# Re: st: RE: xtlogit

 From Tim Wade <[email protected]> To [email protected] Subject Re: st: RE: xtlogit Date Thu, 10 Mar 2005 09:42:24 -0500

```panels.  My panel sizes were very large (2000 +), but I don't really
understand how panels could be completely determined in the same way
an individual observation can be in logit.  Moreover, I get no such
error fitting the same random effects model in other programs,
(gllamm, or Proc NLMIXED in SAS, ) (see results of the two models for
xtlogit and gllamm below). Thanks very much for any insight!

. xtlogit hcgi m_l10count8 if anycontact==1, i(beachnum)

Fitting comparison model:

Iteration 0:   log likelihood = -4026.8522
Iteration 1:   log likelihood = -4020.8548
Iteration 2:   log likelihood = -4020.8325
Iteration 3:   log likelihood = -4020.8325

Fitting full model:

tau =  0.0     log likelihood = -2562.2351
tau =  0.1     log likelihood = -2556.2705
tau =  0.2     log likelihood = -2558.3731
Iteration 0:   log likelihood = -2556.2705
Iteration 1:   log likelihood = -2555.8389
Iteration 2:   log likelihood = -2555.2593
Iteration 3:   log likelihood = -2555.2165
Iteration 4:   log likelihood = -2555.2117
Iteration 5:   log likelihood = -2555.2117

Random-effects logistic regression              Number of obs      =     13927
Group variable (i): beachnum                    Number of groups   =         4

Random effects u_i ~ Gaussian                   Obs per group: min =      1305
avg =    3481.8
max =      7502

Wald chi2(1)       =      3.38
Log likelihood  = -2555.2117                    Prob > chi2        =    0.0659

------------------------------------------------------------------------------
hcgi |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
m_l10count8 |   .1326954   .0721532     1.84   0.066    -.0087223    .2741131
_cons |  -2.628946   .2267516   -11.59   0.000    -3.073371   -2.184521
-------------+----------------------------------------------------------------
/lnsig2u |  -2.390896   .9681572                     -4.288449   -.4933428
-------------+----------------------------------------------------------------
sigma_u |   .3025684   .1464669                      .1171588    .7813974
rho |   .0270738    .025502                      .0041549    .1565414
------------------------------------------------------------------------------
Likelihood-ratio test of rho=0: chibar2(01) =  2931.24 Prob >= chibar2 = 0.000

Note: 2 completely determined panels

. gllamm hcgi m_l10count8 if anycontact==1, i(beachnum)

number of level 1 units = 13927
number of level 2 units = 4

Condition Number = 5.8527904

gllamm model

log likelihood = -4012.7333

------------------------------------------------------------------------------
hcgi |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
m_l10count8 |   .1129098   .0586056     1.93   0.054     -.001955    .2277747
_cons |  -2.566529   .1207458   -21.26   0.000    -2.803186   -2.329872
------------------------------------------------------------------------------

Variances and covariances of random effects
-----------------------------------------------------------------------------

***level 2 (beachnum)

var(1): .04480799 (.018906)
-----------------------------------------------------------------------------

On Wed, 9 Mar 2005 16:15:00 -0600, Gustavo Sanchez <[email protected]> wrote:
>
> >I've run a random effects model using xtlogit (attaching output below) and
> I
> >have two questions:
>
> >1) what does the note at the end of the output "77 completely determined
> >panels" means?
> >2) How are sigma_u and rho estimated and what could be the reasons to not
> >get an estimate?
>
> >I can give more details about what I'm trying to do if necessary.
>
> >Does anyone can help me?
>
> >Thanks, Maria
>
> Please, look at the FAQ for completely determined observations for -logit-
> (it also applies to completely determined panels for -xtlogit-):
>
>        http://www.stata.com/support/faqs/stat/logitcd.html
>
> Regarding rho and sigma:
>
>        - rho is obtained as: rho = (sigma_u)^2 / ((sigma_u)^2 + (sigma_e)^2)
>
>          Where sigma_u and sigma_e correspond to the unobserved individual
>          specific component  and the idiosyncratic component of the error term:
>
>                error = u_i + e_i_t
>
>         (See page 135 of the cross-sectional time-series manual for details)
>
>        - sigma_u, is the standard deviation of the random effect term, which
>          measures the degree of heterogeneity in u_i
>
> Sincerely,
>
> Gustavo
> ([email protected])
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```