Re: st: Model fit-statistic for zero-inflated negative binomial models

[Joseph Luchman <jluchman@gmail.com>]

Hi Neeraj,

  The most of the pseudo-R2's produced by - r2c - describe the whole
model - however, the "dp" statistics produced for negative Binomials
compare the fit of the negative Binomial to that of the same
specification using a Poisson distribution (and is thus quite similar
in spirit to the Vuong test without a null hypothesis significance
test component).

  The - r2c - was developed from internal research by my employer that
was worth sharing with the Stata community.  Our internal research
using the command is not cite-able.  I'm not familiar with any papers
that have cited it yet - but, as you note, it's only 4-or-so months
old.  Presuming someone has decided to cite the command in their
manuscript, the publication lag is likely to still be "in effect."
That said, the SSC entry is cite-able and can be downloaded from
ideas.repec.org (http://ideas.repec.org/c/boc/bocode/s457592.html).

  A good source for reading over the different glm-based residuals is
Pierce & Schafer (1986) - full APA-style citation is below.

- joe

Pierce, D. A., & Schafer, D. W. (1986). Residuals in generalized
linear models. Journal of the American Statistical Association,
81(396), 977-986.

Joseph Nicholas Luchman, M.A.
----
Senior Research Associate | Fors Marsh Group
Email: jluchman@forsmarshgroup.com
forsmarshgroup.com
----
Doctoral Candidate
Industrial Organizational Psychology
George Mason University
https://www.researchgate.net/profile/Joseph_Luchman/


2013/6/10 Neeraj Iyer <iyeranna@gmail.com>
>
> Hello Joe,
>
> Thank you.  I was able to install the 'r2c' package and obtain outputs
> for my ZINB models.  I am currently reading about R-Squared measures
> based on deviance residuals.  I wanted to ask a few questions that
> might help me understand the fit statistics better:
>
> 1) Are the fit-statistics output by 'r2c' a measure of overall model
> fit or are these a measure of comparative fit (like the Vuong test)?
>
> 2) Since the 'r2c' program has been released very recently, have any
> studies been published that report using the 'r2c' generated
> fit-statistics?  I would like to cite those studies in my
> dissertation.
>
> 3) Could you direct me to some other papers (apart from Cameron and
> Windmeijer 1996) or annotated output that might explain the basis and
> interpretation of deviance residuals?
>
> Thanks again.
>
> Regards,
> Neeraj
>
> On Fri, Jun 7, 2013 at 8:48 AM, Joseph Luchman <jluchman@gmail.com> wrote:
> > Hi Neeraj,
> >
> >   My guess is that what you should report will be (sub-)discipline
> > specific.  That said, I have a postestimation module - r2c - (type:
> > ssc install r2c) that computes some fit statistics following count
> > models, including a ZINB model.  This may help in terms of your point
> > #3/other fit statistics.
> >
> >   Hope this helps,
> >
> > - joe
> >
> > Joseph Nicholas Luchman, M.A.
> > Senior Research Associate | Fors Marsh Group
> >
> > Email: jluchman@forsmarshgroup.com
> > forsmarshgroup.com
> > https://www.researchgate.net/profile/Joseph_Luchman/
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/faqs/resources/statalist-faq/
> > *   http://www.ats.ucla.edu/stat/stata/
>
>
>
> --
> Neeraj
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/