xtregar命令得出的结论差异怎么这么大
发布网友
发布时间：2022-04-27 08:52
共2个回答
热心网友
时间：2023-09-15 12:00
下面是STATA的输出结果，可发现，运用xtserial命令发现存在一阶自相关，然后，运用xtregar命令，得出的变量 lf_p的估计系数为 .0262005， lf_p的平方项为 f2，其估计系数为-.0105111，但是其他的命令如xtscc以及考虑了截面异方差的命令都与xtregar得出的系数符号完全相反；也就是说，根据xtregar得出一个倒U型曲线，而其他命令（包括xtreg,fe）得出结论是U型曲线，如何处理这个问题啊？？？？？
local fin "lf_p"
. *local fin "lf_r"
. *local fin "lf_g"
. dropvars f2
. gen f2=`fin'*`fin'
(28 missing values generated)
. xtreg vol `fin' f2 lurban le lstu,fe
Fixed-effects (within) regression Number of obs = 644
Group variable: id Number of groups = 28
R-sq: within = 0.2958 Obs per group: min = 23
between = 0.0083 avg = 23.0
overall = 0.1280 max = 23
F(5,611) = 51.32
corr(u_i, Xb) = -0.6454 Prob > F = 0.0000
------------------------------------------------------------------------------
vol | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lf_p | -.0529708 .0102403 -5.17 0.000 -.0730814 -.0328603
f2 | .021836 .0051751 4.22 0.000 .0116729 .031999
lurban | -.0403943 .0093836 -4.30 0.000 -.0588223 -.0219664
le | -.6707866 .1402669 -4.78 0.000 -.9462504 -.3953229
lstu | .0070028 .0030566 2.29 0.022 .0010001 .0130056
_cons | .0611197 .0066801 9.15 0.000 .048001 .0742385
-------------+----------------------------------------------------------------
sigma_u | .01223642
sigma_e | .01482639
rho | .40516623 (fraction of variance e to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(27, 611) = 7.46 Prob > F = 0.0000
. xtserial vol `fin' f2 lurban le lstu //存在一阶相关
Wooldridge test for autocorrelation in panel data
H0: no first order autocorrelation
F( 1, 27) = 1189.581
Prob > F = 0.0000
. xtregar vol `fin' f2 lurban le lstu,fe twostep //考虑一介相关
FE (within) regression with AR(1) disturbances Number of obs = 616
Group variable: id Number of groups = 28
R-sq: within = 0.0311 Obs per group: min = 22
between = 0.0115 avg = 22.0
overall = 0.0397 max = 22
F(5,583) = 3.74
corr(u_i, Xb) = -0.4803 Prob > F = 0.0024
------------------------------------------------------------------------------
vol | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lf_p | .0262005 .0126274 2.07 0.038 .0013998 .0510013
f2 | -.0105111 .0065709 -1.60 0.110 -.0234167 .0023945
lurban | -.020102 .0106012 -1.90 0.058 -.0409233 .0007193
le | -.5069736 .2250153 -2.25 0.025 -.9489129 -.0650342
lstu | -.0004315 .0033019 -0.13 0.896 -.0069166 .0060537
_cons | .0202119 .0019293 10.48 0.000 .0164227 .0240011
-------------+----------------------------------------------------------------
rho_ar | .7705868
sigma_u | .00814859
sigma_e | .00905381
rho_fov | .44752335 (fraction of variance because of u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(27,583) = 0.88 Prob > F = 0.6375
. xtreg vol `fin' f2 lurban le lstu,fe cluster(id) // 考虑截面相关
Fixed-effects (within) regression Number of obs = 644
Group variable: id Number of groups = 28
R-sq: within = 0.2958 Obs per group: min = 23
between = 0.0083 avg = 23.0
overall = 0.1280 max = 23
F(5,27) = 16.67
corr(u_i, Xb) = -0.6454 Prob > F = 0.0000
(Std. Err. adjusted for 28 clusters in id)
------------------------------------------------------------------------------
| Robust
vol | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lf_p | -.0529708 .0130689 -4.05 0.000 -.079786 -.0261557
f2 | .021836 .0054313 4.02 0.000 .0106918 .0329801
lurban | -.0403943 .0127516 -3.17 0.004 -.0665585 -.0142302
le | -.6707866 .2175747 -3.08 0.005 -1.117213 -.2243603
lstu | .0070028 .0058296 1.20 0.240 -.0049586 .0189643
_cons | .0611197 .0109355 5.59 0.000 .038682 .0835575
-------------+----------------------------------------------------------------
sigma_u | .01223642
sigma_e | .01482639
rho | .40516623 (fraction of variance e to u_i)
------------------------------------------------------------------------------
.
. xtscc vol `fin' f2 lurban le lstu,fe lag(1) //同时考虑异方差、自相关、截面相关
Regression with Driscoll-Kraay standard errors Number of obs = 644
Method: Fixed-effects regression Number of groups = 28
Group variable (i): id F( 5, 27) = 15.60
maximum lag: 1 Prob > F = 0.0000
within R-squared = 0.2958
------------------------------------------------------------------------------
| Drisc/Kraay
vol | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lf_p | -.0529708 .0102028 -5.19 0.000 -.0739052 -.0320364
f2 | .021836 .0048789 4.48 0.000 .0118253 .0318466
lurban | -.0403943 .0147506 -2.74 0.011 -.0706601 -.0101286
le | -.6707866 .2246327 -2.99 0.006 -1.131695 -.2098784
lstu | .0070028 .0057805 1.21 0.236 -.0048578 .0188635
_cons | .0611197 .0111259 5.49 0.000 .0382913 .0839482
------------------------------------------------------------------------------
. xtivreg2 vol `fin' f2 lurban le lstu,fe bw(1) robust small
FIXED EFFECTS ESTIMATION
------------------------
Number of groups = 28 Obs per group: min = 23
avg = 23.0
max = 23
OLS estimation
--------------
Estimates efficient for homoskedasticity only
Statistics robust to heteroskedasticity and autocorrelation
kernel=Bartlett; bandwidth= 1
time variable (t): .
group variable (i): id
Number of obs = 644
F( 5, 611) = 51.44
Prob > F = 0.0000
Total (centered) SS = .1907200624 Centered R2 = 0.2958
Total (uncentered) SS = .1907200624 Uncentered R2 = 0.2958
Resial SS = .1343111987 Root MSE = .01483
------------------------------------------------------------------------------
| Robust
vol | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lf_p | -.0529708 .0087502 -6.05 0.000 -.0701549 -.0357868
f2 | .021836 .003894 5.61 0.000 .0141888 .0294831
lurban | -.0403943 .008565 -4.72 0.000 -.0572148 -.0235738
le | -.6707866 .1250726 -5.36 0.000 -.9164111 -.4251622
lstu | .0070028 .003097 2.26 0.024 .0009208 .0130849
------------------------------------------------------------------------------
Included instruments: lf_p f2 lurban le lstu
------------------------------------------------------------------------------
建议你在 FE 模型设定中加入年度虚拟变量，看看结果有何变化，在此基础上再执行序列相关检验和后续分析。
加入年度虚拟变量后，发现，最核心的解释变量变得完全不显著了，整个理论预期完全错误了，无法得到证实。
后来，对被解释变量取了自然对数，再按照上面的步骤，仍然发现，最核心的解释变量是完全不显著的。
如何处理这个问题呢？是不是要放弃这个研究呢？
建议你采用 xtreg,fe robust 命令即可。
热心网友
时间：2023-09-15 12:00
