分组检验中，如果数据中control变量，分别以1，2，3，4代表中央国有、地方国有、私有、其他产权性质，那么两两分组比较是否可行？

例如：中央国有和地方国有分组比较

          xtreg y x1 x2 x3 if control==1,fe 
          mat b1 = e(b)
         xtreg y x1 x2 x3 if control==2,fe        
          mat b2 = e(b)
        mat D0 = b1 - b2
        mat list D0, title(系数估计值的真实差异)

但我运算后发现，以上回归结果与单独运算xtreg y x1 x2 x3 if control==1,fe和xtreg y x1 x2 x3 if control==2,fe的结果不一致，这是为什么？

恳请答复，谢谢！

其实，从Bootstrap的原理来看，对于unbalanced Panel Data, 可以按照视频上介绍的方法计算 empirical p-value.

之所以在视频中仅介绍了balanced Panel 中的 经验p值 的获取方法，是因为它比较容易理解。

不知你所言的“不一致”具体是什么情况。我想两个结果是一样的，差别可能仅在于小数点后的第4位以后的数值。

你可以把相应的结果贴出来，我看一下。

分别运算结果：

（1）

xtreg y x1 x2 x3 x4 x5 if family==1,fe

Fixed-effects (within) regression               Number of obs      =       630
Group variable: id_bs                           Number of groups   =       147

R-sq:  within  = 0.5228                         Obs per group: min =         1
       between = 0.3872                                        avg =       4.3
       overall = 0.4029                                        max =         5

                                                F(5,478)           =    104.74
corr(u_i, Xb)  = -0.2919                        Prob > F           =    0.0000

------------------------------------------------------------------------------
      y      |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         x1  |  -.0409221   .0312997    -1.31   0.192    -.1024241      .02058
         x2  |   .1107661   .0221773     4.99   0.000     .0671891    .1543432
         x3  |  -2.729631   .1490609   -18.31   0.000    -3.022527   -2.436735
         x4  |   .1448393   .1076614     1.35   0.179    -.0667088    .3563875
       x5    |   .3000972   .0267498    11.22   0.000     .2475355    .3526589
       _cons |  -2.932661   2.179672    -1.35   0.179    -7.215585    1.350263
-------------+----------------------------------------------------------------
     sigma_u |  .68934858
     sigma_e |  .59934227
         rho |  .56950417   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(146, 478) =     3.11            Prob > F = 0.0000

（2）

 xtreg y x1 x2 x3 x4 x5 if localstate==1,fe

Fixed-effects (within) regression               Number of obs      =      1367
Group variable: id_bs                           Number of groups   =       301

R-sq:  within  = 0.1628                         Obs per group: min =         2
       between = 0.2076                                        avg =       4.5
       overall = 0.1844                                        max =         5

                                                F(5,1061)          =     41.27
corr(u_i, Xb)  = -0.4603                        Prob > F           =    0.0000

------------------------------------------------------------------------------
        y    |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         x1  |  -.0528663   .0767341    -0.69   0.491    -.2034341    .0977015
         x2  |    .008672   .0105656     0.82   0.412    -.0120599    .0294039
         x3  |  -.9128659   .1300254    -7.02   0.000    -1.168002   -.6577297
         x4  |   .6642081   .0641612    10.35   0.000     .5383109    .7901053
         x5  |   .0262315   .0055112     4.76   0.000     .0154174    .0370455
       _cons |  -13.76913    1.31836   -10.44   0.000    -16.35602   -11.18224
-------------+----------------------------------------------------------------
     sigma_u |  .56350595
     sigma_e |  .41748773
         rho |  .64562096   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(300, 1061) =     6.09           Prob > F = 0.0000

计算真实差异的结果：

（3）

 xtreg y x1 x2 x3 x4 x5  if family==1,fe

Fixed-effects (within) regression               Number of obs      =       656
Group variable: id_bs                           Number of groups   =       147

R-sq:  within  = 0.3294                         Obs per group: min =         1
       between = 0.2852                                        avg =       4.5
       overall = 0.2892                                        max =         5

                                                F(5,504)           =     49.51
corr(u_i, Xb)  = -0.2760                        Prob > F           =    0.0000

------------------------------------------------------------------------------
       y     |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         x1  |  -.0112587   .0281915    -0.40   0.690    -.0666461    .0441287
         x2  |  -.0210558   .0199427    -1.06   0.292    -.0602368    .0181252
         x3  |  -.9141472   .1121804    -8.15   0.000    -1.134546   -.6937485
         x4  |   .6123647     .11324     5.41   0.000     .3898841    .8348454
         x5  |   .2046955    .033289     6.15   0.000     .1392931    .2700978
       _cons |  -12.27952   2.295819    -5.35   0.000    -16.79007   -7.768964
-------------+----------------------------------------------------------------
     sigma_u |  .62550323
     sigma_e |   .5312164
         rho |  .58097425   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(146, 504) =     5.89            Prob > F = 0.0000

.           mat b1 = e(b)

.         xtreg y x1 x2 x3 x4 x5 if localstate==1,fe

Fixed-effects (within) regression               Number of obs      =      1315
Group variable: id_bs                           Number of groups   =       288

R-sq:  within  = 0.1353                         Obs per group: min =         2
       between = 0.2694                                        avg =       4.6
       overall = 0.2242                                        max =         5

                                                F(5,1022)          =     31.98
corr(u_i, Xb)  = -0.5170                        Prob > F           =    0.0000

------------------------------------------------------------------------------
      y     |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        x1  |   .0010771   .0728987     0.01   0.988    -.1419712    .1441254
        x2  |  -.0066045   .0099919    -0.66   0.509    -.0262114    .0130024
        x3  |  -.8302784   .1651426    -5.03   0.000    -1.154336    -.506221
        x4  |   .7078328   .0652889    10.84   0.000     .5797173    .8359483
        x5  |   .0120042   .0055449     2.16   0.031     .0011235    .0228849
       _cons |  -14.52583   1.322331   -10.99   0.000    -17.12063   -11.93104
-------------+----------------------------------------------------------------
     sigma_u |  .56598978
     sigma_e |  .39176357
         rho |  .67608441   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(287, 1022) =     6.61           Prob > F = 0.0000

.           mat b3 = e(b)

.         mat D0 = b1 - b3

.         mat list D0, title(系数估计值的真实差异) 

（1）和（2）与（3）中样本数不相同，不明白为什么？

family样本数有955个，localstate样本数有1950个，但各个解释变量数据有残缺。