I knocked doors for an independent redistricting commission. I wrote a explainer about it that got passed around. When it passed I was genuinely happy, and then I spent two cycles looking at the maps it produced and I have to report that the effect on outcomes was much smaller than I told people it would be.
Not zero. Smaller. And smaller in a specific way that I think matters more than the size.
Here is what I expected: partisan mapmakers were producing maps with a systematic tilt, so removing the mapmakers would remove the tilt. Straightforward, and it is roughly what the literature said, and it is what I said.
Here is what happened: the commission's map has fewer of the shapes that make people angry — the long tendrils, the county cut eleven ways — and produces a partisan outcome close to what the previous map produced. Fewer outliers, similar aggregate.
The reason, as best I can reconstruct it, is that the binding constraint was never the mapmaker. It is that voters are geographically sorted to a degree that makes a large class of maps produce similar results. If one party's voters are packed into dense areas at very high concentrations and the other party's are spread across a much larger area at moderate concentrations, then any districting rule that values compactness and contiguity — which every commission's rules do, because those are the rules people can agree on — will produce something with a tilt in it. Not because anybody drew it that way. Because that is what compact districts do to that distribution.
I could have known this beforehand. Some people did know it and said so, and I read them and thought they were being defeatist, which is a way of not engaging with an argument.
The counter to my own disappointment, which I have come around to, is that outcomes were the wrong scoreboard and I chose it myself. The commission did produce things: a process that can be watched, maps whose criteria are stated in advance, an end to the specific practice of a legislature drawing its own seats behind a closed door, and — this one is not small — a mechanism that no longer requires trusting whoever won the last election. Those are legitimacy goods. They were in the campaign material. I emphasised the seat count because the seat count is what moved people, and then I judged the result by the thing I had used as a lure.
The narrower lesson I have taken is about my own forecasting rather than about commissions. When I predicted a large effect, I was reasoning from the mechanism I could see — a mapmaker with an intent — and not from the distribution the mechanism was operating on. That is a general failure mode and I do not think I have fixed it. I think I have only learned to notice when a story has a villain in it, because a story with a villain is a story where I have probably located the variable in the person rather than in the terrain.
