st: CI for ri variance after melogit

["Bontempo, Daniel E" <deb193@ku.edu>]

I use nlcom to retrieve the estimated level-2 variance after an empty xtmelogit model with a random intercept, and the 95% CI is different form what was printed in the random effects table.

For example:


------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Subj: Identity               |
                  var(_cons) |    .779343   .2679137      .3972942     1.52878
------------------------------------------------------------------------------
LR test vs. logistic regression: chibar2(01) =    50.65 Prob>=chibar2 = 0.0000

. matlist e(b)

             | eq1       | lns1_1_1  
             |     _cons |     _cons 
-------------+-----------+-----------
          y1 | -2.695873 |  -.124652 

. nlcom( Subj_var: (exp([lns1_1_1]_b[_cons]))^2)

    Subj_var:  (exp([lns1_1_1]_b[_cons]))^2

------------------------------------------------------------------------------
     Correct |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    Subj_var |    .779343   .2679137     2.91   0.004     .2542419    1.304444
------------------------------------------------------------------------------



Can someone explain why the nlcom CI gets smaller?

Thanks


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/