[Limdep Nlogit List] Problems in Nested Multinomial Logit Model
William Greene
wgreene at stern.nyu.edu
Wed Sep 10 22:01:38 EST 2008
Dear Ester Ramos Pérez: The line search error indicates that it is
not possitle to fit your model with the weights. The log likelihood
has changed only slightly from entry to exit. This is an indicator
of a problem in your specification of the model - here it because
of the weights. Without the data in hand, I cannot be any more specific
than that. For the second question you raised, your model is not
identified. I would guess that the problem is the constant term in
the utility function for Public. If you remove BP from the equation, I
suspect that the rest of the model will be estimated OK. When a model is
overspecified, it will predict the outcomes as well as one that is
not - that is why you are finding that the unidentified model is
"very good."
/B. Greene
----- Original Message -----
From: To: limdep at limdep.itls.usyd.edu.au
Cc: Sent: Wednesday, September 10, 2008 7:50:12 AM GMT -05:00 US/Canada Eastern
Subject: [Limdep Nlogit List] Problems in Nested Multinomial Logit Model
We're fitting a nested multinomial logit model. As a response variable we have three alternatives (public transport, car and motorcycle). The syntax of the model, which we use, is:
NLOGIT;Lhs=MODE,N_ALT,ALT; Wts=ELEV;
Choices=Public,Cotxe,Moto;
Tree=Travel[TP(Public),VP(Cotxe,Moto)];
Model:
U(Cotxe)=BC+BCC*CGCO+BCA*DIA+BCR*RISC/
U(Moto)=BMC*CGMO/
U(Public)=BP+BPC*CGTP+BPI*INT1;
PrintVC;Crosstab;
Describe;List;ShowTree;Effects:;Pwt;Means$
When we obtain the results after the execution of the model, we can see these two errors.
Line search does not improve fn. Exit iterations. Status=3
Error 806: Line search does not improve fn. Exit iterations. Status=3
Function= .11853876438D+06, at entry, .11812641048D+06 at exit
Error 1025: Failed to fit model. See earlier diagnostic.
Our first question is what's the meaning of these errors, and how can we solve them?
According to the above, We're fitting the same model without weight. The syntax of the model, which we use, is (without weight):
NLOGIT;Lhs=MODE,N_ALT,ALT;
Choices=Public,Cotxe,Moto;
Tree=Travel[TP(Public),VP(Cotxe,Moto)];
Model:
U(Cotxe)=BC+BCC*CGCO+BCA*DIA+BCR*RISC/
U(Moto)=BMC*CGMO/
U(Public)=BP+BPC*CGTP+BPI*INT1;
PrintVC;Crosstab;
Describe;List;ShowTree;Effects:;Pwt;Means$
Now, the results in all parameters of Public transport are not significant.
+---------+--------------+----------------+--------+---------+
|Variable | Coefficient | Standard Error |b/St.Er.|P[|Z|>z] |
+---------+--------------+----------------+--------+---------+
Attributes in the Utility Functions (beta)
BC .66774715 .08919300 7.487 .0000
BCC -.07808303 .01075774 -7.258 .0000
BCA -2.51885645 .15944988 -15.797 .0000
BCR -.06917638 .00737276 -9.383 .0000
BMC -.13290383 .02072193 -6.414 .0000
BP -.76015104 158.734729 -.005 .9962
BPC -.03802969 7.94136259 -.005 .9962
BPI -.38789100 80.9993955 -.005 .9962
IV parameters, tau(j|i,l),sigma(i|l),phi(l)
TP 2.10762224 440.113628 .005 .9962
VP 2.35485967 .12374261 19.030 .0000
We don't understand why, because if we apply the model the degree of success is very good
Could you help us?
Ester Ramos Pérez
Gabinet d'Estudis
Transports Metropolitans de Barcelona
Tf. 93 298 75 57
Fax: 93 298 71 21
E-mail: eramosp at tmb.net <mailto:eramosp at tmb.net>
__________________________________________________________________________
Aquest missatge electrònic està dirigit únicament a les adreces indicades. El seu caràcter confidencial, personal i intransferible està protegit legalment. Qualsevol revelació, ús o reenviament no autoritzat, total o parcial, està prohibit. Si ha rebut aquest missatge per equivocació, notifiqui-ho immediatament a la persona que l'ha enviat i esborri el missatge original juntament amb els seus fitxers annexos sense llegir-lo ni gravar-lo.
TMB (Transports Metropolitans de Barcelona)
Informació interactiva de transport: www.tmb.net
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au
More information about the Limdep
mailing list