[Limdep Nlogit List] RUM NLOGIT Land-Use Change Modeling

Mike Mann mmann1123 at gmail.com
Thu Dec 16 03:03:05 EST 2010


I have few general questions about random utility models as they apply to
land use change in a NESTED LOGIT.

Here is a stylized version of the data.  But general idea is predicting
current land-use based on rents, slope (and other choice invariant
properties of the parcels), and previous land use.

U(i) = rent(i,j) + G(i) - C(i),  where i= parcel and j = LU choice,
 G=choice invariant physical conditions (slope), C=conversion cost proxy
(previous land cover dummy variables).

Definitions-     ID: Point representing a parcel   LUse: Landuse options
 Choice: LU Choosen   Rent:$NPVpotential income  Slope: slope of parcel

ID      LUse          Rent       Slope          PrevLUse      Choice
1       Forest          20            2                 Forest           0
1       Cerrado        0              2                 Forest           0
1       Pasture        50            2                 Forest           1
1       Agricult        120           2                 Forest           0
1       Deforest       60             2                Forest            0

2       Forest          40             10              Agri                0
2       Cerrado        0               10              Agri                0
2       Pasture        80             10              Agri                0
2       Agricult        240           10               Agri                1
2       Deforest       10             10               Agri                0
.


Tree->     Human(pasture, agriculture, deforest), Natural(Forest, Cerrado)

*My general problem is that it is not clear to me which level of the nest
parameters should be estimated.*  In particular I am confused by the
implications of the fact that ID identifies a particular parcel, and all
variables are particular to that parcel (whether choice invariant or
variant).  Choices then reflect a decision made not for different plots,
with varying parcel charateristics, but for one plot with different land use
options (and therefore different rents).


Is it more appropriate to apply parcel-level choice variant and invariant
components at the leaf level?

U(past, agri, def, for, cer) = rent(i,j), slope(i), prevLU(i)


Or, (i,j) variables at leaf, and (i) variables at branch level?

U(past, agri, def, for, cer) = rent(ij)
U(Hum, Nat) = slope(i), prevLU(i)


Or all at branch level since choice reflects options for a particular plot
(i)?

U(Hum, Nat) = rent(ij), slope(i), prevLU(i)


Thank you for taking to time to look at this!   Any suggestions or comments
would be helpful!


Sincerely,
Mike


More information about the Limdep mailing list