[EM] Re: Automated districting

[Alex Small]

I'd like to jump into a long thread with a few thoughts:

If I were in charge of redistricting for a state, I'd proceed from the
large scale to the small scale, starting with the largest natural
"communities of interest" and progressing to smaller units.

In the first stage I'd identify large "communities of interest" (probably
counties) and aggregate them into geographically contiguous units so that
each aggregate has a population corresponding to an integral number of
districts (within some tolerance specified in advance).  I'd draw, say,
two or three different possible maps, and pick whichever one has the
smallest total perimeter (summing the perimeters of each district).

Next I'd look at smaller communities of interest within each aggregate,
perhaps cities or some other municipal division.  Once again, identify
geographically contiguous aggregates of communities that have populations
corresponding to an integral number of districts.  Draw two or three
different possible maps that satisfy the same population constraints, and
pick whichever has the smallest total perimeter.

Continue this process until we have single districts with equal
populations within some specified tolerance.


The variables are:

1)  The population tolerance.
2)  The number of configurations to try at each level of refinement.  The
more configurations you try, the more likely you are to find a case with
minimum total perimeter.

This keeps districts "logical", in the sense of corresponding to natural
divisions within the state, and also keeps them from assuming fractal
shapes (often a sign of gerrymandering) by minimizing total perimeter. 
The very act of starting on a large scale and only going to smaller scales
as needed also serves to keep the district shapes relatively innocuous.

Anyway, just a thought.



Alex