r-重现Stata合并OLS结果

Question

我正试图重现Stata的结果。数据集是一个不平衡的面板，看起来像

ï..region id year grpmlnr   grppc   cpi
1  region1   1 1998 18245.5 12242.8 167.7
2  region1   1 1999 32060.6 21398.0 140.8
3  region1   1 2000 42074.5 27969.5 120.9

Stata中的原始回归被合并为reg y x1 x2 x3 x4形式的OLS，并给出以下输出

Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------------+----------------------------------------------------------------
x1 |  -.0045519   .0070808    -0.64   0.520    -.0184413    .0093376
x2 |  -.1598071   .0345597    -4.62   0.000    -.2275982    -.092016
x3 |   4.08e-06   4.16e-06     0.98   0.327    -4.08e-06    .0000122
x4 |  -.0000874   .0000244    -3.58   0.000    -.0001354   -.0000395
_cons |   .2899655   .0655542     4.42   0.000   .1613767    .4185542
Number of obs = 1489, R=squared = 0.0242, Adj R-squared = 0.0216

当我运行时

pooledols<-plm(y~
x1
+ x2
+ x3
+ x4,
data=dataset, index=c('ï..region', 'year'), model='pooling')
summary(pooledols)

我得到

Coefficients:
Estimate  Std. Error t-value  Pr(>|t|)    
(Intercept)         1.1228e-02  6.3812e-02  0.1760 0.8603497    
x1                  3.5982e-03  6.7284e-03  0.5348 0.5928858    
x2                  4.3466e-02  3.1060e-03 13.9943 < 2.2e-16 ***
x3                  1.3737e-05  3.9212e-06  3.5033 0.0004732 ***
x4                  -2.7368e-05  2.3573e-05 -1.1610 0.2458259 

带有

number of obs = 1489, R=squared = 0.12554, and Adj R-squared = 0.12319.

有人有什么建议吗？我相信这两种情况下的数据集是相同的。我在其他地方看到它暗示，对于随机效应模型，Stata和R如何处理不平衡面板很重要，但我不确定这是否与此相关。

编辑：这是我的数据子集，其中x1, x2, x3, x4与回归中使用的变量匹配：

region year x1 x2 x3 x4 y RegionA 1998 9.412693 7.316763 655 212 RegionA 1999 9.412693 4.662889 720 232 0.55836 RegionA 2000 9.412693 3.669467 741 303 0.267817 RegionA 2001 9.412693 3.480852 748 304 0.169225 RegionA 2002 9.412693 3.434518 720 347 0.221187 RegionA 2003 9.412693 3.252523 719 393 0.195911 RegionA 2004 9.412693 2.30941 731 426 0.408409 RegionA 2005 9.412693 2.03653 714 477 0.237577 RegionA 2006 9.412693 1.857329 752 512 0.209052 RegionA 2007 9.412693 1.796764 735 527 0.278823 RegionA 2008 9.412693 1.59614 759 543 0.288872 RegionA 2009 9.412693 1.925464 793 522 -0.04663 RegionA 2010 9.412693 1.685813 779 508 0.267205 RegionA 2011 9.412693 1.570235 767 478 0.241406 RegionA 2012 9.412693 1.689142 787 446 0.068759 RegionA 2013 9.412693 1.819899 810 420 0.03955 RegionA 2014 9.412693 1.859676 814 382 0.083057 RegionA 2015 9.412693 1.860045 806 342 0.11043 RegionA 2016 9.412693 1.921366 822 326 0.048621 RegionA 2017 9.412693 1.911606 823 316 0.074802 RegionB 1998 8.94365 10.81936 633 129 RegionB 1999 8.94365 7.110605 698 152 0.428163 RegionB 2000 8.94365 5.014219 665 192 0.393189 RegionB 2001 8.94365 4.521011 652 208 0.21136 RegionB 2002 8.94365 4.237961 636 276 0.227971 RegionB 2003 8.94365 4.373059 651 301 0.167702 RegionB 2004 8.94365 3.992342 659 320 0.165888 RegionB 2005 8.94365 3.276585 648 345 0.280323 RegionB 2006 8.94365 2.853214 660 392 0.219669 RegionB 2007 8.94365 3.031803 661 401 0.233179 RegionB 2008 8.94365 2.598884 656 457 0.210191 RegionB 2009 8.94365 2.773871 638 472 0.011586 RegionB 2010 8.94365 2.618205 650 443 0.157882 RegionB 2011 8.94365 2.474298 644 410 0.178349 RegionB 2012 8.94365 2.257853 644 387 0.182941 RegionB 2013 8.94365 2.362653 638 336 0.06543 RegionB 2014 8.94365 2.35502 635 320 0.108892 RegionB 2015 8.94365 2.308449 624 282 0.119917 RegionB 2016 8.94365 2.607521 625 252 0.038878 RegionB 2017 8.94365 2.583059 612 223 0.096383 RegionC 1998 9.143153 7.710033 771 120 RegionC 1999 9.143153 4.82562 810 139 0.50267 RegionC 2000 9.143153 4.112946 798 184 0.309938 RegionC 2001 9.143153 3.384044 785 181 0.254107 RegionC 2002 9.143153 3.639285 808 280 0.192077 RegionC 2003 9.143153 3.58782 796 302 0.214723 RegionC 2004 9.143153 2.960462 806 319 0.190094 RegionC 2005 9.143153 2.528599 809 361 0.165926 RegionC 2006 9.143153 2.252368 792 393 0.26823

编辑2：这是第一次回归的结果，与Nick Cox的相同

lm(formula = y ~ x1 + x2 + x3 + x4, data = replicate)
Residuals:
Min       1Q   Median       3Q      Max 
-0.23488 -0.06966  0.00142  0.05492  0.20161 
Coefficients:
Estimate Std. Error t value Pr(>|t|)    
(Intercept) -1.792e+00  8.772e-01  -2.043   0.0475 *  
x1           1.865e-01  1.149e-01   1.623   0.1122    
x2           8.823e-02  1.989e-02   4.437 6.72e-05 ***
x3          -6.175e-05  3.271e-04  -0.189   0.8512    
x4           1.995e-04  2.242e-04   0.890   0.3786` 

Answer 1

发布了49个观察结果(行(；3具有CCD_ 4的缺失值。这是Stata的一个简单回归，没有任何关注面板结构(更不用说任何时间变量(。_cons是估计的截距。我还列出了自动排除的3个观察结果。其他人可能想要发布R结果。

. regress y x1 x2 x3 x4
Source |       SS           df       MS      Number of obs   =        46
-------------+----------------------------------   F(4, 41)        =      7.41
Model |  .287208881         4   .07180222   Prob > F        =    0.0001
Residual |  .397187973        41  .009687512   R-squared       =    0.4197
-------------+----------------------------------   Adj R-squared   =    0.3630
Total |  .684396854        45  .015208819   Root MSE        =    .09843
------------------------------------------------------------------------------
y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
x1 |   .1864634   .1148715     1.62   0.112    -.0455244    .4184511
x2 |   .0882264   .0198852     4.44   0.000     .0480674    .1283854
x3 |  -.0000618   .0003271    -0.19   0.851    -.0007223    .0005988
x4 |   .0001995   .0002242     0.89   0.379    -.0002532    .0006522
_cons |  -1.791928   .8772392    -2.04   0.048    -3.563548   -.0203073
------------------------------------------------------------------------------

. l if !e(sample)
+-----------------------------------------------+
|  region         x1         x2    x3    x4   y |
|-----------------------------------------------|
1. | RegionA   9.412693   7.316763   655   212   . |
21. | RegionB    8.94365   10.81936   633   129   . |
41. | RegionC   9.143153   7.710033   771   120   . |
+-----------------------------------------------+