I am trying to create contiguous territories given polygon data (at the US
ZIP Code level) and need some help ensuring compactness of the territories
I am using the SKATER algorithm implemented in the spdep package to achieve
this – this is a graph partition algorithm on a minimum spanning tree.
I can use the lat / long as one of the variables and this algorithm will
try to minimize the variance of the variables passed (in this case lat /
long) for each territory therefore trying to force compactness in my
solution. The problem is that if the minimum spanning tree has very large
chains with vertices of degree = 2 then there is nothing you can do to
avoid non-compact territories.
This representation shows (--*--*--*--*--*--*--*--*) there is nothing you
can do to avoid ending with a territory in a very long straight line.
My question revolves around adjusting the tree created to avoid as much as
possible the creation of such large chains and favor the creation of a tree
(with slightly higher cost than the minimum spanning tree) but with less of
these chains of nodes with degree = 2.