Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: xtnbreg, fe does not converge when xtpoisson does


From   Klaus Pforr <[email protected]>
To   [email protected]
Subject   Re: st: xtnbreg, fe does not converge when xtpoisson does
Date   Wed, 22 Aug 2012 21:33:02 +0200

<>

I hope that this does not confuse the statalist-thread, but you should always ask at the list, and not via private communication.

Considering your problem, I should have added, that the lack of variance might be the cause, but as Maaren already suggested, there might be other reasons. If the variance is the problem, then it does not necessarily mean, that you don't have overdispersion.

All I wanted to say is, that the fixed-effects of both models are completely different. For xtpoisson its a fixed effect for the scale parameter, and for the nbreg its a fixed effect for the shape parameter. Paul Allison recommends on his blog (http://www.statisticalhorizons.com/fe-nbreg), to either use the nbreg with individual dummies or to use the hybrid-model-strategy. At his blog, he also recommends some literature, which gives straight-forward implementations.

I forgot to add one comment on the overall modelling strategy. To your advantage you gave information about your dependent variable. Although it is a count variable, it ranges from 0 to 800000. Scott Long (1997) recommends for cross-sectional count-data models with high average counts, to switch to linear models. I don't know of any recommendations about panel models, but the argument should also apply there. To be on the safe side, you should use robust standard errors.

Literature:
J. Scott Long, 1997, Regression Models for Categorical and Limited Dependent Variables (Advanced Quantitative Techniques in the Social Sciences). Sage Publications. ISBN 0-8039-7374-8.

Klaus



Am 22.08.2012 20:57, schrieb Anastasiya Zavyalova:
Klaus,

If I do not have enough variance for the overdispersion-factor in the xtnbreg, does this mean the dependent variable is NOT over-dispersed? Alternatively, if the xtnbreg model does not "run," am I correct to assume that it's not an appropriate estimation for my dependent variable?

Thank you.

Annie
*
*







On Aug 22, 2012, at 9:01 AM, Klaus Pforr wrote:

<>

xtpoisson und xtnbreg look at different fixed-effects. xtpoisson adds a
fixed effect to Xb, i.e. an additional possible correlated
within-group-invariant independent variable. xtnbreg models a fixed
over-dispersion-factor (in contrast to the simple nbreg-model): <<Here
<random effects> and <fixed effects> apply to the distribution of the
dispersion parameter, not to the x term in the model.>> (Manual entry for xtnbreg. This means, that your xtpoisson converges because you have enough variance for the Xb-fixed-effect, but you may have not enough variance for
the fe for the overdispersion-factor.

Klaus Pforr

__________________________________

Klaus Pforr
GESIS -- Leibniz Institut für Sozialwissenschaft
B2,1
Postfach 122155
D - 68072 Mannheim
Tel: +49 621 1246 298
Fax: +49 621 1246 100
E-Mail: [email protected] <mailto:[email protected]>
__________________________________


-----Ursprüngliche Nachricht-----
Von: [email protected] <mailto:[email protected]>
[mailto:[email protected]] Im Auftrag von Maarten Buis
Gesendet: Mittwoch, 22. August 2012 11:41
An: [email protected] <mailto:[email protected]>
Betreff: Re: st: xtnbreg, fe does not converge when xtpoisson does

On Tue, Aug 21, 2012 at 8:08 PM, Anastasiya Zavyalova wrote:
I have a dependent count variable. The range is from 0 to 800,000.
When I run an xtpoisson with fixed effects it runs well and, in fact,
all my coefficients are highly signifiant.
However, when I run an xtnbreg with a fixed-effects option (turns out
the variable is over-dispersed), I receive the following error
message: "discontinuous region encountered; cannot compute an
improvement; r(430);" What could be the reason? Thank you.

A negative binomial model is a much more complicated model than a Poisson,
so the fact that a Poisson converges does not guarantee that a negative
binomial will converge as well.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   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/


*
*   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/




--
__________________________________
Klaus Pforr
GESIS -- Leibniz Institut für Sozialwissenschaft
B2,1
Postfach 122155
D - 68072 Mannheim
Tel: +49 621 1246 298
Fax: +49 621 1246 100
E-Mail:[email protected]
__________________________________

*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index