[Limdep Nlogit List] zero expenditures - Grouped data regression model

[William Greene]

Steven. This one is not supported for the GROUPED data model.
The natural case is the truncated regression model, for which the
model is already in place.  The density is automatically. for the
grouped data (discrete outcome) model, it would be necessary to
divide all the probabilities by 1 - Prob[y=0].  You could write
the likelihood and use MINIMIZE. But, it is not a built in
procedure.
Regards,
Bill Greene

----- Original Message -----
From: "Steven Vos" <Steven.Vos at faber.kuleuven.be>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Monday, June 21, 2010 4:51:01 PM GMT -05:00 Colombia
Subject: Re: [Limdep Nlogit List] zero expenditures - Grouped data regression model

Dear Mr. Green,
Thank you for your response to my question.
If I understand it right the second model would exclude the subsample op people with zero-expenditures.
You suggest to scale up the probabilities in my grouped data model by the probability that y>0.
However, I'm not sure how to relate the first model to the second model (probabilities)?

Many thanks,

Kind regards,
Steven Vos

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au [mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of William Greene
Sent: maandag 14 juni 2010 12:35
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] zero expenditures - Grouped data regression model

Steven.
The suggestion was that the zero values be treated as one of two
decisions in a hurdle (two part) model. Your equations would be
D = 0 if y = 0 <---- possibly a probit model
D = 1 if y > 0
Then a model for y|y>0, such as a truncated regression in the 
continuous case.  For your grouped data, with the discrete outcome,
probabilities for your GROUPED model are scaled up by the 
probability that y > 0.
/Bill Greene

----- Original Message -----
From: "Steven Vos" <Steven.Vos at faber.kuleuven.be>
To: limdep at limdep.itls.usyd.edu.au
Sent: Monday, June 14, 2010 4:31:25 AM GMT -05:00 Colombia
Subject: [Limdep Nlogit List] zero expenditures - Grouped data regression model

Dear Limdep user,





I've got another question concerning grouped data regression models (@Andreas: Thank you for your reaction to my previous question):



My dependent variable are expenditures (0€, < 100€, 100-200€, 200-400€, >400 €).

In a first model model I've collapsed the two first categories. So the first category (< 100€) included zero-expenditures.

However I've got the remark that these zero-expenditures should be a category on its own.

So my idea is to use
  y=1  if y* < 0.1 €
  y= 2  if 0.1€ ≤y* < 100 €
  y= 3  if 100€ ≤y* < 200 €
  y= 4  if 200€ ≤y* < 400 €
  y= 5  if 400€ ≤y*


In my opinion it seems reasonable to assume that the latent variable y* can take negative values.



Is this straight forward, or are there better solutions to threat the zero-expenditures as a category on its own?



Many thanks,



Steven Vos

K.U.Leuven

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au