st: Fwd: xtmelogit predict generates missing values

[Rachel Cole <rachel.cole@nyu.edu>]

Hi there,
I'm using xtmelogit and predict (reffects, mu, and deviance) with a
dataset with no missing values.  In a couple of instances when I use
predict, Stata informs me that it is generating missing values--an
example is below.  In this example I have 2248 observations nested in
135 schools (schid) and bsttrn is an indicator variable that equals 1
for 14%.
xtmelogit bsttrn || schid:
Refining starting values:

Iteration 0:   log likelihood = -575.44982
Iteration 1:   log likelihood =  -476.2408
Iteration 2:   log likelihood = -463.97972

Performing gradient-based optimization:

Iteration 0:   log likelihood = -463.97972
Iteration 1:   log likelihood = -450.95784
Iteration 2:   log likelihood = -446.10531
Iteration 3:   log likelihood =  -446.0206
Iteration 4:   log likelihood = -446.01967
Iteration 5:   log likelihood = -446.01966

Mixed-effects logistic regression               Number of obs      =      2248
Group variable: schid                           Number of groups   =       136

                                                Obs per group: min =         1
                                                               avg =      16.5
                                                               max =        40

Integration points =   7                        Wald chi2(0)       =         .
Log likelihood = -446.01966                     Prob > chi2        =         .

------------------------------------------------------------------------------
      bsttrn |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |  -7.026729   1.224348    -5.74   0.000    -9.426407   -4.627051
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
schid: Identity              |
                   sd(_cons) |   4.311155   .6488151      3.209893    5.790243
------------------------------------------------------------------------------
LR test vs. logistic regression: chibar2(01) =   962.05 Prob>=chibar2 = 0.0000

. predict re1_cons, reffects
(557 missing values generated)
Predictions are only made for schools with 0% bsttrn or a very few
(less than 5%).  No predictions are made for schools with some
variation in bsttrn or 100% bsttrn.

So, besides missing data, under what circumstances would xtmelogit's
predict produce missing values? In this example, what causes the
missing values for schools with variation in bsttrn?

Thanks for any help you can offer!
Rachel
*
*   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/