Hi Maarteen,
Thank you for responding. I think I have several follow up questions.
First, I did wonder if using OR was causing my problem but decided it
must be something else because the examples in the Stata manual show a
constant when using OR. I agree this is an important statistic. Please
correct me if I am wrong, but when I report my results this constant
would represent the grand mean of model, correct? I guess one way
around this is to run it without OR and then report both statistics with
an asterisk explaining the grand mean is not generated when using odds
ratios. Would this be acceptable?
I have been hesitant to use laplace because of the potential bias. As
I noted earlier I have over 56,000 observations so each run can take
several days even using the difficult command. If I can't report the
difference in the models using laplace and not using laplace then I
don't suspect I should use laplace. Would that be a safe assumption?
Here is example model not using diff: ....as you can see there are so
many valleys that I wasn't getting anywhere.
xtmelogit dumlib dumfem dumhighi dumcolle dumestee dummarri divdum
prosperi nedum sed
> um mwdum southdum pacificd || citycode: || uniqresp:, or
Refining starting values:
Iteration 0: log likelihood = -33824.627 (not concave)
Iteration 1: log likelihood = -33790.565
Iteration 2: log likelihood = -33778.46
Performing gradient-based optimization:
Iteration 0: log likelihood = -33778.46
Iteration 1: log likelihood = -33765.905
Iteration 2: log likelihood = -33764.782
Iteration 3: log likelihood = -33764.634
Iteration 4: log likelihood = -33764.633 (not concave)
Iteration 5: log likelihood = -33764.633 (not concave)
Iteration 6: log likelihood = -33764.633 (not concave)
Iteration 7: log likelihood = -33764.633 (not concave)
Iteration 8: log likelihood = -33764.633 (not concave)
Iteration 9: log likelihood = -33764.633 (not concave)
Iteration 10: log likelihood = -33764.633 (not concave)
Iteration 11: log likelihood = -33764.633 (not concave)
Iteration 12: log likelihood = -33764.633 (not concave)
Iteration 13: log likelihood = -33764.633 (not concave)
Iteration 14: log likelihood = -33764.633 (not concave)
Iteration 15: log likelihood = -33764.633 (not concave)
Iteration 16: log likelihood = -33764.633 (not concave)
Iteration 17: log likelihood = -33764.633 (not concave)
Iteration 18: log likelihood = -33764.633 (not concave)
Iteration 19: log likelihood = -33764.633 (not concave)
--Break--
r(1);
My final model seems to be too complex for my computer but I believe it
is the best way to model my question to understand the differences in
regional influences (where i.regroup is the different regions). xi:
xtmelogit dumlib dumfem dumhighi dumcolle ageraw dummarri dumestee
i.regroup dumcanad || citycode: i.regroup || uniqresp: , diff or
cov(unstr) var
This model has been running for six days. It has not generated an
error but it has not generated initial values either. Do you have any
suggestions for me? Should I be patient or should I return to the
simpler model given the volume of data?
Thank you very much for any light you can shed on these concerns. I am
looking forward to reporting results
Susan
>>> Maarten buis <[email protected]> 1/26/2008 3:05 AM >>>
--- Susan Mason <[email protected]> wrote:
> For some reason I do get a constant in the random coefficients when
I
> run my HLM model.
<snip>
> xi: xtmelogit dumlib dumfem dumhighi dumcolle ageraw dummarri
> dumestee divdum prosperi i.regroup dumcanad || citycode: ||
> uniqresp:, diff or
Thank you for this question, now I can repeat my favourite feature
request ;-)
The reason that you don't get the constant is that you specified the
-or- option in order to display the odds ratios. With the -or- option
Stata never displays the constant. The constant in this constant is
not
an odds ratio, but the baseline odds, so the column label would be
off,
but that could just be fixed with a star after _cons and a footnote,
or
some other labeling of the column.
Substantively, I think that the baseline odds are very important. The
odds ratio will tell you that the odds of success for men may be twice
that of women, but that tells a very different story if the odds of
success for women is .0001 success for every failure of 1 success for
every failure.
For many programs there is a trick to get the baseline odds: Generate
a
variable that is always one, add that variable to your model and
choose
the -noconst- option. However, there is, for good reason, no -noconst-
option in -xtmelogit-, so this trick won't work. You can display the
baseline odds separately by typing -nlcom exp([eq1]_cons)- after you
have estimated you -xtmelogit- command.
Finally if you are in a model building stage and want to have quick
results to see if something has to be changed to your model you can
specify the -laplace- option, which will be quicker.
Hope this helps,
Maarten
*-------------- begin example ----------------
webuse bangladesh
xtmelogit c_use urban age child* || district:
// replay last estimation with -or-
// notice no constant
xtmelogit , or
// the baseline odds
nlcom exp([eq1]_cons)
*-------------- end example ------------------
(For more on how to use examples I sent to the Statalist, see
http://home.fsw.vu.nl/m.buis/stata/exampleFAQ.html )
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
__________________________________________________________
Sent from Yahoo! Mail - a smarter inbox http://uk.mail.yahoo.com
*
* 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/
