standard error = 0

[lisa carnicelli]

I am doing DID. when I run the normal regression without clustering I get standard errors of 1.2 or 2.3, but when I use cluster the standard error becomes 0. is that normal to have such a low standard error ?

[Steve Samuels]

Twice you've been asked to read FAQ 12 about how to ask good questions. Please do so. Without seeing all your code and results, it's impossible to judge what has happened. As the FAQ requests, put everything between code delimiters [CODE] and [/CODE]

[Steve Samuels]

I apologize for the tone of my statement above.

To answer your question: a zero standard error for a cluster-based standard error is not common. Most of the time a cluster-based standard error will be larger than the standard error that ignores the clustering.. Still the opposite can happen. Here's a simple example of a very small standard error with two clusters of three points each. The points in each lie on a line and the two lines are parallel.

. clear

. input group x y

  1.      1   3   9
  2.      1   2   8
  3.      1   1   7
  4.      2   0   4
  5.      2   2   6
  6.      2   4   8
  7. end

. scatter y x  




. . regress y x

      Source |       SS           df       MS      Number of obs   =         6
-------------+----------------------------------   F(1, 4)         =      6.67
       Model |          10         1          10   Prob > F        =    0.0612
    Residual |           6         4         1.5   R-squared       =    0.6250
-------------+----------------------------------   Adj R-squared   =    0.5312
       Total |          16         5         3.2   Root MSE        =    1.2247

------------------------------------------------------------------------------
           y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           x |          1   .3872983     2.58   0.061    -.0753126    2.075313
       _cons |          5   .9219544     5.42   0.006     2.440244    7.559756
------------------------------------------------------------------------------

.
. . regress y x, vce(cluster group)

Linear regression                               Number of obs     =          6
                                                F(1, 1)           >   99999.00
                                                Prob > F          =     0.0000
                                                R-squared         =     0.6250
                                                Root MSE          =     1.2247

                                  (Std. Err. adjusted for 2 clusters in group)
------------------------------------------------------------------------------
             |               Robust
           y |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
           x |          1   1.88e-08  5.3e+07   0.000     .9999998           1
       _cons |          5   1.118034     4.47   0.140    -9.205969    19.20597
------------------------------------------------------------------------------

.