Every hates -- or says they hate -- partisan redistricting. So why not try its opposite? Random districting.

Xavier Marquez, a political theorist from New Zealand, has a fascinating (and never-gonna-happen) proposal to do just that. Here's how it would work: Suppose a congressman wins a safe Democratic seat in, say, Massachusetts. If that congressman wants to be reelected, under this system he has to win an election in another district selected at random. That district could be a Republican seat in Mississippi, or a swing seat in Ohio, or another Democratic seat in Hawaii. A month or so before the election, the random placements would be revealed, and legislators would have to justify their records to a new constituency.

Why? Systems like the one currently in place in the United States and most countries provide an incentive for legislators to advance the interests of their particular constituency. But that means that they're often responsive to the interests of specific industries in their districts, resulting in policy outcomes that reflect parochial interests rather than the national interest. Presumably, if legislators didn't know where they were going to have to run for reelection, they would appeal to the median national voter, rather than their particular district. As Sarah noted in response to a similar idea back in April, there's a long history of randomization in legislatures, dating back to ancient Greece.

The drawbacks are obvious. The lack of responsiveness of local special interests would come with a lack of responsiveness to local interests, period. Congressmen wouldn't be as motivated to do constituent service, or to fund bread-and-butter infrastructure projects for their districts, like roads and bridges. The plan would likely be unconstitutional. But as a thought experiment, would randomization actually change Congress's ideology?

Perhaps not. If you just shuffled existing members into different districts the average ideology of Congress wouldn't change, assuming those members' ideology doesn't shift due to the policy change (a big assumption, but any other one seems speculative). The median member would still be the median member, and the average change that a given district would experience in their representative's ideology would be zero, since any change in one district is canceled out in another. But individual districts could still experience significant shifts in ideology, even if Congress as a whole didn't.

Or think about it this way: Suppose that, randomly, every safe seat Congressman was swapped with a representative from a swing district. Even though, if you averaged all the swings, nothing changed, every single district would experience a huge swing. To see what a typical district's swing would be, I created a simulation that randomly shuffled Congressmen to different districts. It then compared the ideology of each district's new Congressman -- as measured by DW-NOMINATE, the standard political science metric for ideology, to the ideology of the Congressman being replaced. I computed both the average left-right change, and the average absolute change (that is, how much a district's representation would change, in whichever direction). After running the simulation a number of times (100,000, to be precise) I averaged out the results to cancel out differences in the different random swappings.

Unsurprisingly, districts don't get more liberal or more conservative on average. Under DW-NOMINATE, the most conservative score is 1, and the most liberal score -1. I found a typical district gets a representative with a score 0.0023 points higher, or more conservative. That's a really small amount — it's about 100 times smaller than the difference between Nancy Pelosi and George Miller, her closest ally in Congress.

The really interesting results are the absolute changes. The typical district sees a swing of 0.485 in one direction or another. In the last Congress, the typical Democrat had a score of -0.366 and the typical Republican a score of 0.60. The most conservative Democrat, Walter Minnick of Idaho, had a score of 0.082, and the most liberal Republican, Charles Djou of Hawaii, had a score of 0.215. So a swing of 0.485 is more than enough to swap a moderate for a more liberal/conservative member (or vice versa). It's certainly enough for seats to flip parties. In short: randomization would mean most Congressional districts would see serious ideological change. But how frequently would that happen? Here's how the probabilities look:

The median district only sees about a 0.485 point change. But the biggest groupings are around very minimal change, and extremely dramatic change of around 1 point. This makes sense: the parties are about 1 point apart in scores, on average, so a seat flipping parties would see its score change by 1 point, and one that doesn't flip usually sees a minimal change.

Sure, this is a highly theoretical exercise. But it's a reminder of how regional even Congressional-level politics are. If you take region out of it, everything gets shaken up dramatically.