hey, boostrap is used when your estimates are generated indirectly, for example, if you have used iv. So you simple need to add ",vce(boot)" after the regression command. That will do.

这个问题比较有意思，我也是一知半解

欢迎讨论。

我正在看呢，哪位高手指点一二，听说bootstrap能修正估计的偏差，下面是我估计的结果，用的是最简单的命令

(running xtreg on estimation sample)

Bootstrap replications (50)

----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5

.................................................. 50

Fixed-effects (within) regression Number of obs = 232

Group variable (i): n Number of groups = 29

R-sq: within = 0.1165 Obs per group: min = 8

between = 0.0087 avg = 8.0

overall = 0.0070 max = 8

Wald chi2(8) = 21.60

corr(u_i, Xb) = -0.5687 Prob > chi2 = 0.0057

------------------------------------------------------------------------------

| Bootstrap

y1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]

-------------+----------------------------------------------------------------

x11 | .0299174 .037529 0.80 0.425 -.0436382 .103473

x2 | -.001709 .0035098 -0.49 0.626 -.008588 .00517

x3 | -.1030626 .0728897 -1.41 0.157 -.2459237 .0397985

x5 | .0120006 .0037163 3.23 0.001 .0047168 .0192845

x6 | -.0006948 .0007497 -0.93 0.354 -.0021642 .0007745

x7 | -.0002269 .0002382 -0.95 0.341 -.0006938 .0002399

x9 | -.0040276 .0019661 -2.05 0.041 -.0078812 -.0001741

x10 | -.0019916 .0009764 -2.04 0.041 -.0039054 -.0000779

_cons | -.039287 .0561457 -0.70 0.484 -.1493306 .0707566

-------------+----------------------------------------------------------------

sigma_u | .02322199

sigma_e | .0228256

rho | .50860756 (fraction of variance due to u_i)

------------------------------------------------------------------------------

而采用的面版数据固定效应模型中的估计结果是

Fixed-effects (within) regression Number of obs = 232

Group variable (i): n Number of groups = 29

R-sq: within = 0.1165 Obs per group: min = 8

between = 0.0087 avg = 8.0

overall = 0.0070 max = 8

F(8,195) = 3.22

corr(u_i, Xb) = -0.5687 Prob > F = 0.0019

------------------------------------------------------------------------------

y1 | Coef. Std. Err. t P>|t| [95% Conf. Interval]

-------------+----------------------------------------------------------------

x11 | .0299174 .0294872 1.01 0.312 -.0282374 .0880722

x2 | -.001709 .0023167 -0.74 0.462 -.006278 .00286

x3 | -.1030626 .0593385 -1.74 0.084 -.2200903 .0139651

x5 | .0120006 .0032738 3.67 0.000 .005544 .0184572

x6 | -.0006948 .000594 -1.17 0.244 -.0018663 .0004767

x7 | -.0002269 .0002763 -0.82 0.412 -.0007718 .0003179

x9 | -.0040276 .0015183 -2.65 0.009 -.0070221 -.0010331

x10 | -.0019916 .0009712 -2.05 0.042 -.0039071 -.0000762

_cons | -.039287 .0534978 -0.73 0.464 -.1447955 .0662215

-------------+----------------------------------------------------------------

sigma_u | .02322199

sigma_e | .0228256

rho | .50860756 (fraction of variance due to u_i)

------------------------------------------------------------------------------

F test that all u_i=0: F(28, 195) = 3.48 Prob > F = 0.0000

1。上面两个结果的系数估计值是一样的，只是标准误差不同，这是什么意思呢？是不是表明没必要用Bootstrap方法啊？

2．如果增加选择项，应该怎么增加呢？bca、mse

3．“exp_list”什么意思？如何运用？

请高手指点一下

bootstrap可以估计置信区间，是通过重复抽样的方法进行。一般来说，对小样本用bootstrap simulation一般可以达到比较好的效果。我也是一知半解，不一定正确。请高手指点。