Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Two Part Model

From   "Naji Nassar (MIReS)" <>
To   <>
Subject   st: RE: Two Part Model
Date   Fri, 25 Jun 2004 17:18:39 +0200


I used to model a twopart model in marketing (Buy/NoBuy as cloglog, and
quantity as truncated Poisson)
The likehood for each observation
I think you'll have to take into account all the observations (even those
who never incur an expense)..
Please pay attention to the percentage of 'patients' (if less than 5% you'll
get rare events biases..)
Normally, as the 'non-patients' show no variance for the expenses, the two
models show no significat correlations

You can also try a two step modelling (one estimation for the decision -all
observations-, an other for the expenses -patients-..)
Hope this helps

-----Message d'origine-----
De :
[]De la part de Simone Boyce
Envoye : vendredi 25 juin 2004 17:01
A : Stata
Objet : st: Two Part Model


I'm interested in modeling a two part model, which
seems to go by many names in different fields:
two-tiered and hurdle models.  Essentially, I'm trying
to model medical cost expenditure, but the data
consist of many individuals who never incur an
expense.  Hence, there are many zeroes in the data.

As a result, I'm separating the decision into two
1) the decision to see a doctor and 2) the amount of
the expense:

P(y>0) =X_1B_1+e_1                       (1)
E(y|y>0) = log(y)|y>0= X_2B_2+e_2  (2)

where the error terms between equations (1) and (2)
are independent.

I know that this is usually estimated as a probit (eqn
1) and then OLS (eqn 2) to compute
E(y)= P(y>0)E(y|y>0).  However, how do I determine the
marginal effects and the associated standard errors?
Especially, if one of the variables that I'm
interested in is in both equations.  Has anyone
programmed the two part model?


Do you Yahoo!?
New and Improved Yahoo! Mail - Send 10MB messages!
*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index