Dear STATAlist users,
I am trying to decide a model for a count dependent variable. Based on
the summary and tabulation of my dependent variable, I found the evidence
of both overdisperson (variance > mean) and excess-zero (about 10 percent
is zero.) I did some test about poisson, nbreg, zip and zinb but not sure
which model I should use.
Because pweight cannot be used for LR test, I took weight out in all
estimation II am not sure this will affect the result a lot). The results
are:
1) poisson vs. nbreg: LR test of alpha=0: chibar2(01) = 1047.64
Prob>=chibar2 = 0.000
2) poisson vs. zip (inflate all regressors used for estimating count
outcome):
log likelihood for poisson: -6214.3391
log likelihood for zip: -5168.085
3) poisson vs. zip (inflate part of the regressors used for estimating
count outcome):
log likelihood for poisson: -6214.3391
log likelihood for zip: -5257.639
(*** I don't understand the inflate option for ZIP/ZINB estimation. STATA
help menu says inflate is not optional, and is used to "specify the
equation that determines whether the observed count is zero". In my
model, there are X number of regressors used to predict the outcome.
Shall I use all of them in the varlist of "inflate"? I did test to use a
different cohort of regressors. What should be put in the varlist of
"inflate" anyway"???)
4) nbreg vs. zinb (inflate part of the regressors used for estimating
count outcome):
Vuong Test of Zinb vs. Neg. Bin: Std. Normal = 16.20 Pr> Z =
0.0000
5) nbreg vs. zinb (inflate part of the regressors used for estimating
count outcome):
Vuong Test of Zinb vs. Neg. Bin: Std. Normal = 10.70 Pr> Z =
0.0000
I am lost in the jungle of the test. Can anybody tell me how to decide
the model? Or should I conduct more test?
THanks a lot...........
Ying
*
* 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/
