[Limdep Nlogit List] simulation

[Branka Valcic]

Greetings,

I am modeling fishing location choices using an HEV model. There are 30 
fishing areas (alternatives) that each fisherman can choose to go to 
fish to. The model runs well and its predicted probabilities are good 
too. Now, I want to simulate some closures, in particular I want to 
close 10 areas to fishing. I have used the Simulation command and have 
omitted from it the 10 alternatives assuming that the simulation will 
thus not use information from those 10 alternatives. My Simulation code 
is as follows:

NLOGIT;
Lhs=CHOICE;
Choices=4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33;
Rhs=DIST10km,EREV00,NEIGHBOR,BOATSSCA,TEXPDUM,EXPSCALE,STDEV00,IEXP_TRI;
HET;
Simulation=4,6,7,9,10,12,13,15,16,18,19,21,22,24,25,27,28,30,31,33;
Scenario: BOATSSCA(*)=[*]1$

The base per-area shares and % shares in the simulation output did not 
show the 10 omitted areas, which was as as I was hoping, but the output 
wrote (see below) that the scenario specification affected (among other 
alternatives) alternative 5. But alternative 5 was omitted from the 
choice set, and as such its attributes (expected revenue etc.) should 
not have been included in the simulation either. Or so I thought.
-------------------------------------------------------------------------
Specification of scenario 1 is:
Attribute  Alternatives affected            Change type             Value
---------  -------------------------------  ------------------- ---------
BOATSSCA   4        *5   *     6        more  Scale base by value     1.000
-------------------------------------------------------------------------

Basically, I thought the simulation worked as follows: The predicted 
probabilities are recalculated using the formula for the HEV probability 
of individual n choosing alternative i, plugging in it the attribute 
data for those alternatives in the simulated choice set and the 
estimated coefficients from the original model. Hmm, am I wrong on this?

Since the base % shares are so off the actual ones (I do have the data 
with the closures and can compare the simulation and the actual case) 
even though the original model is really good, I am worried that the 
attribute information on the 10 omitted alternatives is still being used 
and that this is causing the disparity. I would simulate all attributes 
to be zero in all closed areas, but since there are 10 closed areas and 
8 explanatory variables, this exceeds by far the number of possible 
attribute specifications in the Scenario command.

Can somebody, please, help me?

Thank you so much in advance,

Branka Valcic

------------------------
Branka Valcic
Assistant Professor
Economics Program
University of Alaska Fairbanks

PO Box 756080
Fairbanks, Alaska 99775-6080

Phone: (907) 474 2754
Fax: (907) 474 5219
URL: www.faculty.uaf.edu/ffbv
-------------------------