看看一样不一样啊

郁闷啊，我的结果确实不同，版主请看。
另外，注意我分别使用robust和cluster时，F统计值是不一样的，而上面版主的结果里F统计值一直没有变化，不只是否与此有关。

*************stata10.0的结果

. webuse nlswork
(National Longitudinal Survey.  Young Women 14-26 years of age in 1968)

. xtreg ln_wage hours,fe

Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,23756)         =      3.14
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.0764

------------------------------------------------------------------------------
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0002525     1.77   0.076    -.0000475    .0009423
       _cons |   1.658941   .0094249   176.02   0.000     1.640468    1.677415
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(4709, 23756) =     8.30         Prob > F = 0.0000

. xtreg ln_wage hours,fe robust

Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,23756)         =      1.20
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.2724

                                 (Std. Err. adjusted for clustering on idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0004076     1.10   0.272    -.0003516    .0012464
       _cons |   1.658941   .0152638   108.68   0.000     1.629023    1.688859
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------

. xtreg ln_wage hours,fe vce(cluster  idcode)

Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,4709)          =      0.81
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.3696

                              (Std. Err. adjusted for 4710 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0004986     0.90   0.370    -.0005301     .001425
       _cons |   1.658941   .0182299    91.00   0.000     1.623202     1.69468
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------

***************stata12的结果


. webuse nlswork
(National Longitudinal Survey.  Young Women 14-26 years of age in 1968)

. xtreg ln_wage hours,fe

Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,23756)         =      3.14
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.0764

------------------------------------------------------------------------------
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0002525     1.77   0.076    -.0000475    .0009423
       _cons |   1.658941   .0094249   176.02   0.000     1.640468    1.677415
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0:     F(4709, 23756) =     8.30         Prob > F = 0.0000

. xtreg ln_wage hours,fe robust
Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,4709)          =      0.81
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.3696

                              (Std. Err. adjusted for 4710 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0004986     0.90   0.370    -.0005301     .001425
       _cons |   1.658941   .0182299    91.00   0.000     1.623202     1.69468
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------


. xtreg ln_wage hours,fe vce(cluster  idcode)

Fixed-effects (within) regression               Number of obs      =     28467
Group variable: idcode                          Number of groups   =      4710

R-sq:  within  = 0.0001                         Obs per group: min =         1
       between = 0.0314                                        avg =       6.0
       overall = 0.0074                                        max =        15

                                                F(1,4709)          =      0.81
corr(u_i, Xb)  = 0.0976                         Prob > F           =    0.3696

                              (Std. Err. adjusted for 4710 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       hours |   .0004474   .0004986     0.90   0.370    -.0005301     .001425
       _cons |   1.658941   .0182299    91.00   0.000     1.623202     1.69468
-------------+----------------------------------------------------------------
     sigma_u |   .4229084
     sigma_e |  .32040339
         rho |  .63532952   (fraction of variance due to u_i)
------------------------------------------------------------------------------

.

update 25feb2008 -------------------------------------------------------------------------------------------------        -

Ado-files

1.  bootstrap has the new jackknifeopts() option for passing options to jackknife when the bca option is specified.

2.  The documentation for glm and binreg did not mention that the constraints() option was allowed when fitting models
using ml optimization.  The help files and dialogs have been updated.

3.  glm, when used with the binomial family and either the log or identity link, sometimes issued a spurious warning
message regarding inadmissible predictions.  This has been fixed.

4.  When using a saved scatterplot matrix that was created using the by() option and that had previously been edited
with the Graph Editor, Observation Property edits might be ignored for some observations.  This has been fixed.

5.  The Y axis tab has been removed from the dialog box for graph pie.

6.  kdensity displayed a bandwidth of zero on the graph even though the bandwidth was not exactly zero.  kdensity now
displays more significant digits for the bandwidth in the resulting graph.

7.  notes did not attach notes to a variable name with exactly 32 characters.  This has been fixed.

8.  predict with the scores option after mlogit was displaying an error when specifying the correct number of
variables.  This has been fixed.

9.  predict with the scores option after ologit and oprobit produced incorrect equation-level score values when the
estimation results were computed under version control less than 9.  This affected the standard errors produced by
svy: ologit and svy: oprobit when they were called under version control less than 9.  This has been fixed.

10.  ssc whatsnew and ssc whatshot have been renamed ssc new and ssc hot.  The old subcommand names continue to work
but are undocumented.

11.  The following changes are made to sts graph and stcurve for kernel-based hazard estimation (when option hazard is
used).

A.  The use of the boundary-adjusted kernel functions epan2, biweight, or rectangular can sometimes lead to
negative estimates of the hazard function.  In such cases, the negative hazard estimates are replaced with 0.

B.  When a hazard estimation grid included a time point equal to the value of a bandwidth, the hazard estimates
obtained using an epan2, biweight, or rectangular kernel were incorrectly shifted by one time point in the
right boundary region.  This could result in a slightly different curve being displayed in the right boundary
region with these kernels.  This has been fixed.

C.  The left boundary region is changed to be the region [L,L+h) instead of the old region [0,h).  Here L is the
minimum analysis time at which failure occurred, and h is the bandwidth.  As such, the default plotting range
is restricted to [L,R] (R is the observed maximum failure time) for the epan2, biweight, and rectangular
kernels.  For other available kernels, the default plotting range is [L+h,R-h].

12.  svy results will now report the population and subpopulation sizes out to a larger number of significant digits,
reserving scientific notation for sizes greater than 99 trillion.

13.  svy: tabulate twoway, when used with data that contained a stratum with a single sampling unit, would report
zero-valued test statistics (with p-values of 1) when it should have reported missing values.  This has been
fixed.

14.  svy: tabulate twoway now recognizes value labels assigned to extended missing values.

15.  svyset would not check for invalid syntax beyond the stage that contained an fpc() option.  This has been fixed.

16.  The dialog box for xtabond did not allow vce(robust) to be specified with the twostep option.  This has been
fixed.

17.  xtgee terminated with error 198, Unable to identify sample, if pweights were used with the noconstant option.
This has been fixed.

18.  xtgee did not properly mark the estimation sample (contained in e(sample)) when using the autoregressive,
stationary, and nonstationary correlation structures, and observations were dropped because they did not have
equally spaced time intervals.  This has been fixed.

19.  xtmixed, xtmelogit, and xtmepoisson now support time-series operators.

20.  xtreg, fe now uses vce(cluster id) when vce(robust) is specified, in light of the new results in Stock and Watson,
"Heteroskedasticity-robust standard errors for fixed-effects panel-data regression," Econometrica 76 (2008):
155-174.

stata10都升级了，你看最后一条
升级以后两个就是一样的了。

所以结果是否一样，你必须知道软件后面是怎么算的。
同时，你的看手册对更新的命令的介绍和解释

我的问题是xtreg y x1 x2,fe robust 在stata10和stata12里面的结果是不一样的，主要是标准误的算法

恩，我也发现版主说的STATA12现在只要是ROBUST就默认是cluster panelid这个做法了，但是stata12里面能不能做出之前stata10中robust的结果呢，那个是没有cluster on any group的，这两种差别还蛮大的，而且我看计量教材上面也说过ROBUST的基本原理，似乎没有说要CLUSTER ON GROUP啊

版主的意思是stata12或者升级之后的stata10以上版本基于2008年的这篇文章，认为robust检验应该是要cluster on group的对不。看来之前所有的教科书都老套了，难怪在书上看到的ROBUST都还是NON CLUSTER的

stata10的robust应该是有问题的
否则没有必要stata升级到stata10.1
而且人家还提供了文章。

软件需要升级，最初的命令或者选项可能并不是正确的结果

你的stata10没有升级

应该是采用升级后的更加可靠

这个已经试验过了，在两个版本上结果是一样的。
stata11.2和stata12

请问这里有连老师？能否告诉他的号啊

哇  来看看  太厉害了都