Show HN: Antimander – Optimize Congressional Districts with Genetic Algorithms | Hacker News Reader

Show HN: Antimander – Optimize Congressional Districts with Genetic Algorithms

antimander.org

68 pointsjoelS50 comments

"I cut, you choose" is really the only politically viable alternative to the current state of affairs.

https://slate.com/news-and-politics/2018/02/how-the-i-cut-yo...

It would require a more fundamental recognition of political parties than the constitution provides.

Gerrymandering is not a technological problem.

Why can't you just split elected representatives into two groups, regardless of stated political affiliation? One group then cuts, the other chooses.

Defections. Electeds will happily maximize their own personal power at someone else's expense. There are many examples of getting opposition support for inequity in exchange for safe seats.

People outside of (partisan) politics don't often get to see that the party orgs and the politicians are usually in conflict. More so on the left than on the right.

Interestingly enough, this works even for defectors. They still have to participate in a turn-based game against a competitor.

There is a whole field of study around efficient "cake cutting" algorithms, many of which contemplate and account for defectors.

https://en.wikipedia.org/wiki/Fair_cake-cutting

Unfortunately, given Duverger's law [0], first past the post electoral systems eventually resolve into two party systems.

Essentially, in a first past the post system, voting for any other candidate other than the one both most likely to win AND whose politics most closely aligns with your own is effectively a vote for the major party candidate who least aligns with your politics.

[0] https://en.wikipedia.org/wiki/Duverger%27s_law

edit for link.

I've been trying to dream up ways to use general purpose strategies like this to mitigate regulatory capture. Applicable to two party systems with no clear obvious third party to forge a balance of powers, a durable trilemma.

There are fair algorithms for multi-way division... but yeah, it's going to be real hard for a third party to break in when the two ruling parties can conspire to block them.

Ah that's very clever, like when my sister and I split a candy bar in grade school. It really worked then as it would now.

>Some interesting patterns emerge, such as how with you can have compact and republican-biased but not compact and democrat-biased maps for some numbers of cities. This is a direct result of one party being forced into a small geographic area (cities).

That's quite an interesting finding. I always understood that gerrymandeting is bad because the artificially winning parties get congressmen that are not representative of their districts. This guy uses "competitiveness" as a property worth optimizing for, achieving exactly that.

I've long advocated maximal competitiveness. Mostly to motivate voter participation.

I love how antimander illuminates the tradeoffs.

Along the lines of "preserving community", preserving continuity can also be a factor. On the presumption that voters don't want to have their congressional district designation changed every 10 years. For example. I know this is a consideration during redistricting.

Antimander is real progress. The cites are terrific. I'm delighted that I even learned some new things, like the Seats-Votes Curve.

I've spoken with the Dave Bradlee many times over the last 15 years. He created a redistricting app that got some national media attention last cycle. https://en.wikipedia.org/wiki/Dave%27s_Redistricting

Some of those talks were pretty bleak. I'm so glad so many more people have engaged with this issue.

This is really well done! Do you have analysis of what this would do nationwide or across multiple states? Optimizing for competitiveness is a metric I hadn’t considered but seems particularly valuable especially towards increasing voter turnout.

How is it valuable? Having a "red" or "blue" district means the congressman is somebody most people can get behind, "our guy in the parliament".

Being around 50% gives you a compromise that nobody trusts and nobody feels the stake in their country.

On Android and Chrome, I can't read the page. The graphic takes up half the screen and the text underneath is barely visible.

For me also.

I had to scroll all the way to the bottom to see if anyone could even read the site on mobile.

Gerrymandering is a danger to democracy and it is great that researchers shine light on the different trade-offs of districting criteria.

This website seems like a great demonstration of the fairness and competitiveness trade-offs.

I'd argue that every voting system that has gerrymandering as a possibly error mode is inherently dangerous. No amount of antimandering could ever be more than a temporal band-aid. There are many examples that successfully combine local representation with proportional outcome.

Interesting. GA's have some interesting use cases. We used a GA to optimize assignment of work crews to areas as part of a scheduling optimization, and achieved significant benefits

Steve

Seen similar benefits - many years ago applied GA's to schedule technicians to fix computers (based on location, part availability, skill set, etc) and also for manufacturing production lines (labor rules, parts, color changeovers, demand, margin)

As Congressional elections become more and more a national affair, why do we still have districts? Say a state gets 24 reps. Give me a ballot with all the candidates and let me mark the 24 I want. Yes the ballot will be more complicated, but this way we entirely avoid districting and the process is entirely fair (well until you consider states not being fairly drawn, but that’s one level up).

That of course defeats the purpose of smaller districts getting a voice and not being crushed by larger districts.

Yes. That’s a feature, not a bug. 10m people in a city should get 10x the say than 1m people in the surrounding rural land. Why would you ever want it to be different other than trying to skew elections towards giving rural residents disproportionate amounts of voting power?

Likely it means that big cities would get all the reps and the rest of the state gets nothing.

A better route would be proportional representation. You vote for a party-ranked list of candidates, and the top n% of the list gets into office based on the fraction of the vote the party received. Unlike single-member plurality voting, this makes third parties viable.

As that is where most of the people live, why is it wrong? Why does largely unpopulated land get a vote?

This. Why do voters living in rural areas get more weight to their votes? This is obviously more pronounced in the Senate, but clearly also a problem in the House.

Why stop there? Why do small states get the same number of Senators as large states? All the Senators should just be from California, Texas, New York, and Florida.

The small states getting the same number of senators as large states is partly a legacy of slavery. Interestingly, the constitution explicitly prohibits the 2-senators-per-state rule from being amended away (Article V).

Can you amend that prohibition?

Good question. If you want to get into the weeds, there are a couple articles worth looking at: http://law2.umkc.edu/faculty/PROJECTS/FTRIALS/CONLAW/unamend... https://scholarship.law.marquette.edu/cgi/viewcontent.cgi?ar...

The short answer is yes. The long answer is that the Supreme Court would have to get involved. And they are very reluctant to make any kind of new interpretation of the Constitution for something like this.

Any answer to the question would be a new interpretation, since the clause hasn’t previously been litigated.

But you can clearly amend the powers of the Senate, including abolishing all of them. And/or create a third house of Congress, without equal per-state representation.

I mean, it's fun to think about at a purely theoretical/academic level. But practically speaking, none of this could ever happen.

I wouldn’t be so sure; it would require getting states to vote against their narrow parochial interests, but it's not inconceivable that it could happen. It would probably take a crisis of legitimacy in which the disaffection of the underrepresented states was sufficient to threaten the viability of the union, though.

I've imagined a workaround where states would be de-stated if they fell below a threshold percentage of the total population or be required to split if they grew above an upper bound percentage. The only reason there are two Dakotas is to give the Republicans two extra senate seats (this is true, you can look it up).

> The only reason there are two Dakotas is to give the Republicans two extra senate seats

First of all, you have no idea what you're talking about. The two Dakota territories gained statehood in 1889, long before there was any kind of notion of the modern-day Republican party.

I know you are trying to be sarcastic but you are correct. There is zero reason for the Senate to work the way it does as it is very clearly an unfair system. Instead, Senators should be nationally elected: 100 spots and every voter votes on every one of them.

> 100 spots and every voter votes on every one of them

So then all the Senators should just be from California, Texas, New York, and Florida.

What would be the problem with that?

IMO, lifting the cap on the number of representatives we can have as a nation is the best way to fight this issue. Representatives are supposed to represent the people in their district. If each representative represented no-more than, say, 200k people, then gerrymandering would almost be impossible and the people would be better represented.

Yes, we'd have a LOT more representatives. But we also now have the technology to support such a thing.

> As a result [of the Reapportionment Act of 1929], the average size of a congressional district has tripled in size—from 210,328 inhabitants based on the 1910 Census, to 710,767 according to the 2010 Census. https://en.wikipedia.org/wiki/Reapportionment_Act_of_1929

That would help somewhat but all it would do is require more computational power to draw imperfect districts. The fundamentals of gerrymandering do not change with doubling the number of districts. And my proposal is based on the idea that today House reps do not represent their district. Look at people like Pelosi and AOC and Nunes. They represent voters from across the whole country but are only elected in their districts. The debates we have are also largely national, not district-level. Most reps spend little time in their districts. They are national level politicians, so why do we pretend like they are the politicians representing just your neighborhood?

Along the same thread, genetic algorithms have been used to evaluate fair land allocation in Brazil [1], which have also historically been done by hand. I agree with another user that pointed out the very same tools can be used to gerrymander even more effectively than those districts are currently, perhaps in less obvious ways, without the oddly shaped borders that stretch wildy from place to place. It would be important not to have this occur behind closed doors but left open to scrutiny where the fitness function and biases are in full view.

[1] https://www.lume.ufrgs.br/bitstream/handle/10183/174950/0010...

How hard would it be to change the congressional election system to proportional voting by party?

In California you would vote for 53 representatives, each representative would be part of a party. First the seats are divided among all parties based on some system. Then within each party the seats are given to the representatives that have the most votes within the party.

This would introduce new parties to the system, no party would be able to govern alone.

Would this require me to make 53 different votes? This sounds a like pretty complicated situation to expect educated votes for all 53 votes.

In practice, in mostly countries that use STV PR, you have smaller electoral districts, for this reason. For instance, in Ireland, electoral districts elect 3-5 TDs depending on population. This is big enough to make gerrymandering difficult and not very effective, but small enough to avoid the situation where someone has to cast 53 different votes. And 53 would be the low end here, really. That would assume you only want to vote for one candidate per seat, which is not typically the way it works out. I'm in a four seat constituency and voted about ten preferences in the last election, say.

In practice, in STV PR it is generally optimal to vote for almost everyone on the ballot; you give low preferences to people you don't care about to avoid people you actively don't want getting in (it's commonly claimed that you should optimally give a preference to everyone, even those you actively don't want, but this is incorrect).

Usually each party would provide a list with their preferred representatives that can be modified.

Where I life, you can give up to 2 votes per representative. You don't need to choose 54 different politicians because "empty" votes will still count for the party.

Most people will just use the list of their party and maybe remove some that they don't like or vote double for someone they like.

I've always like this article's thoughtful approach to what we should value when districting: https://www.politico.com/magazine/story/2015/06/could-gerrym...

Interesting to note that the very same tool can be used to optimize the 'worst' maps as well.

Now do Maryland.

Reply on news.ycombinator.com