October 2017

S M T W T F S
1234567
89 1011121314
151617 18192021
22232425262728
293031    

Style Credit

Expand Cut Tags

No cut tags
Thursday, February 16th, 2017 02:31 pm
As it does on a regular basis, the subject of gerrymandering has come up again. And, as always, I'm seeing people make the perfectly reasonable suggestion that we deal with it algorithmically. I'm all for that... until it is claimed that this would somehow make it non-political. And that's just bullshit. Dangerous bullshit.

The Gerry-Mander Edit.png

Districting is hard because it's very hard to define an obvious set of criteria by which to rate potential districts. We have some basic parameters set for the federal level: break each state into n districts, each containing roughly 700K people, and don't allow the districting to artificially limit the political power of racial minorities. Specifically, it wants to avoid "cracking" (breaking a group's voting power over many districts, so their votes are overwhelmed everywhere) or "packing" (lumping all their voting power into a small number of districts, giving them a few safe seats but still reducing their representation far under what it should be going by population).

The problem is, obviously, that these are very squishy guidelines. So can't we firm them up with some hard mathematical definitions, write up a segmentation algorithm, and let it do its absolutely objective magic? Sure! We just need to define some kind of scoring system to judge how good or bad a potential set of districts for a state is.

Here are some factors I can think of that such a system could use for its scoring:
* District shape -- the eponymous gerrymander was a point of satire because of how strung out it was. Keeping districts reasonably compact is usually a good thing.
* Geography -- we don't want a district that extends across a mountain range or other significant barrier, since the people on either side probably have little contact with each other and don't make sense as a single political unit.
* Road/train networks -- same as above.
* Racial composition
* Cultural composition
* Age composition
* Education level
* Religion
* Types of economic activity
* Economic ties to other parts of the country/world
* Climate
* Soil types
* Favorite NFL team
* Literally a billion other possible options

But which of these factors does that is best? In what ratios? I could certainly come up with a solution I like, but it wouldn't be perfect. There isn't a perfect solution to this problem. It's not the kind of problem where "correct" even has a meaning. Given a defined algorithm, math can give you a perfectly objective answer, but it can't choose the algorithm in the first place.

And that's where I find this talk gets really dangerous. It wants to pretend we live in a world with provably perfect solutions to messy human problems. If we just let some smart math/computer types work on it, they can fix everything, and save us from the dreaded specter of politics. But that would just be putting the imprimatur of unquestionable objectivity on yet another arbitrary decision. Governments based on that kind of thinking tend to get all great-leap-forwardy and mass-starvationy.

The real error in this thinking is that it assumes politics is a bad thing. It isn't. Being political isn't a bad thing. Politics just means the process by which we come to a decision when there are conflicting human desires. You have to accept that deciding on something as hopelessly complex as districting is, yes, going to be political. And that's okay!

Personally, my solution would be to set up a framework for an official, national algorithm, running against standardized data provided by the Census Bureau. Let the politicians fight over the definition of the algorithm, let them tweak it as much as they see fit. Just use the same algorithm for the entire nation and make its definition public. Would that process be political? Fuck yeah it would be! But it would be transparent and it wouldn't undermine faith in democracy. That is what is important here.
Saturday, February 18th, 2017 12:04 am (UTC)
Indeed, it's a test of success, not the algorithm you are proposing. However, if single precincts are taken as a unit, you have enough past election data to determine whether a lump of particular precincts is going to cause a significant efficiency gap or not.
Friday, February 17th, 2017 02:42 am (UTC)
Huh, that's a really good, thoughtful short essay. Would you post this on g+ so I could reshare it?
Friday, February 17th, 2017 11:21 pm (UTC)
Arg. See my comment in [livejournal.com profile] randomdreams's repost about the Iowa solution. It's crude, but reasonably effective.
Saturday, February 18th, 2017 08:12 pm (UTC)
Alternatively, switching to a proportional system would fix this remarkably well...