全部版块 我的主页
› 论坛 › 提问 悬赏 求职 新闻 读书 功能一区 › 经管百科 › 爱问频道
stata 中面板数据存在异方差怎么办？
zhonghh2
32144 15
2012-02-04
stata结果如下：
. xtreg p1 dg ds dt fg fs ft, re
Random-effects GLS regression Number of obs = 330
Group variable: i Number of groups = 30
R-sq: within = 0.2257 Obs per group: min = 11
between = 0.7572 avg = 11.0
overall = 0.6088 max = 11
Wald chi2(6) = 109.77
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000
p1 Coef. Std. Err. z P>z [95% Conf. Interval]
dg .0111848 .0029551 3.78 0.000 .0053928 .0169767
ds -12.64256 4.125702 -3.06 0.002 -20.72878 -4.556329
dt .7687046 .3501209 2.20 0.028 .0824803 1.454929
fg .0076898 .0045073 1.71 0.088 -.0011442 .0165239
fs -1.079189 1.334719 -0.81 0.419 -3.69519 1.536813
ft -.2296786 .4661017 -0.49 0.622 -1.143221 .683864
_cons 760.8861 73.09908 10.41 0.000 617.6146 904.1577
sigma_u 293.58827
sigma_e 149.46782
rho .79416142 (fraction of variance due to u_i)
. est store re
. xtreg p1 dg ds dt fg fs ft, fe
Fixed-effects (within) regression Number of obs = 330
Group variable: i Number of groups = 30
R-sq: within = 0.2304 Obs per group: min = 11
between = 0.7418 avg = 11.0
overall = 0.5704 max = 11
F(6,294) = 14.67
corr(u_i, Xb) = 0.6598 Prob > F = 0.0000
p1 Coef. Std. Err. t P>t [95% Conf. Interval]
dg .0086808 .0028059 3.09 0.002 .0031586 .0142031
ds -7.226417 4.020715 -1.80 0.073 -15.13945 .6866149
dt .5928148 .3307276 1.79 0.074 -.0580789 1.243708
fg .0078365 .0043386 1.81 0.072 -.0007021 .0163751
fs -.9647257 1.238477 -0.78 0.437 -3.40213 1.472679
ft -.2333571 .4358886 -0.54 0.593 -1.091215 .6245004
_cons 724.6549 41.60323 17.42 0.000 642.777 806.5328
sigma_u 532.70478
sigma_e 149.46782
rho .92701884 (fraction of variance due to u_i)
F test that all u_i=0: F(29, 294) = 63.95 Prob > F = 0.0000
. est store fe
. hausman re fe
---- Coefficients ----
(b) (B) (b-B) sqrt(diag(V_b-V_B))
re fe Difference S.E.
dg .0111848 .0086808 .0025039 .0009272
ds -12.64256 -7.226417 -5.416139 .9248032
dt .7687046 .5928148 .1758898 .1149081
fg .0076898 .0078365 -.0001467 .0012216
fs -1.079189 -.9647257 -.1144631 .497644
ft -.2296786 -.2333571 .0036785 .1650815
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg
Test: Ho: difference in coefficients not systematic
chi2(6) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 110.91
Prob>chi2 = 0.0000

选择固定效应模型
固定效应异方差检验结果如下：
. xttest3
Modified Wald test for groupwiseheteroskedasticity
in fixed effect regression model
H0: sigma(i)^2 = sigma^2 for alli
chi2 (30) = 42557.67
Prob>chi2 = 0.0000

表明存在异方差了。
怎么修正呢？
难道是用FGLS估计吗？
. xtgls p1 dg ds dt fg fs ft, panel(heteroskedastic)
Cross-sectional time-series FGLS regression
Coefficients: generalized least squares
Panels: heteroskedastic
Correlation: no autocorrelation
Estimated covariances = 30 Number of obs =330
Estimated autocorrelations = 0 Number of groups =30
Estimated coefficients = 7 Time periods =11
Wald chi2(6) =590.00
Prob > chi2 =0.0000
p1 Coef. Std. Err. z P>z [95% Conf.Interval]
dg .042909 .0035188 12.19 0.000 .0360123.0498056
ds -29.51432 3.37747 -8.74 0.000 -36.13404-22.8946
dt .1694073 .4048436 0.42 0.676 -.6240716.9628861
fg .0108368 .0054424 1.99 0.046 .0001699.0215037
fs -3.017731 1.849226 -1.63 0.103 -6.642148.6066857
ft .4162204 .4940019 0.84 0.399 -.55200551.384446
_cons 822.8147 52.76306 15.59 0.000 719.401926.2284

这个结果还可以说是固定效应模型吗？
另外PCSE的结果如下：
. xtpcse p1 dg ds dt fg fs ft
Linear regression, correlated panels correctedstandard errors (PCSEs)
Group variable: iNumber of obs = 330
Time variable: tNumber of groups = 30
Panels: correlated (balanced)Obs per group: min = 11
Autocorrelation: no autocorrelationavg = 11
max = 11
Estimated covariances = 465R-squared = 0.6425
Estimated autocorrelations = 0Wald chi2(6) = 289.94
Estimated coefficients = 7Prob > chi2 = 0.0000
Panel-corrected
p1 Coef. Std. Err. zP>z [95% Conf. Interval]
dg .0383731 .0054828 7.000.000 .027627 .0491193
ds -32.65586 3.48672 -9.370.000 -39.48971 -25.82202
dt -.0715003 .6054877 -0.120.906 -1.258234 1.115234
fg .0172674 .0052236 3.310.001 .0070293 .0275054
fs -4.399319 1.756286 -2.500.012 -7.841576 -.9570619
ft 1.14975 .6960338 1.650.099 -.2144515 2.513951
_cons 896.0825 38.4172 23.330.000 820.7862 971.3789

以上结果有什么区别？怎么选择？异方差问题怎么处理啊？
请高手指点一下，俺的课程论文快要的deadline啦。
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

全部回复
夸克之一
2012-2-4 11:54:30
最简单的办法在 , fe后面写上 r

xtreg p1 dg ds dt fg fs ft, fe r

如果你是省级面板，令省id为 prov,

xtreg p1 dg ds dt fg fs ft, fe cluster(prov)
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

zhonghh2
2012-2-4 12:52:52
夸克之一 发表于 2012-2-4 11:54
最简单的办法在 , fe后面写上 r

xtreg p1 dg ds dt fg fs ft, fe r
为什么按照您的方法弄出来的结果没有一个变量是显著的呢？
在固定效应模型下，就只有这种方法修正异方差吗？
前面提到的FGLS和PCSE估计结果到底是什么模型？
虽然检验出来使用固定效应模型比较好，那么还可以选择用混合模型的吗？因为选择模型的检验是在异方差情况下的结果啊，能不能在修正异方差的情况下在检验那种模型比较好呢？
. xtreg p1 dg ds dt fg fs ft, fe
Fixed-effects (within) regression Number of obs = 330
Group variable: i Number of groups = 30
R-sq: within = 0.2304 Obs per group: min = 11
between = 0.7418 avg = 11.0
overall = 0.5704 max = 11
F(6,294) = 14.67
corr(u_i, Xb) = 0.6598 Prob > F = 0.0000
p1 Coef. Std. Err. t P>t [95% Conf. Interval]
dg .0086808 .0028059 3.09 0.002 .0031586 .0142031
ds -7.226417 4.020715 -1.80 0.073 -15.13945 .6866149
dt .5928148 .3307276 1.79 0.074 -.0580789 1.243708
fg .0078365 .0043386 1.81 0.072 -.0007021 .0163751
fs -.9647257 1.238477 -0.78 0.437 -3.40213 1.472679
ft -.2333571 .4358886 -0.54 0.593 -1.091215 .6245004
_cons 724.6549 41.60323 17.42 0.000 642.777 806.5328
sigma_u 532.70478
sigma_e 149.46782
rho .92701884 (fraction of variance due to u_i)
F test that all u_i=0: F(29, 294) = 63.95 Prob > F = 0.0000

. xtreg p1 dg ds dt fg fs ft, fe r
Fixed-effects (within) regression Number of obs = 330
Group variable: i Number of groups = 30
R-sq: within = 0.2304 Obs per group: min = 11
between = 0.7418 avg = 11.0
overall = 0.5704 max = 11
F(6,29) = 1.93
corr(u_i, Xb) = 0.6598 Prob > F = 0.1094
(Std. Err. adjusted for 30 clusters in i)
Robust
p1 Coef. Std. Err. t P>t [95% Conf. Interval]
dg .0086808 .009626 0.90 0.375 -.0110065 .0283682
ds -7.226417 6.351198 -1.14 0.265 -20.21608 5.763241
dt .5928148 .6395464 0.93 0.362 -.7152045 1.900834
fg .0078365 .0138696 0.57 0.576 -.02053 .036203
fs -.9647257 1.322179 -0.73 0.471 -3.668886 1.739434
ft -.2333571 .5973154 -0.39 0.699 -1.455004 .9882901
_cons 724.6549 48.23277 15.02 0.000 626.0078 823.302
sigma_u 532.70478
sigma_e 149.46782
rho .92701884 (fraction of variance due to u_i)
. xtreg p1 dg ds dt fg fs ft, fe cluster( i)
Fixed-effects (within) regression Number of obs = 330
Group variable: i Number of groups = 30
R-sq: within = 0.2304 Obs per group: min = 11
between = 0.7418 avg = 11.0
overall = 0.5704 max = 11
F(6,29) = 1.93
corr(u_i, Xb) = 0.6598 Prob > F = 0.1094
(Std. Err. adjusted for 30 clusters in i)
Robust
p1 Coef. Std. Err. t P>t [95% Conf. Interval]
dg .0086808 .009626 0.90 0.375 -.0110065 .0283682
ds -7.226417 6.351198 -1.14 0.265 -20.21608 5.763241
dt .5928148 .6395464 0.93 0.362 -.7152045 1.900834
fg .0078365 .0138696 0.57 0.576 -.02053 .036203
fs -.9647257 1.322179 -0.73 0.471 -3.668886 1.739434
ft -.2333571 .5973154 -0.39 0.699 -1.455004 .9882901
_cons 724.6549 48.23277 15.02 0.000 626.0078 823.302
sigma_u 532.70478
sigma_e 149.46782
rho .92701884 (fraction of variance due to u_i)
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

夸克之一
2012-2-4 13:11:03
你的数据本来就存在异方差，当标准误校正之后，系数不显著不是很正常么？

https://bbs.pinggu.org/forum.php? ... ble&tid=1238175
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

zhonghh2
2012-2-4 13:36:39
夸克之一 发表于 2012-2-4 13:11
你的数据本来就存在异方差，当标准误校正之后，系数不显著不是很正常么？

https://bbs.pinggu.org/forum. ...
其它问题呢？
数据处理没问题的，而且有人也做出结果来了。变量都不显著，这算是什么问题呢？
谢啦。
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

weberxu
2012-2-4 15:35:23
看过
二维码

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

点击查看更多内容…
相关推荐
栏目导航
热门文章
推荐文章

说点什么

分享

微信QQ空间QQ微博
扫码加好友，拉您进群
各岗位、行业、专业交流群