连老师及各位同学 计量高手们！求助啊！
我的数据是country industry year三维面板数据。
为了能在stata里运行，我采用了egen id=group(country industry) 然后xtset id year
其中industry变量（trade cost intensity）只随着行业变化而不随时间变化 trade cost随着国家和时间变化，而主要的自变量是两者的交叉项，即x=trade cost*trade cost intensity（即自变量随着国家行业时间变化）。y也是随着不同行业不同年份取不同数值的。
由于是模仿一篇外文文献，所以倾向于固定效应。
附件的图片是原英文文献的回归结果。外国学者是做了两个双向固定效应吗？由于他的研究视角是立足于世界，而我将视角转化为中国（country代表的是中国不同的贸易伙伴，industry是中国的各个行业），不知道我的是不是要比他的稍微简单一点。
首先我进行了hausman检验
xtreg exportshare interaction,fe
Fixed-effects (within) regression Number of obs = 8568
Group variable: id Number of groups = 612
R-sq: within = 0.0055 Obs per group: min = 14
between = 0.0472 avg = 14.0
overall = 0.0441 max = 14
F(1,7955) = 43.92
corr(u_i, Xb) = 0.1254 Prob > F = 0.0000
------------------------------------------------------------------------------
exportshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
interaction | -.1009949 .0152393 -6.63 0.000 -.130868 -.0711218
_cons | .0395886 .0028298 13.99 0.000 .0340414 .0451359
-------------+----------------------------------------------------------------
sigma_u | .04657092
sigma_e | .01240273
rho | .93377138 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(611, 7955) = 194.29 Prob > F = 0.0000
est store FE
xtreg exportshare interaction,re
est store RE
hausman FE RE
结果如下：
chi2(1) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 9.13
Prob>chi2 = 0.0025
因P值较小，所以选择固定效应。
下面加入时间虚拟变量考察是否存在双向固定效应：
quietly tab year,gen(dum_t)
xtreg exportshare interaction dum_t*,fe
note: dum_t14 omitted because of collinearity
Fixed-effects (within) regression Number of obs = 8568
Group variable: id Number of groups = 612
R-sq: within = 0.0351 Obs per group: min = 14
between = 0.0472 avg = 14.0
overall = 0.0462 max = 14
F(14,7942) = 20.66
corr(u_i, Xb) = 0.0603 Prob > F = 0.0000
------------------------------------------------------------------------------
exportshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
interaction | -.1721275 .0158927 -10.83 0.000 -.2032814 -.1409736
dum_t1 | .0062445 .0007103 8.79 0.000 .0048521 .0076369
dum_t2 | .0054034 .0007027 7.69 0.000 .0040259 .0067809
dum_t3 | .0050666 .0007004 7.23 0.000 .0036936 .0064395
dum_t4 | .0049283 .0007007 7.03 0.000 .0035547 .0063019
dum_t5 | .0044475 .0006996 6.36 0.000 .003076 .005819
dum_t6 | .0039501 .0006994 5.65 0.000 .002579 .0053212
dum_t7 | .0033088 .000699 4.73 0.000 .0019387 .004679
dum_t8 | .0022767 .0006995 3.25 0.001 .0009054 .003648
dum_t9 | .0016724 .0006991 2.39 0.017 .000302 .0030428
dum_t10 | .0007357 .0007 1.05 0.293 -.0006364 .0021078
dum_t11 | .0006051 .0006994 0.87 0.387 -.000766 .0019761
dum_t12 | .0002728 .0006995 0.39 0.697 -.0010983 .001644
dum_t13 | .0000351 .0006993 0.05 0.960 -.0013357 .0014059
dum_t14 | 0 (omitted)
_cons | .0500007 .0029747 16.81 0.000 .0441695 .055832
-------------+----------------------------------------------------------------
sigma_u | .04627994
sigma_e | .01222642
rho | .9347601 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(611, 7942) = 199.82 Prob > F = 0.0000
.xi: xtreg exportshare interaction i.year,fe(这个命令和上面的等价吗？区别何在？)
i.year _Iyear_1999-2012 (naturally coded; _Iyear_1999 omitted)
Fixed-effects (within) regression Number of obs = 8568
Group variable: id Number of groups = 612
R-sq: within = 0.0351 Obs per group: min = 14
between = 0.0472 avg = 14.0
overall = 0.0462 max = 14
F(14,7942) = 20.66
corr(u_i, Xb) = 0.0603 Prob > F = 0.0000
------------------------------------------------------------------------------
exportshare | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
interaction | -.1721275 .0158927 -10.83 0.000 -.2032814 -.1409736
_Iyear_2000 | -.0008411 .000701 -1.20 0.230 -.0022152 .0005331
_Iyear_2001 | -.0011779 .0007037 -1.67 0.094 -.0025573 .0002014
_Iyear_2002 | -.0013162 .0007031 -1.87 0.061 -.0026945 .0000621
_Iyear_2003 | -.001797 .0007054 -2.55 0.011 -.0031798 -.0004142
_Iyear_2004 | -.0022944 .000706 -3.25 0.001 -.0036784 -.0009104
_Iyear_2005 | -.0029357 .0007089 -4.14 0.000 -.0043253 -.001546
_Iyear_2006 | -.0039678 .000716 -5.54 0.000 -.0053714 -.0025642
_Iyear_2007 | -.0045721 .0007131 -6.41 0.000 -.00597 -.0031743
_Iyear_2008 | -.0055088 .000718 -7.67 0.000 -.0069162 -.0041013
_Iyear_2009 | -.0056394 .0007154 -7.88 0.000 -.0070418 -.0042371
_Iyear_2010 | -.0059717 .0007158 -8.34 0.000 -.0073748 -.0045685
_Iyear_2011 | -.0062094 .0007147 -8.69 0.000 -.0076104 -.0048084
_Iyear_2012 | -.0062445 .0007103 -8.79 0.000 -.0076369 -.0048521
_cons | .0562452 .0030997 18.15 0.000 .050169 .0623214
-------------+----------------------------------------------------------------
sigma_u | .04627994
sigma_e | .01222642
rho | .9347601 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(611, 7942) = 199.82 Prob > F = 0.0000
由此可以判定应该使用双向固定效应模型吗？
如果结果是可以接受的，上述结果该如何解释呢？（尤其是时间虚拟变量部分）
本人刚刚接触stata不久，所提问题难免白痴，还望各位朋友耐心解答，真心感谢！
扫码加好友，拉您进群
全部版块
我的主页
收藏
