[EM] Another auto districting proposal (Crystal districting?)

[Brian Olson]

I agree with Juho. Define what a good redistricting result is, preferably in terms that produce a single valued numeric score, and then produce maps by whatever means you like and let the best map win.
I haven't seen a procedure defined that I was sure would always produce good districts. While many procedures have been defined where every step is fair, that doesn't entail that the result would be good. There's always some side effect of the method which results in odd unnaturally shaped districts.

My current favorite score function for district maps is*:
Minimize the average distance per person to the geographic center of their district.
(This is currently what I've solved for at http://bolson.org/dist/ )

But this also feels tantalizingly good and might be better:
Minimize the average distance per person to the population center of their district.

Unfortunately when laying down criteria like this, it's a more philosophical argument rather than something that can be measured and engineered.

I've seen lots of proposals that are purely geometric over the land encompassed by a district, but I think that is a mistake. We're not redistricting land. We're redistricting spatially distributed people.


*This doesn't include the _requirements_ which a map must first pass, otherwise it is simply thrown out. 1. Contiguous districts. 2. District populations must be within 0.5% of the average district population.

On Nov 19, 2009, at 11:00 AM, Juho wrote:

> My thinking is that it might be easier to agree about the targets rather than the whole procedure. The targets can be simpler to define.