I am curious where you read that GOF tests on Poisson data are best if done
on compressed data - if this is what you are saying. There are certain GOF
tests that require binomial (e.g., logistic regression) data to be compressed
or grouped, but I have never seen a reference to compression being better for
count data. Since many count response models use an offset, this amounts to
nearly the same thing as grouping. Regardless though, I see no reason for
it. Do you recall the source?
Joe Hilbe
===================================================
Dear All,
I am a novice learing stata......this is probably very basic but Im
struggling a bit.
I have a cohort study of miners developing TB who are HIV +ve or -ve
over a 8 year study period.. I have found the best poisson regression
model testing with LRTs I have included a lexis expansion for age of
subject effects. The Stat manual on poisgof makes no mention of
compressing the data. I have read that ungrouped data is not correct
for this command.
my final model looks like this
xi:poisson _d i.hiv ageband2 i.prevtb i.sil if sil~=.,e(y) irr
I have used the following commands to compress the data before doing the
poisgof
preserve
. collapse (sum) _d y,by(hiv ageband2 job1 prevtb sil)
. xi:poisson _d i.hiv ageband2 i.job1 i.prevtb i.sil if sil~=.,e(y) irr
(output ommitted)
then poisgof
I am sorry if there is insufficient information I am not sure how much
to include.
Many thanks for help in this matter.
sincerely
Mark Tracy
Director and Neonatologist
Nepean Newborn Intensive Care Unit
SWAHS
Lecturer Department of Paediatrics and Childhealth
Sydney University
*
* 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/
