[Limdep Nlogit List] Endogenous Stratification- Help with Code

[William Greene]

Peter. TLQ is a function of T and T is one of your parameters.
The problem is being caused by TLQ becoming nonpositive. Since the
function in the second command is different from the one in the 
first, it's not possible to predict what will happen to T. I would
say you have to force T to be positive, either by estimating ET
= exp(T) or T2=T^2. (I would use the first.)  There
are a couple oddities in the second code. It appears that y*log(1-w)
became y*log(1) (which equals zero, of course), and the last term
is log(exp(b1'x)) = b1'x.  Finally, since you computed a1=Exp(b1'x),
you should be using a1 in the subfunctions that follow.  Also, did
you recompute K before you fit the second equation, if you changed
the namelist?
/Bill Greene

----- Original Message -----
From: "Peter Edwards" <pedwards at UDel.Edu>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Tuesday, September 23, 2008 9:12:54 PM GMT -05:00 Columbia
Subject: [Limdep Nlogit List] Endogenous Stratification- Help with Code 

Hello All

I am still trying to figure out how to get the code to work for a log 
likelihood function for an on site travel cost study.
I am trying to model the data using a negative binomial regression including 
accounting for overdispersion and endogenous stratification.
This is the code Iv'e used for overdispersion as per the Limdep Manual and 
Greene's text.

NAMELIST; X=ONE,TRIPCST4,BIRDHRS,AGE,GENDER,EDUC,INC0000,EQUIP00$
calc; k=Col(x)$
create;y=totaltri$
negbin;Lhs=Y;Rhs=X$
Matrix; b0=b;t0=1/alpha$
Calc; q0=0$
maximize;start=b0,t0,q0
;labels=k_b,t,q
;Fcn=a1=Exp(b1'x) |
tlq=t*(a1^q) |
w=tlq/(a1+tlq) |
Lgm(y+tlq)-Lgm(tlq)-Lgm(y+1)+tlq*Log(w)+y*Log(1-w)$

This liklihood function works fine.


However when I adjust the code (see below) to account for endogenous 
stratification as per the literature (Haab and McConnell 2002, page 181)


Matrix; b0=b;t0=1/alpha$
Calc; q0=0$
maximize;start=b0,t0,q0
;labels=k_b,t,q
;Fcn=a1=Exp(b1'x) |
tlq=t*(a1^q) |
w=tlq/(a1+tlq) |
y*Lgm(y+tlq)-Lgm(tlq)-Lgm(y+1)+tlq*Log(w)+y*Log(1)-Log(tlq+Exp(b1'x))+(y-1)*log(exp(b1'x))$



I get the following message. with this ....NAMELIST; X=ONE,TRIPCST4$ or

Warning 137: Iterations: function not computable at crnt. trial estimate

Error 590: Obs.= 1 Cannot compute function: GMA of-#

Maximum iterations reached. Exit iterations with status=1.

Error 143: Models - estimated variance matrix of estimates is singular

Error 447: Current estimated covariance matrix for slopes is singular.



And the following messages with this...  NAMELIST; 
X=ONE,TRIPCST4,BIRDHRS,AGE,GENDER,EDUC,INC0000,EQUIP00$

Warning 137: Iterations: function not computable at crnt. trial estimate

Line search does not improve fn. Exit iterations. Status=3

Check derivatives (with ;OUTPUT=3). This may be a solution

if several iterations have been computed, not if only one.

Error 806: (The log likelihood is flat at the current estimates.)

Error 143: Models - estimated variance matrix of estimates is singular

Error 447: Current estimated covariance matrix for slopes is singular.

I am not sure where I am going wrong.

I would appreciate any assistance.



PETER


++++++++++++++++++++++++++++++++
Peter E.T. Edwards
PhD. Candidate
University of Delaware
Graduate College of Marine and Earth Studies
Room 312 Robinson Hall
Newark, DE 19716
http://www.ocean.udel.edu/




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