Mark Ludwig replied to Austin Nichols:
> Here's what I get when I try Austin Nichols' suggestion. The crux of the
> problem seems to be that when I try to create lags, it generates missing
> observations.
>
> sort id date
>
> .. tsset id date, monthly
> panel variable: id (weakly balanced)
> time variable: date, 2007m7 to 2008m11
> delta: 1 month
>
> ..
> .. loc v "war income male PID education L.war_casualties L.dem_iraq_email
> L.GOP_iraq_email L.dem_
>> iraq_web L.GOP_iraq_web L.media_iraq"
>
> .. su `v' if tin(2007m9, 2008m3)
[...]
You didn't answer my previous question, but here's what I get using
some toy data from Stata 9.2 (something you were also asked to do, but
didn't):
. webuse quad1
. bysort id: g time=_n
. tsset id time, monthly
panel variable: id (strongly balanced)
time variable: time, 1960m2 to 1961m9
. g x7=invnorm(uniform())
. g x8=invnorm(uniform())*-10
. g x9=invnorm(uniform())^2
. xtlogit z x7 x8 x9 l.x8 l.x9 if tin(1960m2,1961m9)
Fitting comparison model:
Iteration 0: log likelihood = -3944.2542
Iteration 1: log likelihood = -3943.45
Iteration 2: log likelihood = -3943.45
Fitting full model:
tau = 0.0 log likelihood = -3943.45
[...]
tau = 0.8 log likelihood = -3225.5103
Iteration 0: log likelihood = -3206.189
Iteration 1: log likelihood = -3203.0869
Iteration 2: log likelihood = -3203.0812
Iteration 3: log likelihood = -3203.0812
Random-effects logistic regression Number of obs = 5700
Group variable (i): id Number of groups = 300
Random effects u_i ~ Gaussian Obs per group: min = 19
avg = 19.0
max = 19
Wald chi2(5) = 3.65
Log likelihood = -3203.0812 Prob > chi2 = 0.6014
------------------------------------------------------------------------------
z | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
x7 | .0190738 .0332985 0.57 0.567 -.04619 .0843375
x8 | .0018772 .0032997 0.57 0.569 -.00459 .0083444
x9 | .0269834 .024339 1.11 0.268 -.0207202 .074687
x8 |
L1. | -.0011123 .0032993 -0.34 0.736 -.0075789 .0053542
x9 |
L1. | .0335917 .0247367 1.36 0.174 -.0148914 .0820748
_cons | .0850576 .1091967 0.78 0.436 -.1289641 .2990792
-------------+----------------------------------------------------------------
/lnsig2u | 1.028655 .1077276 .817513 1.239797
-------------+----------------------------------------------------------------
sigma_u | 1.672514 .0900879 1.504945 1.85874
rho | .4595406 .0267556 .4077355 .512235
------------------------------------------------------------------------------
Likelihood-ratio test of rho=0: chibar2(01) = 1480.74 Prob >= chibar2 = 0.000
It could well be that -xtlogit- doesn't like weakly balanced
time-series data, or (more likely) it could be because of the
particularities of your own longitudinal survey dataset.
--
Clive Nicholas
[Please DO NOT mail me personally here, but at
<[email protected]>. Please respond to contributions I make in
a list thread here. Thanks!]
"My colleagues in the social sciences talk a great deal about
methodology. I prefer to call it style." -- Freeman J. Dyson.
*
* 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/
