[Limdep Nlogit List] Nlogit4 and bimodal distributions

[Stephen Beville]

 I'm a new-comer to choice modelling and I'm investigating anglers'
recreational fishing site choices using  Nlogit4.0 and stated preference
data. I'm having trouble getting a good fit, or in other words significant
t-stats, for the attribute in the model, "bag limit" (i.e. # of fish anglers
can take home) which in my design has been specified with levels (0,2,4).
The attribute is insignificant with an MNL specification and also when I try
a random parameter specification using normal, uniform, triangular
distributions. I've also tried constraining the variances of the
distributions to (to allow for a wide variance), and still have not achieved
a desired fit-though the fit improved over the MNL model usign a triangular
distribution constrained so that the variance was eqaul to twice the mean.

The underlying problem I believe, as to why the parameter is insignificant
in all mentioned cases, is due to extreme heterogeneity in angler
preferences, or preference poles.   Where a large portion of the angler
respondents view (prefer) the attribute level (0), or in other words catch
and release, because it may in their minds represent a proxy for high
fishing quality and where some anglers view the attribute level (4) as
preferable because they can take more fish home.

Hence, I believe what I really need to use is a bimodal distribution which
has two peaks-one for catch and release-oriented anglers and one for
consumption-oriented anglers. Can anyone please advise me if Nlogit is
capable of specifying a bimodal distribution or bimodal-like distribution.
Or offer any suggestions how I may solve this conundrum.

Thank you so much for any help you can offer.


-- 
Stephen Beville
PhD Candidate ESD
Lincoln University
(027) 338 8438