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

[Juho]

On Nov 19, 2009, at 6:55 PM, Raph Frank wrote:

> On Thu, Nov 19, 2009 at 4:00 PM, Juho <juho4880 at yahoo.co.uk> 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.  
>> Following
>> Raph Franks model it would be thus enough to say that any N points  
>> and the
>> kn values and then derive the border lines and the jointly agreed  
>> value of
>> the solution from this data. That would not leave much space for  
>> strategies
>> and gerrymandering. The proposed solutions would be evaluated and  
>> the one
>> with best value would be declared the winner.
>
> Well, ideally the method should be a well defined process rather than
> an optimisation method.

This is true from the point of view that it would be good to have a  
known algorithm that can automatically (and in reasonable time) find  
the best result. On the other hand the optimization approach is a  
superset of the procedural approach. The optimization approach works  
also in the case where finding the best answer (or proving that some  
answer is the best) is not computationally feasible (but when  
optimization can find good enough answers).

It may also be easier to define and agree just the targets /  
criterion. I also like the idea of defining the ideal outcome instead  
of defining a procedure (that might or might not yield a good result)  
(the agreed criterion is closer to defining the actual targets).

>  It would take as its input a set of points
> and output a map.  Splitline also requires a description of the State
> boundary.

I think the optimization approach that I proposed would as well  
require very similar data, except that the points could be picked at  
random and not given as input.

>
> However, it would be perfectly valid to give a measure and then allow
> anyone submit a map districting.

Yes, for splitlines one could e.g. just set a requirement that the  
borders should be straight lines and there should be n-1 lines (to get  
n districts) (+ the border length and even population distribution  
requirements).

>
> I think that if the block boundaries are decided before the census and
> the number of blocks is large enough (say 100-300 people per block on
> average), then it would be hard to gerrymander using block boundaries.

Yes, some suitably small size should be set to reduce gerrymandering.

>
> The process could be something like
>
> - based on the old census, define the blocks for the new census
> -- A group of contiguous old blocks with population < 500 may be
> combined into a new block
> -- Old blocks may be split into pieces
> ---- (if > 1000*N, it must be split into N+1 parts)
> -- otherwise, the blocks shall remain the same as previously

Yes. In the case that they are already small enough there would not  
even be any interest to find "politically appropriate" blocks.

>
> - Geographic data is released
>
> - Hold census
>
> - Population data is released
>
> - Format for maps is published
>
> - Anyone can submit a map
>
> - best map after 6 months wins.

Yes. With computers (and if working sw is already available to all and  
all the rules are already well known) also shorter period should be  
enough.

>
> Ofc, that requires that the SC is able to determine which map wins
> based on the description of the measure in the legislation.

Yes, the comparison method should be well defined and feasible to  
compute. The optimization algorithms should be able call this  
subroutine thousands of times.

Juho