Inverse Dist SWM - why do nearby features receive a lesser weight than those far off?

Discussion created by chesnokovaa on Nov 6, 2012
Latest reply on Nov 9, 2012 by chesnokovaa

I generated a SWM for the state of PA (census tracts = polygons) using the inverse distance conceptualization with a 0 threshold (everyone is everyone's neighbor), but something funky is happening. When I look at the SWM transformed to dbf, weights don't make sense. Nearby, directly neighboring features, receive a MUCH lesser weight than those far off. E.g. looking at one area, it's immediate neighbor got a weight of 0.0009122, while it's much more distant neighbor (10 areas away) received a weight of 0.0187211 - that's a 20 fold difference!

In another instance, neighbors with the highest weight for a particular area are located at the opposite end of the state. The area in this instance is located on the SW corner, while its neighbors with the highest weights are located on the NW corner of the state.

I expected for there to be some variation due to row standardization (e.g. I expected the matrix to not be symmetrical, i.e. the weight between X & Y is not necessarily the same as weight between Y&X), but I'm confused by these weight/distance relationships.

Am I reading the file wrong? I assumed that in the resulting .dbf file "uniqueid" marked the id of the area, "nid" marked the id of the neighbor, and "weight" marked the relationship between them.

Any help much much appreciated!