The Scarymander Series: Part 2

Currently, gerrymandered districts are known for their ugly, weird appearances such as many of the ones I showed in my last blog post (which I recommend you read before this one). However, gerrymandered districts don’t have to look this way. A youtuber named AlphaPhoenix released a video two years ago demonstrating that you can gerrymander North Carolina more severely in Republicans favor than it already was. Moreover, he showed that you could do it while making every district look very compact. The process through which he does this is not particularly new, but it is very technical.

Essentially, the main algorithm at play here is a Monte Carlo simulated annealing. What this means in english is taking random maps, and adjusting by randomly nudging them tiny sections at a time. Then you judge if the random nudge was any good. If we like it we keep it, and if we didn’t like it we usually get rid of it (but sometimes we don’t because a little more randomness helps us optimize our maps). There is a lot more to process, which he explains in the video but that is the core. It sounds pretty simple but the results are terrifying.

Before getting into why this is so scary it is worth pointing out that this could not be used as is because it doesn’t comply with the VRA (see my previous post for details). That aside, this is a horrifying algorithm. By far the most common attack on gerrymandering is the shape of the districts, and this entirely removes that complaint. When the computer judges if a change it makes is good one of the features we can include is how compact the district is. To most people who have very little in depth knowledge of gerrymandering a district’s shape is the main part they look at. However as you can see in the gerrymandered maps above, they don’t need to be irregular with this new method.

Similarly, it makes gerrymanders more effective as it removes biased humans from the equation. Consider that in North Carolina Republicans were only able to draw a 10-3 majority while AlphaPhoenix drew an 11-2. What is crazy to me is that AlphaPhoenix’s data is very poor, only having data down to the precinct level while Republicans had access to far better data. If this was made by the Democratic or Republican Party it would be far more granular allowing them consider voter by voter level data. The consequence of this would be even more gerrymandered maps that look far more normal and control election outcomes far more.

How is the algorithm outperforming humans so much? When a person uses REDMAP, they look for blue areas and try to either crack them apart or pack them together. People are not particularly good at making tiny changes or differentiating between an area that is 66% Democratic or 75% Democratic so they are only so precise in how they drawing lines and seeing the results. Every change always affects at least two districts and they have a lot to consider. Moreover, they can only balance so many variables with their squishy human brains. Computers don’t have to do this. They can check the difference of moving a line an imperceptible amount to cleave population centers perfectly such that the district stays compact but maintains a desired partisan split.

Perhaps more importantly, a computer could be given parameters that are more complex than what humans are capable of. You could run another layer of Monte Carlo simulations on top of the mapping to try to make more resilient districts under various scenarios. If you have the maps examined under the conditions of countless different possible swings in the electorate there is the possibility to have gerrymanders do more later into the decade which has been a weakness of them in the past.

It is my belief that state parties could reasonably take advantage of this in 2030 and gerrymander state legislatures and Congressional districts even worse than they already have done. Project REDMAP only cost the GOP 30 million dollars, and this could potentially be far less costly seeing as one youtuber was able to do it on his own. In preparation for this there are two steps we need to take.

Firstly, we need to emphasize voter efficiency when talking about why gerrymandering is bad. Shape is not going to be a useful metric for analyzing gerrymandering in the future. This is a major shift in messaging and thinking so it is key to start adapting now. While voter efficiency may be a harder message to get across, if I am right it is the only metric we will be able to point to.

Secondly, it is even more dangerous than before for Democrats to unilaterally stop gerrymandering. That is to say if Democratic states install nonpartisan or bipartisan methods of drawing districts and Republicans don’t we will suffer grievous consequences. For instance, as I noted in my last post if California was gerrymandered we likely would have won the house this cycle.

We can’t afford to leave seats on the table. If the federal government was in a position to ban gerrymandering that would be one thing. However, given the filibuster and the fact that most House members directly benefit from gerrymandering it is hard to imagine anything moving on that front anytime soon. Rather we need to fight back with everything we have to win the types of majorities to make this fair nationwide. This means fighting fire with fire. In the words of the evil Margaret Thatcher, “There is no alternative“.