st: xtmelogit post-estimation

[Vincenzo Carrieri <vincenzo.carrieri@gmail.com>]

Dear statalisters,
I am trying to estimate a two-level logit model using the command
xtmelogit in stata 10 (It' s the first time that I use this command,
please forgive my unexperience..)
 This is the model that I tried to estimate:

xtmelogit  hpoor age male agesq age3 centro sud istr1 istr2 istr3
istr4 istr5  quintpos1 quintpos2 quintpos3 quintpos4 escluso || reg:
quintpos1 escluso

and I tried to test if is the case that the variables quintpos1 (that
is first quintile of income) and escluso (that is an indicator of
social exclusion) have a random slopes with regard to the id variable
reg (geographical area)

This is the output that I got:


Mixed-effects logistic regression               Number of obs      =     39779
Group variable: reg                             Number of groups   =        21

                                                Obs per group: min =       705
                                                               avg =    1894.2
                                                               max =      4855

Integration points =   7                        Wald chi2(16)      =   9093.30
Log likelihood = -20589.315                     Prob > chi2        =    0.0000

------------------------------------------------------------------------------
       hpoor |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |   .1448671   .0237825     6.09   0.000     .0982542    .1914799
        male |  -.1313132   .0265634    -4.94   0.000    -.1833765   -.0792499
       agesq |  -.0017283   .0004864    -3.55   0.000    -.0026816    -.000775
        age3 |    .000012   3.15e-06     3.79   0.000     5.78e-06    .0000181
      centro |   .1413327   .1091925     1.29   0.196    -.0726807    .3553461
         sud |   .1948152   .0924097     2.11   0.035     .0136955    .3759348
       istr1 |   1.152087   .0770777    14.95   0.000     1.001018    1.303157
       istr2 |   .6678377   .0474157    14.08   0.000     .5749045    .7607708
       istr3 |   .4098109   .0440482     9.30   0.000     .3234779    .4961439
       istr4 |   .1752302   .0430783     4.07   0.000     .0907982    .2596621
       istr5 |   .2834899   .0732636     3.87   0.000     .1398958     .427084
   quintpos1 |   .3710548   .0476594     7.79   0.000     .2776442    .4644654
   quintpos2 |    .289854   .0416092     6.97   0.000     .2083015    .3714065
   quintpos3 |   .3432798   .0392819     8.74   0.000     .2662886     .420271
   quintpos4 |   .2017893   .0376625     5.36   0.000     .1279722    .2756065
     escluso |   .1486454   .0200471     7.41   0.000     .1093538     .187937
       _cons |  -5.452556   .3787138   -14.40   0.000    -6.194821    -4.71029
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
reg: Independent             |
                sd(quintp~1) |   .0797235   .0618246      .0174377    .3644875
                 sd(escluso) |   .0576097   .0192887      .0298881    .1110434
                   sd(_cons) |   .1716145   .0330082      .1177157     .250192
------------------------------------------------------------------------------
LR test vs. logistic regression:     chi2(3) =    83.05   Prob > chi2 = 0.0000


I have basically 4 questions:
1) the random effects parameters that I got are the mean variation of
the fixed coefficients among regions? So basically for the variable
quintpos1: I'll have 0.37 (fixed part) + 0.079 (mean variation across
regions). Is it right?
2) there is a way to test if the random parameters are significant. I
would say yes, because confidence interval does not include zero. Is
it ok?
3) LR test at the end of the output said me that multilevel model
should be preferred to a standard logistic regression?
4) I would like to explore in which way the slopes of "quintpos1" and
"escluso" vary across regions. I executed this command:
 predict re_quintpos1 re_escluso re_cons, reffects
but they are different from random parameters estimation in the output
table. I have that the mean of re_quintpos 1 is  -.0047964.
My purpose is to plot the effect of quintpos1 in the fixed part + the
fitted residual for each region, to see if there is a systematic
pattern across some regions (for instance in all the southern
regions..). How to do this in stata?

I appreciate any hint.
Many thanks in advance

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