Gerrymandering, the practice of stringing together voter precincts in convoluted ways to create a district that favors an incumbent political party, is nearly as old as voting in America. (The term traces to 1812, when then Massachusetts Governor Elbridge Gerry made a salamander-shaped district to perpetuate a hold on government by his Democratic-Republicans.) But just because an idea is long-tolerated in politics doesn’t make it good for democracy. Gerrymandering weakens the value of one-voter, one-vote, suppresses voter turnout because election outcomes are predictable, and reduces the incentive of parties to govern according to the will of their constituents.
But Puget Sound computer science professor Randy Bentson and Walker Lindley ’08 may have a solution. The teacher-student team is closing in on a way to draw fair, evenly packed voter precincts by using a technique called a genetic algorithm. Here’s how it works.
Fairly drawn voter districts should have two characteristics: equal apportionment and compactness. Equal apportionment means that the number of voters in each district should be about the same. That’s easy to figure out by simply counting. Compactness relates to consistent physical shape, and that is much harder to measure. Bentson and Lindley found a formula for this, though, which they call the Gerrymander Factor, or GF. Using this measure, oddly shaped districts like Governor Gerry’s salamander get a high number, more than 80. Nice, evenly shaped ones—a circle is the theoretical optimum—get a much lower number. (A circle is 12.56.) But you can’t make a good map from a grid of circles because, when placed side-by-side, circles only touch at one point, leaving gaps. Polygons, which can be pieced together like tile, with all boundaries touching, are the best that can be hoped for.
So how do you figure out the best shapes for the polygons? Using a search technique first devised in the mid-1970s by John Holland, one of the pioneers in computer science, our professor and student assigned values to the boundaries of a set of test precincts, encoding the data in a series of 0s and 1s like a DNA sequence. The technique is called a genetic algorithm because it works a lot like a population of living things evolving over millions of generations of natural selection—it tries out possibilities over and over, keeping and combining good “fitness” values and allowing not-so-good ones to go extinct. The longer the program runs, the better the solution becomes, and sure enough, the test data Bentson and Lindley ran on a computer for several days produced districts with very low GFs.
Now they’re ready to try it on the voter precincts of King County. This presents a new set of challenges, since physical features such as lakes and rivers interfere with the tiling of precincts. But the researchers have come up with a way to compensate: By assigning artificial zones with no population to bodies of water, precincts can be tied together across such obstructions.
Bentson says there are other problems to work out, too, but stay tuned. Sometime soon, computers may offer a way to draw voter districts that removes the opportunity for human mischief—if only the politicians adopt it.
— Chuck Luce