[EM] Re: Automated districting

[Ernie Prabhakar]

Hi Matt,

On Jan 10, 2004, at 10:37 AM, matt at tidalwave.net wrote:
> For this to be a generally applicable approach I think we need a 
> reasonably objective procedure for dividing large municipalities into 
> smaller units and some reasonably objective procedure for identifying 
> when a municipality should be divided into smaller units.

So, why not always choose the generic unit "census tracts"?   Does 
anyone know exactly how those are defined?  They should at any rate be 
small enough to be immune from gerrymandering, but easier to manage 
than block-level data (which I don't think is always well-defined, 
anyway).

> That makes more sense than .... road traffic and bandwith, which are 
> poor measures of community and are manipulable

Um, I don't see either of those claims as true, so I wonder if we're 
talking about the same thing.

> poor measures of community
My idea of "community" is that, all else being equal, we should group 
people into the same district who have strong interactions with each 
other.   In terms of which neighbors I hang out with, where I shop, and 
the relation between home and work, the connections are very strongly 
correlated with the road network.   I talk more with neighbors on my 
street than those in a different subdivision, even if that subdivision 
is physically closer but doesn't have a connecting road.   I also have 
more connection with the stores 20 miles down the Interstate than those 
5 miles away down a country road.

Could you elaborate on why you don't think road traffic is a good 
measure of community?  For what definition of community?

> are manipulable
I don't see how road traffic is "manipulable" in a political sense.    
First of all, most traffic decisions are made locally, but the 
calculations are done globally.   It would be extremely difficult to 
predict how a given road-building project would affect congressional 
groupings.   Plus, road-building issues are *extremely* democratic at 
the local level; neighbordhood groups get extremely worked up about 
them, and if there was even a hint that the road was being built for 
political purposes rather than serving the community it would blow up 
in their face.

The only area where I could imagine politics playing a role is in 
Highways, since those are influenced at the state level.    However, in 
most areas the highway pattern is already well-established, and new 
construction is a relatively small fraction of the total traffic.   New 
highways are also highly visible and contentious issues; people may not 
notice when districts are redrawn, but they certainly notice when a 
highway is built (or not built) near them.    Large-scale manipulation 
-- such as building a highway where one isn't needed -- would be very 
difficult to accomplish.   Small scale deviations are indeed possible, 
but unlikely to affect the actual results.

In fact, given that edge-weighting algorithms would generally be N! 
factorial, then with small enough districts there is no computationally 
tractable deterministic solution.   That means there is no reliable way 
to predict how a new road would actually affect the relevant districts. 
   Don't forget, gerrymandering relies on precise control over the edges 
and contours of a district.  As long as districts are always 
recalculated from scratch every time, rather than just tweaked based on 
new data, it shouldn't be possible to game the system with any degree 
of certainty.

Don't forget - while we may worry about things like minimize variance, 
I suspect most people would want districts that 'look like how I view 
my community', so we need to take that into account.   I believe we 
should pick the algorithm most likely to win public acceptance, as long 
as it makes gerrymandering infeasible in practice.

-- Ernie P. 
On Jan 10, 2004, at 10:37 AM, matt at tidalwave.net wrote:

> Toplak Jurij wrote"
>
> Matemathically there are many ways we can arrange these 50 
> municipalities in 4 districts, but there is only one under which the 
> population variance is smallest possible.  This procedure does not 
> involve decisions of human factor, except for =
> the decision on the procedure used and political units used. However, 
> the decision on which procedure should be used will always be taken by 
> humans.  It could be said that it was humans who drew the borders of 
> the municipalities (or other political units used), but these borders 
> have usually been drawn long ago and without intent of gerrymandering.
>
> Matt replies:
>
> Thank you.  That makes more sense than my inappropriate integer 
> programming for perimeter compactness suggestion, which at best isn't 
> practical and is indifferent to community, or road traffic and 
> bandwith, which are poor measures of community and are manipulable.  
> Since there is a best solution we cannot argue about method bias.  
> Indeed, it seems to address most of Micah Altman's criticisms (or at 
> least those he mentioned in the several dozen pages of his 
> dissertation that I read).  However, I do see a problem.  What if the 
> largest municipality is larger than all of the other municipalities 
> combined?  For this to be a generally applicable approach I think we 
> need a reasonably objective procedure for dividing large 
> municipalities into smaller units and some reasonably objective 
> procedure for identifying when a municipality should be divided into 
> smaller units.
> ----
> Election-methods mailing list - see http://electorama.com/em for list 
> info