Friday, October 08, 2004
A Solution for Gerrymandering?
Mickey Kaus made a post today that gave me an idea. Gerrymandering is an old problem, and it continues to be a huge problem. Redistricting is necessary from time to time for ensuring that district boundaries are drawn fairly with respect to the changing concentration of the population, but the problem is that the people who do the redistricting are the politicians affected by it. It's been said that gerrymandering lets politicians choose their voters, instead of letting voters choose their politicians.
What if we could come up with an algorithm that creates district boundaries, and all states were required by law to use that algorithm for creating districts? That is, what if we could create an algorithm that takes as input the geographic shape of the state and information on current voter densities and outputs a set of district boundaries? That would take the control entirely out of the hands of any partisan force and place it instead in the hands of mathematics and the voters (by virtue of where they choose to live). Notice that it doesn't take in any information on what party the voters are registered with. To borrow from John Rawls, it would be a set of district boundaries that everyone could agree to if they didn't know what party they were in, only it works by not knowing the party registration of the voters. In essence, a tangible, real-life veil of ignorance.
Of course, there are difficulties. When I took data structures and algorithms, one of the projects involved implementing several different algorithms for taking a group of points distributed on the x,y plane, and clustering them into n groups. Constructing the optimal clustering is not computationally feasible (with all the footnotes that go with that), so these algorithms were mostly based on heuristics and didn't always generate good results (and they could still take a long time to run).
I imagine with some research, it should be possible to construct an algorithm that has desirable properties. For example, it should create districts of roughly equal size in terms of population, connected, non-overlapping, not too concave, and it should try to make district boundaries go through sparsely populated areas. It is essential that it not require any "tuning parameters" or human input (except in the case that it needs firmly set parameters to be tuned once and then used everywhere). That's too many goals, and they can be contrary to each other, I know. Still, the clustering doesn't have to be optimal, it just has to be unbiased and better than what currently happens. Ideally, the key properties and performance boundaries of the algorithm should be proven mathematically, so everyone can be comfortable with it. With some work, I think something promising could be done.
However, I think even if it is possible, it won't be done for the obvious reasons: it requires the people who are affected by it to agree to it, which is the original problem. It's too bad, it would be an interesting combination of law and technology.
Update: Macneil writes in to point out that this is actually an old idea. For example, see here. I did some google searches last night, but nothing came up. Now I'm finding all sorts of stuff, so I must have been using the wrong search terms.
What if we could come up with an algorithm that creates district boundaries, and all states were required by law to use that algorithm for creating districts? That is, what if we could create an algorithm that takes as input the geographic shape of the state and information on current voter densities and outputs a set of district boundaries? That would take the control entirely out of the hands of any partisan force and place it instead in the hands of mathematics and the voters (by virtue of where they choose to live). Notice that it doesn't take in any information on what party the voters are registered with. To borrow from John Rawls, it would be a set of district boundaries that everyone could agree to if they didn't know what party they were in, only it works by not knowing the party registration of the voters. In essence, a tangible, real-life veil of ignorance.
Of course, there are difficulties. When I took data structures and algorithms, one of the projects involved implementing several different algorithms for taking a group of points distributed on the x,y plane, and clustering them into n groups. Constructing the optimal clustering is not computationally feasible (with all the footnotes that go with that), so these algorithms were mostly based on heuristics and didn't always generate good results (and they could still take a long time to run).
I imagine with some research, it should be possible to construct an algorithm that has desirable properties. For example, it should create districts of roughly equal size in terms of population, connected, non-overlapping, not too concave, and it should try to make district boundaries go through sparsely populated areas. It is essential that it not require any "tuning parameters" or human input (except in the case that it needs firmly set parameters to be tuned once and then used everywhere). That's too many goals, and they can be contrary to each other, I know. Still, the clustering doesn't have to be optimal, it just has to be unbiased and better than what currently happens. Ideally, the key properties and performance boundaries of the algorithm should be proven mathematically, so everyone can be comfortable with it. With some work, I think something promising could be done.
However, I think even if it is possible, it won't be done for the obvious reasons: it requires the people who are affected by it to agree to it, which is the original problem. It's too bad, it would be an interesting combination of law and technology.
Update: Macneil writes in to point out that this is actually an old idea. For example, see here. I did some google searches last night, but nothing came up. Now I'm finding all sorts of stuff, so I must have been using the wrong search terms.
Comments:
You could break up the lowest resolution at the zip-code level to both make the algorithm more efficient (10 days on a supercluster could be enough time for every state) and the boundaries a little more grokable.
Though, it would leave politicians some fooling with zip-codes, but any solution they'd want something that would look like a backdoor anyway.
My problem is with the EC, how it's essentially gerrymandering based on states (though state boundaries do not change, so it's not really gerrymandering, but it's the same kind of problem).
Post a Comment
Though, it would leave politicians some fooling with zip-codes, but any solution they'd want something that would look like a backdoor anyway.
My problem is with the EC, how it's essentially gerrymandering based on states (though state boundaries do not change, so it's not really gerrymandering, but it's the same kind of problem).
