[Limdep Nlogit List] simulation
Branka Valcic
ffbv at uaf.edu
Sun Jun 25 05:41:07 EST 2006
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
-------------------------
More information about the Limdep
mailing list