Hi Statalist,

Please consider the following example:

Code:
*model 1*
areg score shock##c.treatment_intensity i.round,a(id) cluster(cluster_id)
Which gives the result:
Code:
-----------------------------------------------------------------------------------------------------------
                                          |               Robust
                                      score |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
------------------------------------------+----------------------------------------------------------------
                                   1.shock |  -2.717576   1.581133    -1.72   0.102    -6.026925    .5917735
                      treatment_intensity |  -.0072408    .006894    -1.05   0.307    -.0216701    .0071884
                                          |
         shock#c.treatment_intensity |
                                       1  |   .0056869   .0012825     4.43   0.000     .0030026    .0083711
                                          |
                                    round |
                                       1  |          0  (empty)
                                       2  |  -22.65148   4.043705    -5.60   0.000    -31.11505   -14.18791
                                       3  |   8.818647   4.789771     1.84   0.081    -1.206459    18.84375
                                       4  |  -5.005754   .9063071    -5.52   0.000    -6.902677   -3.108832
                                       5  |          0  (omitted)
                                          |
                                    _cons |   50.46459   3.150213    16.02   0.000     43.87111    57.05806
------------------------------------------+----------------------------------------------------------------
                                         id |   absorbed                                    (1936 categories)

Now consider this example:
Code:
*model 2*
areg score shock##c.treatment_intensity i.round,a(cluster_id) cluster(cluster_id)
Which gives the following result:
Code:
-----------------------------------------------------------------------------------------------------------
                                          |               Robust
                                      score |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
------------------------------------------+----------------------------------------------------------------
                              1.shock |  -1.724726   1.117468    -1.54   0.139    -4.063613    .6141605
                      treatment_intensity |  -.0062692   .0058986    -1.06   0.301    -.0186151    .0060767
                                          |
         shock#c.treatment_intensity |
                                       1  |   .0020603   .0009882     2.08   0.051    -8.08e-06    .0041287
                                          |
                                    round |
                                       3  |    31.4925   3.739315     8.42   0.000     23.66603    39.31898
                                       4  |   17.50063   2.941436     5.95   0.000     11.34413    23.65712
                                       5  |   22.46466   3.457666     6.50   0.000     15.22768    29.70164
                                          |
                                    _cons |   27.67784    1.87816    14.74   0.000     23.74681    31.60888
------------------------------------------+----------------------------------------------------------------
                                cluster_id|   absorbed                                      (20 categories)
As can be seen, in model 1 , with individual fixed effects, round 5 is the reference category and is omitted.
However, in model 2, with region(cluster_id) fixed effects, round 2 becomes the reference category and gets omitted.
My dependent variable score is available from round 2 to round 5.

Can anyone please explain why this difference occurs?

Thanks,