st: xtmelogit: interpretation of the _cons and level 1 predictors

[Joao Carapinha <joao@carapinha.com>]

Hi List Members,

I'm applying xtmelogit to understand differences within and between
level 1 and level 2 data. The command I'm using is:

xtmelogit s_QA16Arecode c_wealthindex || QHEA: if chronic1==1, variance mle

and the output is (my question follows below this):

************************************************
Refining starting values:
Iteration 0: log likelihood = -1347.4931
Iteration 1: log likelihood = -1331.2782
Iteration 2: log likelihood = -1330.7283

Performing gradient-based optimization:
Iteration 0: log likelihood = -1330.7283
Iteration 1: log likelihood = -1330.7274
Iteration 2: log likelihood = -1330.7274

Mixed-effects logistic regression Number of obs = 1951

Group variable: QHEA Number of groups = 408

Obs per group: min = 1
avg = 4.8
max = 23
Integration points = 7                    Wald chi2(1) = 31.66
Log likelihood = -1330.7274           Prob > chi2 = 0.0000

s_QA16Arecode      Coef.        Std. Err.   z        P>z    [95% Conf. Interval]
c_wealthindex      .2909747 .0517103    5.63    0.000    .1896244 .3923251
_cons                   .0360156 .0525991   0.68    0.494    -.0670768 .1391079

Random-effects Parameters   Estimate   Std. Err.   [95% Conf. Interval]
QHEA: Identity
var(_cons)                             .1628087 .0840827 .0591662 .4480037

LR test vs. logistic regression: chibar2(01) = 5.73 Prob>=chibar2 = 0.0083
******************************************************************

Given that the p-value of _cons suggests that it is not a
statistically significant estimation of the intercept and that the
level 1 predictor (c_wealthindex) is statistically significant, would
a fair interpretation be:

The mean estimate of all groups for the outcome is not significantly
different from zero suggesting that there are no groups that are
either significantly below or above the mean. However, the level 1
predictor is significantly different from zero and it is a great
estimate for the differences between individuals within each group.

- Is it possible to have this situation?
- What would explain this?

Many thanks for your kind assistance,

JC

*
*   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/