[Limdep Nlogit List] double hurdle again

[Brett Smith]

Achilleas

Thank you for this. 

It was exactly those set of commands that started me down this track (chi
squared was critical). I guess that the independent probit and truncated
regression model reproduce the independent double hurdle. I really have
trouble seeing this when the set of variables in the probit regression is to
different to that used in the truncated regression, but ok.

This leads to another question about obtaining the marginal effects (or
obtaining a conditional mean E[y | X, Z] where X are variables in truncated
and Z are vars in probit). My first guess is that the conditional mean is
Prob(probit)*E(y* |X)...but not sure and will look in to it. Any advice is
welcome.

Thanks and Regards,

Brett



-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Vassilopoulos
Achilleas
Sent: Tuesday, 8 June 2010 9:54 PM
To: 'Limdep and Nlogit Mailing List'
Subject: Re: [Limdep Nlogit List] double hurdle again

Brett,

The estimation of double-hurdle models is straightforward since all you need
is a discrete choice model using all observations and estimation of the
parameters of the density f (y|m = 1, x) using only observations with y > 0.
Below is a code to run a double-hurdle model and perform a chi-squared test
to see whether the double hurdle model is more appropriate than Tobit.

Namelist ; x = .......... $
Tobit 	; Lhs = y ; Rhs = x $
Calc	; LTobit  =  logl $
Create; m = y > 0 $
Probit ; Lhs=m; Rhs=x $
Calc ; List ; LProbit=Logl $
Truncate ; Lhs=y ; Rhs=x $
Calc ; List ; LTrunc=Logl
; LRtest=2*((LProbit+LTrunc)-LTobit) $
; prob = Chi(LRtest,df) $


best,
_____________ - _______________

Achilleas Vassilopoulos

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Brett Smith
Sent: 08 June, 2010 06:05
To: 'Limdep and Nlogit Mailing List'
Subject: [Limdep Nlogit List] double hurdle again

Dear List, 

I asked a question about the double hurdle model a week or so ago. The
reason being that it is unlikely (fails likelihood test) that the
restrictions of the Tobit regression apply to the data. 
After fishing though the manuals I cannot see how to run a double hurdle
model in limdep. I am sure it is obvious...but just not seeing it. I have
added some code below which is using the minimize function (based on Jones
1992). 

I would like to ask the group if this is the correct (only, easiest) way to
estimate the double hurdle.

Thank you again,

Brett 



CREATE;ZERO = 0$

NAMELIST	;ProbReg = Z, ZERO, ZERO....? Repeat zero for the number of
vars in X
		;TrunReg = ZERO....,X,ZERO$  ? repeat zero for the number of
vars in Z, and one for SE in truncated regression

?run probit and  censored regresseion for start values

Probit; LHS= Travel; Rhs=Z$
Matrix; Bprob=b$? save probit parameters
Trunc; Lhs=y; Rhs=X$
Matrix; btrunc=b$? 
Matrix; bstart= bprob,_btrunc $ ? can't remember syntax but combine the two
vectors 

?Double hurdle model
MINIMIZE	;START= bstart
		;LABELS = B1 ,B2, B3, .... ,SE
		;FCN = - (1 - Travel)*LOG( 1 - PHI(b1'ProbReg)*PHI(
(b1'TruncReg)/SE) ))          ?log likelihood for travel
			 - Travel*(LOG(PHI(b1'ProbReg)) + LOG(NOl((Y -
(b1'TruncReg)/SE)))/SE)$    ? Log likelihood for observations greater than
zero




_______________________________________________
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