Hello statalisters:
I have a dataset with a binary outcome variable (hcgi) and a "group"
cluster I would like to model with random effects. The main predictor,
x1 is continuous. There are four clusters and they range in size from
about 2,000 to over 10,000 observations. When I fit an xtlogit model,
I get the response that "2 completely determined panels" or
occasionally "4 completely determined panels" when I fit slightly
differnt models. I read the FAQ about this which says it means the
same as logistic regression when observations are completely
determined. While I understand the message in a regular logistic
model, I don't understand how a panel could be completely determined,
or what I can do to avoid this. Any clarifications would be greatly
appreciated!
. xi: xtlogit hcgi x1 gicontact_any i.agecat if anycontact==1, i(group)
i.agecat1 _Iagecat1_1-5 (naturally coded; _Iagecat1_1 omitted)
Fitting comparison model:
Iteration 0: log likelihood = -3937.4203
Iteration 1: log likelihood = -3863.6323
Iteration 2: log likelihood = -3846.2755
Iteration 3: log likelihood = -3846.1091
Iteration 4: log likelihood = -3846.1091
Fitting full model:
tau = 0.0 log likelihood = -2522.9427
tau = 0.1 log likelihood = -2516.1209
tau = 0.2 log likelihood = -2517.9593
Iteration 0: log likelihood = -2516.1209
Iteration 1: log likelihood = -2512.8231
Iteration 2: log likelihood = -2512.8038
Iteration 3: log likelihood = -2512.8033
Iteration 4: log likelihood = -2512.8033
Random-effects logistic regression Number of obs = 13535
Group variable (i): beachnum Number of groups = 4
Random effects u_i ~ Gaussian Obs per group: min = 1294
avg = 3383.8
max = 7239
Wald chi2(6) = 80.85
Log likelihood = -2512.8033 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
hcgi | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
x1 | .1349034 .0622243 2.17 0.030
.012946 .2568608
gicontact_~y | 1.342348 .1607566 8.35 0.000 1.027271 1.657426
_Iagecat1_2 | .0855625 .224333 0.38 0.703 -.3541221 .5252471
_Iagecat1_3 | -.2844456 .2464818 -1.15 0.248 -.767541 .1986498
_Iagecat1_4 | .0568205 .2148344 0.26 0.791 -.3642472 .4778882
_Iagecat1_5 | -.2423311 .2175506 -1.11 0.265 -.6687224 .1840601
_cons | -2.683717 .2941536 -9.12 0.000 -3.260248 -2.107187
-------------+----------------------------------------------------------------
/lnsig2u | -2.458037 1.163423 -4.738303 -.1777706
-------------+----------------------------------------------------------------
sigma_u | .2925796 .1701969 .0935601 .9149505
rho | .0253603 .0287565 .0026537 .2028432
------------------------------------------------------------------------------
Likelihood-ratio test of rho=0: chibar2(01) = 2666.61 Prob >= chibar2 = 0.000
Note: 2 completely determined panels
Tim Wade
*
* 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/
