[EM] Deterministic Districting

[Mike]

Dr.Ernie Prabhakar wrote:

>
> This brings us back to the question of automated redistricting. We've 
> often discussed how the 'fairest' algorithm would use a measure such 
> as  "minimizing lanes of traffic cut by the circumference" by 
> combining census tracts while ensuring equal-population districts.   
> Its easy to get an approximate answer from this criteria using various 
> statistical means, but is there any computationally feasible way to 
> get a *deterministic" answer that is at least near-optimal?  
> Otherwise, I fear that people will either create pseudo-statistical 
> results, or challenge truly random results due to their fear of 
> (random) biases.


Well, it's not really deterministic (in the sense that the results are 
repeatable), but one could could put the districting maps on the ballot 
along with the candidate. Each candidate could provide their own 
districting map (or use their party's map), and have the voters decide 
which one they liked the best. They could then use that map in the 
election to count the votes for various offices.

Of course, there are some potential drawbacks. A candidate might not 
know exactly where he should be campaigning (that auditorium that so 
warmly received him might end up in someone else's district), he might 
not live in the district he represents, and it's even conceivable that a 
person could win in more than one district. I like the fact that it puts 
a bit of uncertainty in an incumbant's mind, though one could always 
have the map go into effect for the *next* election, which would remove 
such concerns.

If you preferred an automated, repeatable method, you might look up 
Voronoi diagrams and Dirichlet tessellations, which are pretty useful 
for breaking a map into simple, compact spaces.

Mike Rouse
mrouse1 at mrouse.com