[Limdep Nlogit List] Nested Logit Sequential (Two step) estimation
Saul Basurto Hernandez
SXB1022 at student.bham.ac.uk
Fri Jul 21 20:09:54 AEST 2017
I am trying to estimate a Nested Logit model using the Sequential estimation as the number of alternatives in my database make the FIML estimation not feasible.
In order to be familiarized with the software, I tried to fit the same models as in the NLogit 6 Reference Guide (pages N-528 and N-529) but, the results I got from the second step considerably differed with respect to those in the Reference Guide.
This is the example in the Reference Guide (using the clogit.lpj project):
NLOGIT ; Lhs = mode
; Choices = air,train,bus,car
; Tree = fly(air), ground(train,bus,car)
; Model: U(air,train,bus,car) = <0,at,ab,0> + bc * gc /
U(fly,ground) = ah * hinca
; Ivb = incvlu
; Conditional $
I got exactly the same results here but, not for the second step:
NLOGIT ; Lhs = mode
; Choices = air,train,bus,car
; Tree = fly(air), ground(train,bus,car)
; Model: U(air,train,bus,car) = <0,at,ab,0> + bc * gc /
U(fly,ground) = ah * hinca + aiv * incvlu
; Sequential $
This is what the reference displays:
-----------------------------------------------------------------------------
Second step estimates of nested logit model
Dependent variable Choice
Number of obs.= 210, skipped 0 obs
--------+--------------------------------------------------------------------
|Model for Choice Among Alternatives
AT| 2.38614*** .36950 6.46 .0000 1.66193 3.11035
AB| .76659** .32387 2.37 .0179 .13182 1.40136
BC| -.07659*** .01004 -7.63 .0000 -.09627 -.05691
|Model for Choice Among Branches
AH| -.01386*** .00428 -3.24 .0012 -.02225 -.00548
AIV| .04165 .05691 .73 .4642 -.06989 .15319
--------+--------------------------------------------------------------------
And, this is what I got using the same command (FIML rather than LIML estimator):
Hessian is not positive definite at start values.
Error 803: Hessian is not positive definite at start values.
B0 is too far from solution for Newton method.
Iterative procedure has converged
-----------------------------------------------------------------------------
Start values obtained using MNL model
Dependent variable Choice
Number of obs.= 210, skipped 0 obs
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
MODE| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Model for Choice Among Alternatives.................................
AT| .67162*** .19719 3.41 .0007 .28513 1.05810
AB| -.32457 .....(Fixed Parameter).....
BC| -.01981*** .5015D-06 ******** .0000 -.01981 -.01981
|Model for Choice Among Branches.....................................
AH| -.01539 .....(Fixed Parameter).....
AIV| 0.0 .....(Fixed Parameter).....
--------+--------------------------------------------------------------------
-----------------------------------------------------------------------------
FIML Nested Multinomial Logit Model
Dependent variable MODE
Warning: Model does not contain a full
set of ASCs. R-sqrd is problematic.
Number of obs.= 210, skipped 0 obs
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
MODE| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Attributes in the Utility Functions (beta)..........................
AT| 2.36903*** .37054 6.39 .0000 1.64278 3.09528
AB| .76469** .32447 2.36 .0184 .12875 1.40064
BC| -.07685*** .01004 -7.65 .0000 -.09653 -.05717
|Attributes of Branch Choice Equations (alpha).......................
AH| .00473 .00681 .69 .4876 -.00863 .01808
AIV| 0.0 .9261D+14 .00 1.0000 -.18151D+15 .18151D+15
|IV parameters (RU1), tau(b|l,r),sigma(l|r),phi(r)...................
FLY| .08202 .06278 1.31 .1914 -.04103 .20507
GROUND| -.04438 .06521 -.68 .4961 -.17220 .08343
--------+--------------------------------------------------------------------
I've tried with different specifications but, the Sequential estimation seems to be unavailable.
Am I doing something wrong (I just copied and pasted the same commands)? Or
Is the Sequential estimator unavailable in NLogit 6 (I am using the student license)?
Best wishes,
Saul Basurto
More information about the Limdep
mailing list