Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Constraint, Poisson, & pseudo R^2


From   Rkaminski@aol.com
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Constraint, Poisson, & pseudo R^2
Date   Wed, 20 Oct 2004 18:22:26 EDT

I usually report one or two pseudo R^2 values as a standard practice  and use 
other methods to assess model fit. I was just surprised that  version 7 
reports a pseudo R^2 value whereas version 8 does not.  
 
In a message dated 10/20/2004 7:58:08 AM Eastern Standard Time,  
maartenbuis@yahoo.co.uk writes:

I do not have the entire answer, but I  think I can narrow the problem
down a bit. The pseudo R^2 that -fitstat-  report require the
loglikelihood of the model with only the constant, and  -poisson- with
constraints do not report or save these. I do not know why.  You could
of course estimate your model with only the constant and use  the
formulas on page 92 of the long and freese book to calculate  them
manually, but I would not do that before someone tells us why  STATA
no longer reports these.

Why do you want to use the pseudo R^2?  If you want to compare
constraint and unconstraint models, the likelihood  ratio test is by
far more appropriate. If you want to asses model fit than I  would
first test for overdispersion (for instance estimate a  negative
binomial -nbreg- and test whether alpha equals zero) since this  is
the most common problem with poisson regression.

Maarten

---  In statalist@yahoogroups.com, Rkaminski@a... wrote:
> Dave - thanks for  your reply.  I'm not using Scott's add on for
the  model
>  estimation. Interestingly, Stata version 7 provides the  pseudo-R^2
value  when
> using the constraint, but version 8 does  not.  This makes me think
Stata
> intentionally eliminated it from  the output in the latest
incarnation.  Hopefully
> someone will  know the reason why. Maybe it is not appropriate to
report the
> R^2  value under these conditions?
>
> Bob Kaminski
> Dept. of  Criminology & Criminal Justice
> University  of South  Carolina
>
>
> In a message dated 10/19/2004 1:50:37 PM  Eastern Standard Time,
> jacobs.184@s... writes:
>
> That's  almost certainly because Long and Freeze use their add on
program
>  that calculates psuedo R squares.  I can't remember the name of  the
add  on
> program, but it is listed repeatedly in the Long and  Freeze  book.
>
> Dave Jacobs
> Ohio  State
>
> At 08:00 AM 10/19/2004 -0400, you  wrote:
>  >In the Long & Freeze book published by  Stata, the   authors
estimate a
> Poisson
> >regression model using the  constraint   command to control for
unequal
> >exposure.  It produces the same  coefficients as  when using
exposure  or
> >offset, but
> >the  pseudo-R^2 value is  substantially larger  when using the
constraint
>  >option
> >(p. 265). However, when I use the constraint   option,  Stata does
not
> >produce a
> >pseudo-R^2  value. Does anyone know  whether  there has been a
change made  to
> >the program by  Stata?
> >
> >Thanks  in  advance.
> >
> >Bob  Kaminski
>  >
> >*
> >*   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