Adam Tarr atarr at purdue.edu
Wed Aug 25 14:28:46 PDT 2004

Dr. Ernie Prabhakar wrote:

>>Basically, commute flows (or road bandwidth) was regarded as the best 
>>measure, as it encapsulates geography and demographics pretty well.
>Yeah, I remember that (lanes of traffic).   I even started writing some 
>Python code to do the calculation.  The tricky part is that it was 
>assuming even sized populations.  I guess in this case, we'd need to 
>simultaneously minimize:
>         a) traffic circumference
>         b) deviations from the 'optimal' size (N=7)
>         c) non-integral population units
>That is, there needs to be some way to compare the goodness, e.g.,  two 
>districts of size 9.95 vs. three of size 6.6.   Which is straightforward, 
>but somewhat arbitrary.   Still, it could be done, and in some ways more 
>parameters generates more interesting options.

Well, I would make it a requirement that population-per-seat be equal 
across disctricts, plus or minus an extremely small margin owing to the 
granularity of census blocks or zip codes or whatever is used as the 
"atoms" of district formation.  So that takes (c) out of the criteria.

As far as deviation from 'optimal' size goes, it might make sense to simply 
require a certain total number of districts, and further require a maximum 
and minimum district size.  I'd suggest 4 and 10.  My intuition is that 
this would cause the algorithm to spit out geographically large rural 
districts with few seats, and geographically small urban districts with 
many seats.  This seems good to me.

So, that leaves only severed traffic connections as a metric to be 
optimized on.  Of course the stipulation that districts can jump in numbers 
of seats (swallowing or spitting out large areas in the process) makes 
optimization considerably more difficult.

More information about the Election-Methods mailing list