Algorithmic Redistricting: Elections made-to-order

Man the VA Dems didn't even try to be subtle

I wonder if you can make maps look fair by using how convex the shapes are, taking the points on the border, drawing lines between them, and seeing if the lines are completely on the inside of said shapes. The percentage of lines completely inside the shape can be used as a score for how "convex" the shape is. The one problem I see with this method is that it probably would take a good amount of computing power because it would have to do some edge detection and computing the lines would be intensive. It may also fall into the Perimeter to Area trap, as it may take a few iterations for the ratio to actually improve.

Well, thanks for destroying America

Dang bro ur content is sick 🔥. How did u get this level of coding and math modeling experience?

I've been a fan for a long time, and I was super curious when I stumbled upon this video that "approaches politics". You did a really tasteful job of sharing solid and unbiased information on a touchy subject.

Thanks for sharing this with everyone, and taking the risk of your comment section devolving to cavemen.

12:22 does this number include districting that just looks like literal noise--each pixel is a random district, and not necessarily related to its neighbors?

bro made a video abt politics and added nothing of value or substance.

question: is it possible to make 13 swing districts? I want chaos

The fairest would be no districts at all. The popular vote should be what counts with a ranked choice system. This is how you solve two parties getting all the votes.

How much money has Elon Musk offered you for this code? 🤔

Watching this years later, I LOVE the headphone untangling method and joke. Thank you

Hey, I heard that Hyrule Castle Theme in the intro!

I think "desired election performance" is a better descriptor

i mean at this point it isnt even gerrymandering? its just each individual district voting for x party and the total statewide results not matching the popular vote. and its not even a fptp problem, its just that at the end of the day 1 person has to win the local election.

in a system where people vote for representatives, all of the "gerrymandered" maps you showed are actually fair because they do reflect the politics of the people inside the districts and the districts are sensible

I wonder why that look like Israel country.

How far can it go if the algorithm doesn't care to make the maps looks reasonable?

I wonder if something could be done that sort of fixes the electoral college nationally so that the way the electoral college votes always approximates the popular vote as closely as possible, and it's impossible for somebody to lose the popular vote and win the electoral vote.

It should almost be that you do the state vote, to see how many seats go to each party first. Then you have how many seats eat should have, then you use a program like yours to make a map where as many people are represented by their local seat as possible. The map would happen after the election.. so political maps during the election would be less interesting... but at least it would actually be fair.

As I see it there are two core evils with gerrymandering: First of all changing the outcome of an election so it no longer is democratic but maybe equally bad is encouraging balkanization by locking in districts to one part or the other. How can you use the algorithm to prevent both those issues. Looking at your maps it feels like the algorithm may naturally cluster the cities into cities again contributing to a rural/urban balkanisation?

Oregon has to be the worst Gerrymandered State of the Union. Criminal theft of voter representation.

I know nothing about statistics and I thought this was an awesome video and I learned a ton! Thanks for making it

I just realized this was 4 years ago. Can you rerun?

AlphaPhoenix just gotta assert his dominance

You know in germany we have a two vote system, where you first vote for a party and then a candidate, the party getting seats in respect to their vote share and the candidate directly elected, that alone would pretty much fix the biggest issues of this US electoral system.

My interest in data science and the recent gerrymandering headlines brought me here. Kudos algorithm

insanely interresting stuff you did here. It's a problem that is almost complitely exclusiv to America, one of the solution would be multiple member districts with some type of proportional representation. Very unlikely to happen sadly.

Now consider something like city or county borders to try to maintain communities within precincts as much as possible, to make actually administering those elections possible. Algorithms such as you have developed are probably how political majorities would initiate their map creation, then they would apply some creative manipulation to "try" to adjust the maps to make more sense to voters and election administrators while still retaining as much political advantage as possible.

It is sad that something like the modern statistic analysis potential made possible by AI is capable of enhancing the quality of life but is immediately weaponized and turned against the people to exploit them. If technology could only be tamed!

I am just guessing but your map drawing abilities might be much more financially rewarding than explaining science on YouTube. Are you willing to move to Texas or California?

This is amazing, now do IL.

Dude! Awesome! Last thing to do is make sure you don't draw lines across apartment complexes or literally break up a small neighborhood. It should be a sellable product.

There is no ideal map. No matter what you draw, there's a minority in each district who's vote doesn't count. The only way to represent as many minorities as possible is some variant of rank choice voting for a pool of candidates. So the handful that want Value A can vote for the Value A candidate no matter where they live in the state, and if there are not enough Value A voters then their votes fall to Value B, C, and D candidates. But its nearly impossible to get this because its against the two party's interests.

As interesting as this is, I think my problem with it is the randomness. I do think that algorithmic redistricting is probably the best solution to gerrymandering. We have alot of debates to hold over exact goals and parameters, but the goal should be to have an open source algorithm that if run with the same data and configs always produces the same outcome. That way you know a map was run under those inputs, and not secretly manipulated.

However, I do like alot of the other features of your algorithm and wonder if a clever guy like yourself could figure out how to do them without randomness. Also, one of the objections to algorithmic redistricting in general is that it doesn't respect 'traditional' or 'natural' boundaries. Mind you, it's a fake objection because the new partisan gerrymandered maps don't either.

But I can imagine an algorithm that starts with an initial state where county lines, major highways, city/town/neighborhood boundaries, rivers, etc subdivide the map much more granularly. Then, in order to get to the 'right' number of districts, it starts combining granular pieces that border each other with some priority that tries to prefer to respect varying kinds of boundaries, achieve a linear partisanship curve like you used, and seeks compact shapes, and similar population sizes. I think operating at the 'pixel' level is just too granular.

But I'd love to see you revisit this topic as it really is an INCREDIBLY important topic. The rise of computer aided gerrymandering during the 2000 redistricting cycle is, I believe, a MAJOR factor in what broke our politics. Not the only, but top three for sure. Prior to that, people working with manual maps could only do so much damage with gerrymandering. You claim politicians are bad at gerrymandering. To the contrary, they are outsourcing it to IT consultants already.

But there's a limitation in terms of how much they believe they can get away with in court. That's the current limiting factor. They're WELL aware that they could stuff ALL of their opponents into one district and win the others comfortably. But as reluctant as SCOTUS has been on gerrymandering, state courts have aggressively replaced gerrymandered maps....with their own gerrymandered maps favoring their party instead of the one controlling the legislature. That's why I don't believe judges are the answer. You may get good ones some of the time, but some courts are every bit as craven as legislators.

hi

can you use this tool in reverse, feed in political redistricting maps, and get the expected outcome/score of what they did? so people can know just how gerrymandered and where the new voted outcome maps will become?

Great video! I'm assuming this or something similar to it is basically what both parties are using right now in the current fight over control of the house in the 2026 midterms. Maybe it would be worth it for states to shift to the use of a tool like this, under the purview of a transparent and independent districting board, to get politicians out of the re-districting process entirely in every state. Wishful thinking I suppose, but it sounds nice. Maybe then the parties would be forced to start representing the entirety of their constituents instead of concentrating on just a few swing districts and leaving all of the others hanging out to dry. Would potentially reduce the influence of money in political campaigns, as well, as the money would have to spread farther and wider.

You have a lot more pixels than the actual number of census districts in NC (2195), if you require contiguous districts, there are even less than 2195^14 maps to consider. You can't have a political map with boundaries that aren't the boundaries of a census district. I love the headphone analogy for annealing, but you might want to use flaking to store your headphones. A super great video, thanx.

12:20 Your math here is wrong. What you should do is multiply by the 13 disticts for each pixel, giving 13^27921, or about 2.584*10^31102. This number is a little too big for me.

Democracy only works when no party has the majority.

I thought you were going to make a simulator to make a histogram of outcomes to test if a proposed map fell within the realm of feasibly random chanced without the intent to gerrymander or improbable to have achieved unless the intent was to gerrymander.

Behaving.