[Openspace] weights for obs that share a location

[Munroe, Darla K]

Hello all,

Here is a sticky question, and if I were a better programmer, I could tackle
this myself, but anyway...

I am conducting an hedonic analysis on residential sales prices to estimate
the benefit of a regional greenway system in NC.  I'm supposed to give
separate estimates for single-family and multi-family housing.

Here's the problem with multi-family - the way the data are recorded in the
parcel layer, multiple records share a location.  Obviously, units within
the same building should be considered "neighbors", as well as units in
proximate buildings (however the apparent spatial structure warrants).  The
problem is, GeoDa can't create weights for the shapefile, because they are
non-unique parcels.

Any suggestions?  The easiest solution would be to throw out all but one
observation per building, but that doesn't seem like the best solution.  Or,
the best solution would probably be to create a weights matrix by hand -
maybe by point distance, and then setting the obs where the distance is 0 to
1, and then finding other obs at a critical distance?

I saw something once where a 3-D weights matrix was specified for a real
estate model in Japan - not only what building you were in, but the floor of
the building (the lower the floor, the higher the price).

Anyway - I thought I'd share this problem, and see if anyone has encountered
this, and come up with creative solutions.

Best,
Darla Munroe