.85
(3)在Eviews软件下,估计变截距固定影响模型得到如下结果:
22
固定影响模型可按最小二乘虚拟变量(LSDV)模型估计,记D2为加拿大(CA)的虚拟变
量;即观测值属于CA时取值为1,其他取值为0;记D3为英国的虚拟变量,取值规律同
D2,所以,LSDV模型的OLS估计结果如下: Yit9.93481.9221D22.011D30.0383X (11.73) (4.12) (4.20) (-4.33)
R20.5048, R20.4783, D.W.=0.664, RSS=117.94
美国(US)没有设定虚拟变量,成为比较的基准。可以看出,该结果与上述固定效应模型的估计结果是一致的。
(4)为了比较以上三个模型,需要进行如下两个F检验。
首先,进行“截距和斜率在不同的横截面样本点和时间上都相同”的假设检验,相应的F检验为: F2S3-S1/n1k1~F[(n-1)(k+1),nT-n(k+1)]
S1/n1k1,nTnk1其中,S3为第二类模型,即合成的大样本模型相应的残差平方和,S1为第一类模型,即按横截面样本点分别估计的各单一方程的残差平方和。
如果接受该假设,则选取第二类模型。如果该假设被拒绝,则再进行“斜率在不同的横截面样本点和时间上都相同,但截距不相同”的假设,相应的F检验为:
F2S2-S1/n1k~F[(n-1)k,nT-n(k+1)] S1/nTnk1其中,S2为第三类模型,即固定效应模型的相应的残差平方和。如果接受该假设,则选取第三类模型。拒绝该假设,则选取第一类模型,即按横截面样本点分别估计的各单一的模型方程。
由上述估计结果,知:
S122.5748.4640.14111.17 S2117.94 S3165.85
于是, F2=6.64, F1=1.64
对于F2,在5%的显著性水平下,自由度为(4,54)的F分布的临界值为F0.054,542.54,可见拒绝“截距和斜率在不同的横截面样本点和时间上都相同”的假设。
对于F1,在5%的显著性水平下,相应的临界值分别为F0.052,543.17,可见接受假设“斜率在不同的横截面样本点和时间上都相同,但截距不相同”,表明应该选取第三类型模型,即固定效应模型来估计。
第7章练习7
用普通最小二乘法(OLS)估计固定一下变系数模型得到:
用广义最小二乘法(GLS)估计固定一下变系数模型得到:
可以看出,变系数固定效应模型OLS估计与GLS估计的参数是相同的,但检验指标不相同。GLS估计只使得美国的斜率项的t检验值变大,但使得加拿大与英国的斜率项的t检验值略有变小。当然,GLS估计使得R变小了。
2
因篇幅问题不能全部显示,请点此查看更多更全内容