[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: Zero-truncated Negative Binomial convergence

From   "Lachenbruch, Peter" <>
To   <>
Subject   RE: st: Zero-truncated Negative Binomial convergence
Date   Tue, 17 Mar 2009 10:16:07 -0700

If the zeros are identifiable from some other information - e.g.,
hospital costs will be 0 if the patient isn't hospitalized or in this
case we would know the visits to a specialist is 0 if the patient is
only seen for physical exams, etc. - then a two-part model might work.
In this case one uses two models: one for the number of visits if visits
are >0 and one (a logistic) for distinguishing 0 vs. non-zero.  I have a
few papers in Statistics in Medicine in 2001 that may be helpful.  

I would emphasize that in this case, some of the 0 visits to a
specialist are in patients who should have seen a specialist but didn't.


Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001

-----Original Message-----
[] On Behalf Of Maarten buis
Sent: Monday, March 16, 2009 2:08 PM
To: stata list
Subject: RE: st: Zero-truncated Negative Binomial convergence

--- Emily Wilson wrote:
> I am having trouble running a zero-truncated negative binomial
> regression. The dependent variable is: # of visits to a
> specialist physician in the past year, and the distribution is 
> something like:
> 0 visits ~= 93,000
> 1 visit ~= 15,000
> 2 visits ~= 1,000
> 3 visits ~= 500

The zero truncated distribution assumes that there are no zeros.
In your case you definately do have zeros. I would start with a 
regular -poisson-, and than I might worry about excess zeros, 
for which you can look at the zero inflated poisson (-zip-), the 
negative binomial (-nbreg-), and zero inflated negative binomial 
(-zinb-). There is a nice discussion of these models and how to
choose between them in this book:

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index