This sounds very much like a two-part model: your 0 values are clearly
identified, so a logistic regression (or probit) can be used to predict
0 or non-zero, and a multiple regression model can be used for the
non-zero responses. See an issue of Statistical Methods in Medical
Research from 2001 that has several articles on this. They are also
called Hurdle models in Econometrics.
Tony
Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Anita
Sent: Friday, February 08, 2008 5:06 AM
To: [email protected]
Subject: st: random effects with truncreg possible
Dear Statalist Members
Sorry for posting again, I learned that I didn't explain my problem
accurately. Sorry for any inconvenience caused.
I am currently estimating a model based on Cragg ("Some statistical
models for Limited Dependent Variable with Application to the Demand of
Consumer Goods", 1971)
My research topic is to compare the models used in the literature to
explain the allocation of bilateral ODA (official development
assistance, so money for development).
So there exist countries who do not recieve ODA (value =0) and others
do. I do have data on both. So it is not a typical truncation problem.
First there is a selection equation to decide if a country gets ODA, I
estimate this via a Probit Model. (y1=1 if they get ODA)
on a second step I estimate for all positive outcomes of the Probit
Model a Regression to estimate how much money they get.
As I am working with a panel (different developing countries over time)
I would like to do this with random effects. For Probit there exists a
xtprobit command, for truncreg (command for the regressionpart) I didn't
find anything.
Is there a possibility to do this?
As I indicated above, I have to compare the models used in the
literature, so I have to do Cragg's Model preferably with random effects
so I can compare with other models.
As a second question: is there a better way to solve the problem as with
truncreg? As I know, the studies used this, but it does not seem to fit
very well
thanks a lot
Anita
*
* 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/
