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]

st: AIC and zero-truncated poisson regression

From   "Ryan Fisher" <>
To   <>
Subject   st: AIC and zero-truncated poisson regression
Date   Tue, 4 Dec 2012 10:46:49 -0700

Hi list,
I'm a fairly new user of Stata and am doing some preliminary analyses (i.e.,
some data exploration and "getting to know Stata" exercises) with some data
that I have on counts of offspring produced at bird nests. Zeroes have been
removed because they are part of another analysis looking at nest survival
(a binary response variable). I'm interested in using a model selection
approach (AIC maybe) for the count data (likely zero truncated poisson or
zero-truncated negative binomial regression) in relation to several
covariates of interest (landcover, "distance to" variables, etc..). However,
I noticed that the zero truncated poisson output indicates that it is using
pseudo-likelihoods. I've been using SAS prior to Stata and several of the
SAS message boards suggest that model comparisons using AIC shouldn't be
attempted with pseudo-likelihoods (and SAS seems to produce a fairly
consistent warning about using pseudo-likelihoods for AIC model comparisons
in procedures such as glimmix).  I can request AIC output from Stata using
estat ic and everything looks fine, but I'm just wondering how reliable that
AIC value is for model comparisons in the type of analysis I explain above?
Note that I'm not including any random effects at the moment. Perhaps there
is just an option that I'm missing or some further reading the list might
Thanks for any help/suggestions!

*   For searches and help try:

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