Oops! Misfire. Sorry about that. Below is the full response.
-----Original Message-----
Bobby and Gary -
Thank you both for your responses to my query. I am just back from some
travels and tackling the data again. After some experimentation, the best
work around to the "initial values not feasible problem appears to be to run
a simple logit to generate start values. It looks like this:
logit suboanw female
mat a = e(b)
xtmelogit suboanw female || country: female, or var from(a)
I have one instance where this approach has failed repeatedly and the final
option that Bobby suggested adding: refineopts(iterate(0)) option to the
xtmelogit. No luck there either, despite several tries.
About the data, my outcome is a binary (0,1) variable representing business
start-up. The extra country values/labels. The data comes from a larger
combined dataset of cross-sectional panels where the countries vary across
years. The country samples range in size from 1500 to 16000 obs. This is
basically what the data looks like for the simplest model and the year in
question:
---------------------------------------------
country of |
origin | mean(suboanw) mean(female)
---------------+-----------------------------
united states | .074097 .466376
russia |
south africa | .04159 .506809
greece | .040339 .5
netherlands | .031722 .541918
belgium | .021397 .555813
france | .047433 .51462
spain | .019844 .488231
hungary | .027218 .506893
italy | .023165 .543777
40 |
switzerland |
43 |
united kingdom | .027073 .598141
denmark | .025733 .517756
sweden | .015271 .48023
norway | .039281 .50188
poland | .03948 .503248
germany | .042635 .541674
peru | .315451 .512876
mexico |
argentina | .103866 .49798
brazil | .049 .512
chile |
57 |
58 |
60 |
australia | .076825 .619048
62 |
63 |
new zealand | .077515 .598817
singapore | .031153 .496625
thailand |
japan | .004173 .497131
korea |
china |
90 |
india |
canada | .055488 .582927
uganda | .157107 .526185
portugal | .020877 .52714
352 |
ireland | .040727 .587719
iceland | .075 .502222
357 |
finland | .026241 .511694
370 |
371 |
372 |
381 |
croatia | .027184 .500971
slovenia | .018463 .50268
420 |
421 |
ecuador | .167164 .534826
598 |
701 |
787 |
852 | .014189 .552745
876 |
jordan | .110608 .427853
971 |
israel | .04314 .500899
Thanks for any help you can give!
Amanda Elam <[email protected]> asks:
> Does anyone have any idea what this error message means? Is there a
> work-around?
> . xtmelogit suboanw female || country: female, or variance
> Refining starting values:
> initial values not feasible
> r(1400);
The error message means that -xtmelogit-'s automatic choice of starting
values
produced values that choked the optimizer. Although rare, it has been known
to happen.
One solution would be to fit a simpler -xtmelogit- model
. xtmelogit suboanw female || country:
and then use these estimates as starting values for the model you are
interested in
. mat b = e(b)
. xtmelogit suboanw female || country: female, or variance from(b)
-from(b)- sets the starting values as those from the simpler model, and
-xtmelogit- is smart enough to fill in a trailing zero as the starting value
for the log of the new variance component.
If that doesn't work, you could also try adding a further option to
eliminate
the initial rough optimization process
. xtmelogit suboanw female || country: female, or variance from(b)
refineopts(iterate(0))
(all on one line)
If that doesn't do the trick, then it would probably be best if we could get
a
look at these particular data. Amanda can feel free to email me directly or
[email protected].
--Bobby
[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/
--
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.5.516 / Virus Database: 269.21.4/1310 - Release Date: 3/4/2008
8:35 AM
*
* 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/
