[Limdep Nlogit List] Scale parameter from Nested Logit

[Koetse, M.J.]

Dear All,

I am trying to estimate a Nested Logit in order to capture scale effects in choices made at three distinct stages (s1,s2,s3) in a choice experiment. For this I based the model on the Nested model provided in:

https://protect-au.mimecast.com/s/SxOwCoVzGQioln9Zi1xNba?domain=people.stern.nyu.edu

My model is given below with additional information. For all the tree structures I have tried I either get (after the CL model has provided normal output, so the problem is with the nested part):

Initial iterations cannot improve function.Status=3
Error    805: Initial iterations cannot improve function.Status=3
Function=  .15883760536D+04, at entry,  .15883760536D+04 at exit
Error   1025: Failed to fit model. See earlier diagnostic.

OR

Error   1026: Singular VC may mean model is unidentified. Check tree.
Error   1027: Models - estimated variance matrix of estimates is singular
Error   1027: Models - estimated variance matrix of estimates is singular

Perhaps my data simply do not allow estimation of the model, but perhaps my syntax is flawed.

Any tips or comments are highly appreciated!

Best regards,
Mark


SYNTAX:

NLOGIT
;lhs=choice,cset,altij
;choices=1,2,3,4,5,6,7,8,9
;tree=mode [ s1(1,2,3) , s2(4,5,6) , s3(7,8,9) ]
;ivset: (S1)=[1.0]
;model:
U(1)  =      asc1 /
U(2)  =      r1*att1 + e1*att2 + p1*att3 + don1*don /
U(3)  =      r1*att1 + e1*att2 + p1*att3 + don1*don /
U(4)  =      asc2 /
U(5)  =      r2*att1 + e2*att2 + p2*att3 + don2*don /
U(6)  =      r2*att1 + e2*att2 + p2*att3 + don2*don /
U(7)  =      asc3 /
U(8)  =      r3*att1 + e3*att2 + p3*att3 + don3*don /
U(9)  =      r3*att1 + e3*att2 + p3*att3 + don3*don $

- Cset is always equal to 3 in my case.
- Choice options in stage 1 were always 1,2,3, those in s2 were always 4,5,6, and those in s3 were always 7,8,9. I would like to estimate the scale at the stage (s1,s2,s3) level, and so I specified the tree accordingly.