双重差分估计值与普通面板回归结果的不一致如何解释？

xinwei1989

7673

收藏 2015-12-30

悬赏 20 个论坛币未解决

首次接触双重差分，利用stata软件跑出来的结果是这样的：
                              DIFFERENCE IN DIFFERENCES ESTIMATION
                              -Including time-invariant covariates-

--------------------- ------------ BASE LINE --------- ----------- FOLLOW UP ---------- --------------
Outcome Variable(s) | Control |  treated  | Diff(BL) | Control |  treated  | Diff(FU) | DIFF-IN-DIFF
---------------------+---------+-----------+----------+---------+-----------+----------+--------------
lper                | -5.937  | -5.362 | 0.576 | -5.937  | -5.362 | 0.576 | 0.000
Std. Error          | 2.233 | 2.157    | 0.149 | 2.233 | 2.157    | 0.149 | .
P>|t|             | 0.008 | 0.014    | 0.000 | 0.008 | 0.014    | 0.000 | .
-------------------------------------------------------------------------------------------------

平均处理效应为 0.576 ，说明事件发生后Y是显著提高了对吗？
但是用普通面板固定效应回归结果：
Random-effects GLS regression                Number of obs    =    559
Group variable: prov                         Number of groups =       47

R-sq:  within  = 0.0621                      Obs per group: min =       11
   between = 0.0037                                     avg =    11.9
   overall = 0.0049                                     max =       12

                                             Wald chi2(5)    =    15.24
corr(u_i, X) = 0 (assumed)                   Prob > chi2       = 0.0094

                                 (Std. Err. adjusted for 47 clusters in prov)
-------------------------------------------------------------------------------
            |             Robust
      lper |    Coef. Std. Err.    z P>|z|    [95% Conf. Interval]
--------------+----------------------------------------------------------------
         t | .1200465 .0554698    2.16 0.030    .0113277 .2287654
   twheat |  -.2245029 .1088458 -2.06 0.039 -.4378368 -.011169
      wheat |  -.1117993 .299732 -0.37 0.709 -.6992632 .4756646
lfarmearnings |  -.0750534 .1522562 -0.49 0.622 -.3734701 .2233632
lpermachinery | .1181777 .1739125    0.68 0.497 -.2226846    .45904
      _cons | .0723345 1.016489    0.07 0.943 -1.919947 2.064616
--------------+----------------------------------------------------------------
   sigma_u |  1.1999658
   sigma_e |  .24129311
      rho |  .96113691 (fraction of variance due to u_i)
-------------------------------------------------------------------------------

交互项twheat却为负，那不是说明政策出台后，Y比之前还小了？且其他估计方法的值均为负，请教高手，这什么原因导致的呢？是不是说明我的数据有问题还是？有些不太明白，请指点，谢谢！