re: re: st: Package -r2c- now available in SSC

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

That is very helpful information, Joe... thanks!

Date: Mon, 4 Feb 2013 08:06:49 -0500
From: Joseph Luchman <jluchman@gmail.com>
Subject: re: st: Package -r2c- now available in SSC

Hi Ariel,

  Certainly.  To my understanding, -countfit- is mainly helpful for
selecting a count model and uses the information criteria and
predicted probabilities to do so.

  I see the -r2c- as useful mainly for explanatory purposes following
model selection and estimation and, although there is a statistic
reported for negative binomial models (i.e., the DP r-square) that is
intended to compare models, -r2c- is not intended for model selection
purposes (though it can be used for that purpose if the analyst
desires).

  The indexes reported are all r-squares (of different flavors, though
several lack properties such as "boundedness" between 0-1, that an
r-square should possess) the recommended statistic, the deviance
r-square is based on the model log likelihood and so it uses the same
information as the AIC and BIC reported in -countfit-.  For what it's
worth, the -r2c- also incorporates probability weights, an
exposure/offset (both in the main count model, and the inflate model
for zero-inflated models), and incorporates more of Stata's built-in
count models (truncated counts, generalized negative binomial).

- - joe


*
*   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/