|
|  |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: xtlogit fe vs. re yields very different results, and Hausman's test doesn't help
|
From |
"avwilson" <[email protected]> |
|
To |
<[email protected]> |
|
Subject |
st: xtlogit fe vs. re yields very different results, and Hausman's test doesn't help |
|
Date |
Thu, 26 Apr 2007 10:56:27 +0300 |
Dear Statalist, I am a rank beginner with stata, and a social
anthropologist, using a panel regression to analyse 3401 year records for
226 women, recording details about their marital and reproductive status in
that year. Intellectually, I think I should use a random effect model, but
I would like some statistical justification for that choice, especially as
the significance of the independent variables in the model differs markedly
from the fixed effects to the random effects model. When I run a Hausman
test on the results I get a warning: "the rank of the differenced variance
matrix (11) does not equal the number of coefficients being tested (12); be
sure this is what you expect, or there may be problems computing the test."
I do not know how to interpret this warning. All outputs are below. Any
advice appreciated! Alexandra Wilson
. tsset PNO YEAR
panel variable: PNO (unbalanced)
time variable: YEAR, 1925 to 1995
. xtlogit BIRTH1 AGE AGESQ PARITY WFNO PREVMNO wstat2 wstat3 wstat4 wstat5
wstat
> 6 wstat7 mmrel2 tooyoung2, fe
note: wstat6 dropped due to collinearity
note: multiple positive outcomes within groups encountered.
note: 39 groups (102 obs) dropped due to all positive or
all negative outcomes.
Iteration 0: log likelihood = -1271.323
Iteration 1: log likelihood = -1009.4002
Iteration 2: log likelihood = -991.86787
Iteration 3: log likelihood = -991.71245
Iteration 4: log likelihood = -991.7124
Conditional fixed-effects logistic regression Number of obs = 3299
Group variable (i): PNO Number of groups = 187
Obs per group: min = 2
avg = 17.6
max = 39
LR chi2(12) = 943.44
Log likelihood = -991.7124 Prob > chi2 = 0.0000
---------------------------------------------------------------------------
BIRTH1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+-------------------------------------------------------------
AGE | -.7781401 .0809084 -9.62 0.000 -.9367176 -.6195625
AGESQ | -.0085769 .0011047 -7.76 0.000 -.0107421 -.0064118
PARITY | 3.912617 .1763788 22.18 0.000 3.566921 4.258313
WFNO | -.9262087 .1469608 -6.30 0.000 -1.214246 -.6381709
PREVMNO | 1.710601 .2524697 6.78 0.000 1.21577 2.205433
wstat2 | .3215757 .2058395 1.56 0.118 -.0818623 .7250138
wstat3 | .2670706 .3395775 0.79 0.432 -.3984891 .9326303
wstat4 | .6557474 .4955069 1.32 0.186 -.3154283 1.626923
wstat5 | .3298397 .8224852 0.40 0.688 -1.282202 1.941881
wstat7 | 2.92791 1.685778 1.74 0.082 -.3761536 6.231973
mmrel2 | .0674363 .4468644 0.15 0.880 -.8084018 .9432744
tooyoung2 | 1.20216 .2155146 5.58 0.000 .7797593 1.624561
---------------------------------------------------------------------------
. estimates store fe
. xtlogit BIRTH1 AGE AGESQ PARITY WFNO PREVMNO wstat2 wstat3 wstat4 wstat5
wstat
> 6 wstat7 mmrel2 tooyoung2, re
note: wstat6 dropped due to collinearity
Fitting comparison model:
Iteration 0: log likelihood = -1811.7739
Iteration 1: log likelihood = -1667.3293
Iteration 2: log likelihood = -1654.3739
Iteration 3: log likelihood = -1653.9127
Iteration 4: log likelihood = -1653.9109
Iteration 5: log likelihood = -1653.9109
Fitting full model:
tau = 0.0 log likelihood = -1653.9109
tau = 0.1 log likelihood = -1651.5408
tau = 0.2 log likelihood = -1657.0675
Iteration 0: log likelihood = -1651.5408
Iteration 1: log likelihood = -1641.4117
Iteration 2: log likelihood = -1638.0263
Iteration 3: log likelihood = -1637.9463
Iteration 4: log likelihood = -1637.946
Iteration 5: log likelihood = -1637.946
Random-effects logistic regression Number of obs = 3401
Group variable (i): PNO Number of groups = 226
Random effects u_i ~ Gaussian Obs per group: min = 1
avg = 15.0
max = 39
Wald chi2(12) = 278.02
Log likelihood = -1637.946 Prob > chi2 = 0.0000
---------------------------------------------------------------------------
BIRTH1 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+-------------------------------------------------------------
AGE | .0734747 .0624513 1.18 0.239 -.0489277 .1958771
AGESQ | -.0070224 .0009241 -7.60 0.000 -.0088337 -.0052112
PARITY | 1.130243 .0771888 14.64 0.000 .9789562 1.281531
WFNO | -.040851 .0820072 -0.50 0.618 -.2015821 .11988
PREVMNO | .3706266 .1405351 2.64 0.008 .0951827 .6460704
wstat2 | -.0870879 .1483159 -0.59 0.557 -.3777817 .2036059
wstat3 | -.044998 .1741492 -0.26 0.796 -.3863241 .2963281
wstat4 | -.1622861 .2776362 -0.58 0.559 -.706443 .3818708
wstat5 | .1411163 .5946235 0.24 0.812 -1.024324 1.306557
wstat7 | -.3090833 1.16006 -0.27 0.790 -2.58276 1.964593
mmrel2 | .2974777 .1392273 2.14 0.033 .0245972 .5703583
tooyoung2 | .7633049 .1909386 4.00 0.000 .3890722 1.137538
_cons | -2.328862 .9760806 -2.39 0.017 -4.241945 -.4157793
-------------+-------------------------------------------------------------
/lnsig2u | -.6380874 .1901891 -1.010851 -.2653236
-------------+-------------------------------------------------------------
sigma_u | .7268438 .0691189 .6032487 .8757612
rho | .1383652 .0226744 .099598 .1890537
---------------------------------------------------------------------------
Likelihood-ratio test of rho=0: chibar2(01) = 31.93 Prob >= chibar2 = 0.000
. estimates store re
. hausman fe re, eq(1:1)
Note: the rank of the differenced variance matrix (11) does not equal the
number of coefficients being tested (12); be sure this is what you expect,
or there may be problems computing the test. Examine the output of your
estimators for anything unexpected and possibly consider scaling your
variables so that the coefficients are on a similar scale.
---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+-------------------------------------------------------------
AGE | -.7781401 .0734747 -.8516148 .0514393
AGESQ | -.0085769 -.0070224 -.0015545 .0006052
PARITY | 3.912617 1.130243 2.782373 .1585919
WFNO | -.9262087 -.040851 -.8853576 .121952
PREVMNO | 1.710601 .3706266 1.339975 .2097399
wstat2 | .3215757 -.0870879 .4086636 .1427316
wstat3 | .2670706 -.044998 .3120686 .2915218
wstat4 | .6557474 -.1622861 .8180335 .4104208
wstat5 | .3298397 .1411163 .1887234 .5682471
wstat7 | 2.92791 -.3090833 3.236993 1.223154
mmrel2 | .0674363 .2974777 -.2300414 .4246216
tooyoung2 | 1.20216 .7633049 .4388553 .099945
---------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtlogit
B = inconsistent under Ha, efficient under Ho; obtained from xtlogit
Test: Ho: difference in coefficients not systematic
chi2(11) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 339.30
Prob>chi2 = 0.0000
*
* 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/
