Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Overflow at level 1 in GLLAMM


From   Xiaoheng Zhang <zhangx@tcd.ie>
To   statalist@hsphsun2.harvard.edu
Subject   RE: st: Overflow at level 1 in GLLAMM
Date   Thu, 23 Nov 2006 15:38:51 +0000

It seems that the problem of overflow is solved by following Anders
Alexandersson <andersalex@gmail.com>'s suggestion of
rescale independent variables (refer to
http://www.stata.com/support/faqs/stat/mfx_scale.html).I rescaled all nine
independent variables and put them in the regression for all 12 alternative
locations.Then the size of the dataset is-173 firms-12 locations-2076
observations-9 random effects.But it took more than 48 hours to run on my
computer(P4 3.00GHz,504 RAM) without anything comes out.Then I followed Cameron
Gillies <cgillies@ualberta.ca>'s suggestion to reduce the number of the random
effects.I set only three of nine variables as random effects but the rest
variables are fixed effects.The three random variables are tax
rate,national consumption of medicines and national pharmaceutical industry
employment.This time it took 1 hour and 41 minutes to calculate all results
that
I need-fixed effects,random effects and their variances.The number of integral
points is 4.Now I am tring integral points 8 on my computer.Cheers!

Thanks for Anders and Cameron's help as well as Nick's comments.

Greetings,

Xiaoheng Zhang(Kevin)


 Cameron Gillies <cgillies@ualberta.ca>:

> Hi Kevin,
> 
> I think your best bet will be to drop some of the random effects, but you
> could keep the fixed effects.  Graphing the data may help decide which to
> drop, but I have also found, with a standard -logit-, that comparing the
> standard errors with and without the -cluster()- option seems to be a good
> place to start.  The variables with an increase in the SE with -cluster-
> seem to be the ones where a random effect most improves model fit.
> 
> Hope that helps,
> Cam Gillies
> 
> 
> Thanks.I will try to kick some of them out so at least I can test if my
> model is
> feasible.
> 
> Kevin
> 
> 
> ÒýÓÃ Nick Cox <n.j.cox@durham.ac.uk>:
> 
> > The simple answer is that this won't help. You
> > are fitting a very difficult model and changing the memory
> > won't make it less difficult. You might get results
> > faster, at best, but the answer will be the same.
> > 
> > Nick 
> > n.j.cox@durham.ac.uk
> > 
> > Xiaoheng Zhang
> >  
> > > Dear Anders,
> > > 
> > > In my sample there are 12 countries as an alternative set and
> > > total 2076
> > > observations(173 firms*12 countries).Frankly I am reluctant
> > > to remove any one of
> > > them from the alternative set.Could I set a large memory for
> > > Stata to tackle
> > > this problem.Currently the memory allocated to Stata is 100mb
> > > and the dataset is
> > > far smaller than 100mb.
> >  
> > 
> > Anders Alexandersson <andersalex@gmail.com>:
> >  
> > > > Kevin (Xiaoheng Zhang <zhangx@tcd.ie>) wonders why overflow
> > > can happen
> > > > when using gllamm for a location choice problem:
> > > > 
> > > > > I am running a multilevel regression model by -GLLAMM-.The initial
> > > > > results are good but then the program reports "overflow
> > > at level 1 (173
> > > > > missing values)".It keeps reporting this msg for a long
> > > time.Anyone has a >
> > > > hint about this? [...] For you information,my command are
> > > > > -gllamm location eatr cdrug2 dist ap sisters2 wage3 edu3
> > > cai share if
> > > > > highgrow==1&deudmy==0, nocons expand(id choice o) i(parentid2)
> > > > > lin(mlogit) family(binom) nrf(10) eqs(eatr cdrug2 dist ap
> > > sisters2 wage3
> > > > > edu3 cai share cons) nip(8) trace adapt-
> > > > 
> > > > I think that Stata's error message for numerical overflow,
> > > r(1400), is
> > > > helpful here, because gllamm uses numerical integration. The error
> > > > description for r(1400) begins with this sentence: "You
> > > have attempted
> > > > something that, in the midst of the necessary calculations, has
> > > > resulted in something too large for Stata to deal with accurately."
> > > > Thus, I guess that if Kevin tries to estimate a simpler model, the
> > > > message may disappear. Kevin, for example how many choice categories
> > > > do you have? Try with fewer choice categories.
> > 
> > *
> > *   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/
> 
> 
> *
> *   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index