Bookmark and Share

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

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

RE: st: When to use Poisson or Negative Binomial

From   "Querze, Alana Renee" <>
To   "" <>
Subject   RE: st: When to use Poisson or Negative Binomial
Date   Fri, 27 May 2011 11:34:37 +0000

Thank you for your thoughts on which to choose. When I run my nbgreg model I get significant results for more variables than the xtpoisson (if the results were similar it wouldn't matter which I used) so I want to have a justification for the one I use.

@Argyn: it does make sense to me if the model is wrong making it robust is not exactly important, but from the books I've read authors seem to believe overdispersion can be overcome using xtpoisson with boot strapped standard errors.

@Maarten: what else do I need to know besides the marginal mean and variance to choose the right command? I chose between random and fixed effects by using the Hausman test... is there a similar test in STATA that would help me know which model better fits the data?

From: [] on behalf of Maarten Buis []
Sent: Friday, May 27, 2011 3:02 AM
Subject: Re: st: When to use Poisson or Negative Binomial

> On Thu, May 26, 2011 at 12:14 PM, Querze, Alana Renee <> wrote:
>> But I don't know whether it is better to sacrifice robustness or efficiency.
>> Anyone know how to justify the use of one over the other? (BTW I have just under 400 districts and 52 months in my panel data).

On Fri, May 27, 2011 at 1:45 AM, Argyn Kuketayev wrote:
> i wouldn't use Poisson if the variance is much greater than the mean,
> e.g. mean = 17, variance = 40.
> who cares how robust is the estimate if it simply doesn't fit

That is not quite all there is to say on this topic. You first need to
know with respect to what aspects of the model a model is "robust".
For example, if we talk about robust standard errors than that
typically is typically robust with respect to the model of the
variance and higher moments, but assumes that the model is correct
with respect to the mean. The comparison of the marginal mean and
variance does not tell us enough to distinguish between the two.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index