The art, science, and math of gerrymandering
October 8, 2017 6:10 AM   Subscribe

"Politics, they say, is a game where whoever’s ahead gets to change the rules on the fly. It’s about winning, not being fair. But this isn’t just a politics story; it’s also a technology story. Gerrymandering used to be an art, but advanced computation has made it a science." MeFi's own Jordan Ellenberg pens an op-ed for the New York Times.
posted by MonkeyToes (22 comments total) 30 users marked this as a favorite
Hmmm.. Interesting. I appreciated the explanation of how algorithms are used to further game an already gamed system (if I'm understanding that right!). The new More Perfect is also about this Supreme Court case. They interview Moon Duchin, a mathematician who has been working on redistricting. If you're a mathematician interested in this topic, you can attend one of her workshops!
posted by latkes at 8:20 AM on October 8, 2017 [2 favorites]

This article is great. I had no idea Ellenberg was one of MeFi's own! Still quite active, too.

2010's maps were drawn with excellent GIS tools and precise data about voting habits. The politicians in control of district maps drew lines to move groups of 50 voters around one group at a time. Not just to put them where they'd voted in the past, but where they were likely to vote in the future.

2010 districts were drawn using voting and broad demographic data. 2020 redistricting will also see much more precise and subtle gerrymandering. Marketing and online profile data of the sort that drives targeted advertising on Facebook will allow very precise selection of voters. Optimization tools like those described in the op/ed will make those districts robust to pertubation: machine learning and randomized trials are particularly strong tools for this application. Politicians will apply significantly more powerful means to choose their voters in two years. It is dangerous to our democracy.

I can't keep recommending the book Ratf**ked strongly enough. It describes in detail how the Republicans engineered a significant political advantage by controlling redistricting in many states in 2010. They're doing it again in 2020, and most likely much more effectively than the Democrats will.
posted by Nelson at 8:51 AM on October 8, 2017 [9 favorites]

The Rat... book is a good one. Jeff Greenfield said two days ago at Politico, though, that the obsession with Republican gerrymandering is overdone.
posted by LeLiLo at 9:19 AM on October 8, 2017

That Greenfield article is weird. He's making the point the Democrats are also losing because, well, they're losing popularity with voters. This may well be true. But in addition Republican gerrymandering is stacking the deck against them. So is the electoral college. These are all factors in the Republicans' current power.

Another aspect of gerrymandering is that it seems to be polarizing politics. Once everyone knows a Congressional district is safely Republican, the election becomes a contest between the Republicans as to who can be the most extremely right wing.

The Wisconsin case before the court may put an end to extreme partisan gerrymandering. I hope so. But we have to wait a few months to know. Here's the SCOTUSblog notes from the hearing. Also RBG slapping down Gorsuch as if he were some second year law student.
posted by Nelson at 9:34 AM on October 8, 2017 [6 favorites]

This article from the NYT is also good: The New Front in the Gerrymandering Wars: Democracy vs. Math.

From the coverage of the Supreme Court hearing, you can tell what the Republican talking point is on this (emphasis mine):
Much of the argument concerned various statistical tests for identifying extreme gerrymandering. Misha Tseytlin, Wisconsin’s solicitor general, said the challengers were relying on flimsy and hypothetical social science evidence.

“Plaintiffs are asking this court to launch a redistricting revolution based upon their social science metrics,” he said.

Chief Justice Roberts told Mr. Smith that courts are poorly equipped to evaluate social science data. “It may be simply my educational background,” the chief justice said of the studies before the court, “but I can only describe it as sociological gobbledygook.”
Math, statistics, regression analysis, and computer modeling are now "sociological gobbledygook" if you don't like the results.
posted by peeedro at 9:35 AM on October 8, 2017 [14 favorites]

People like winners, or at least potential winners. I suspect that gerrymandering may be suppressing Democratic (or would-be Democratic) enthusiasm. Remember that, if the popular vote mattered at all, we'd have Democratic control ofthe Presidency and of many states.
posted by steady-state strawberry at 10:19 AM on October 8, 2017 [7 favorites]

I would suggest a method inspired by John Conway's Game of Life, where an initial line is drawn to divide the state in half, by population, then each new district becomes autonomous after that. The first division is relatively neutral no matter how it is drawn (whether blended or distinctly partisan), and the method only needs two basic rules to stay that way. The first rule is that only people impacted by a new boundary can make a new boundary. The second rule is that all new boundaries are contiguous and must connect to more than one district, allowing new districts to be made from multiple districts, with no isolated enclaves. The results would be resistant to political exploits, with more local control, and satisfy the compact requirement.
posted by Brian B. at 10:46 AM on October 8, 2017 [2 favorites]

The OP article explains the issue quite well, given the subtleties of the math and manipulation. I've already pegged my outrage and frustration quotients on this issue so it didn't 'help there, but does make me want to read his recent book.
posted by mark k at 11:12 AM on October 8, 2017

repeal this 1967 piece of garbage meant to hobble anti-gerrymandering efforts for single statewide districts with at-large congressional representation: 2 U.S. Code § 2c
posted by j_curiouser at 3:52 PM on October 8, 2017 [1 favorite]

Maptitude is purportedly the primary software tool used for the Republican redistricting efforts although there are others including Auto-Redistrict, an open-source pkg that claims to "automatically create fair and compact electoral districts."
posted by bz at 6:50 PM on October 8, 2017

repeal this 1967 piece of garbage meant to hobble anti-gerrymandering efforts for single statewide districts with at-large congressional representation: 2 U.S. Code § 2c

There's a really good reason for that: large multi-member districts were drawn to combine black areas with a majority of white voters to ensure that white members would win as a slate. If you repeal single member districts, you must also put in proportional representation. MMDs without PR means decimation of racial minority representation because race and partisanship are so well tied now.

As a side note, Maptitude is a nonpartisan vendor, they sell to both parties and nonpartisan commissions.
posted by Hollywood Upstairs Medical College at 9:19 PM on October 8, 2017 [4 favorites]

Yeah, Maptitude is the main tool discussed in Ratf**ked. I haven't used it but looked pretty closely at docs and videos. It's pretty much ArcGIS-for-politics, a GIS editor with tools and data specifically aimed at redistricting. The company that makes it, Caliper, has a non-partisan democratic mission. In 2011 they put a lot of effort into a web tool so that ordinary citizens could review districting maps. Here's a video of it in action.

The really valuable thing here isn't the GIS editor, it's the data. It's remarkably difficult to collect election data in the US. There is literally no single place to get a free and open record of the 2016 vote, nor the 2012 vote. Certainly not to the granularity of per-voting-precinct you need to draw district maps. (OpenElections is trying, one county at a time, often having to type in data from PDF scans of hand-written tallies.) And that's just simple election data; never mind demographic data, and census data, and proprietary advertising models.

That's what makes the June 2017 "leak" of GOP voter data so interesting; that data is remarkably valuable and hard to come by. I have not seen the files, nor do I have a copy, only read about it. But it sure sounds interesting. (I put leak in quotes there because a prevailing theory is the GOP left it out there for fellow travelers to find as a way to circumvent campaign finance laws.)

On racial minority representation and redistricting, it's a great example of how gerrymandering is used for progressive ends too. Although even that's been subverted; see the history of the Unholy Alliance from the 1980s for details.
It was about a deal between African-Americans to increase their ranks in Congress and Republicans who wanted to increase their numbers as well. And it worked very well for both sides in that you grew the largest Congressional Black Caucus since the days of Reconstruction. But at the same time, Republicans took over all the rest of those states.
posted by Nelson at 10:38 PM on October 8, 2017 [3 favorites]

For those wondering about proportional representation versus at-large elections, and multiple seat election options, here is a handy downloadable primer.
posted by Brian B. at 7:12 AM on October 9, 2017

"The results would be resistant to political exploits, with more local control, and satisfy the compact requirement."

Any evidence for those three claims?
posted by floppyroofing at 9:37 AM on October 9, 2017

I watched some of the Supreme court hearing, at first out of curiosity, then with a growing sense of trepidation. It seems to me that trying to make gerrymanding "fair" is sort of like trying to fix a flat by wiring a spare tire to the door handle.

One of the judges mentioned "...preservation of one vote to one citizen." Is now the time to have a conversation about doing away with the electoral college? Why not simply use a grid system to determine the number of reps in the lower chamber?--let the parties take care of themselves (in the general election) without crooks using hooks to snag voters.
posted by mule98J at 10:53 AM on October 9, 2017

Why not simply use a grid system to determine the number of reps in the lower chamber?

There is no simple, unbiased algorithm that can allocate a state's population into districts. There's certainly some simple ways to start, like zip codes, but at some point, someone has to decide "the south side of this street is in district A, and the north side is in district B." Someone has to look past the maps and realize, "these eight blocks are a community, and that ninth block belongs to the next community over."

Combine with the hassle of sorting out non-voting residences (like college campuses that vote in their home districts) and transitory populations (homeless, people in long-term care or rehab places, seasonal workers, etc.), and we do need humans, probably in committees, that make those decisions- at least to start.

The biggest problem is that Republicans use the defense of "there is no objective, 100% automated system that will work" to mean "so it's okay for whatever party is in power to skew the playing field as much as they can get away with."

(I'd love to get rid of the electoral college. And we just need another 105 votes worth of states to agree. But it's not likely to happen.)
posted by ErisLordFreedom at 2:18 PM on October 9, 2017 [3 favorites]

Any evidence for those three claims?

Not computational or historical evidence, but there is the assumption to examine. The rule assumes that any local lines are created by those affected. Assuming there is a minority enclave, then at some point the power to draw their own line reaches them. It is analogous to folding a piece of paper in half five times, with the resulting 32 squares, each representing a clean division of population. Each fold requires only those in the current rectangle to achieve it. The principle at stake is that the fold made each time is self-preserving, not malicious, because if someone wants to divide and conquer, they are only punting down to others who will make the next line, and it may backfire. Each division event cannot achieve an immediate political result because any split that benefits one side will strengthen the other conversely. Or they can remain homogeneous with no changes, with the future prospect of creating a third district out of two, because they are permanent boundaries until altered locally. The compactness is ensured by a bordering rule to more than one district. A group would not be able to establish a snaked line across many districts because they would need them all to agree, which would open it up to their interests. I would assume that once a state legislature is carved up this way, that federal districts could be created in a spottily joining fashion, perhaps voluntarily, with no contiguous restrictions. That would be interesting, and going the other way with it.
posted by Brian B. at 6:07 PM on October 9, 2017 [1 favorite]

mule98J : Why not simply use a grid system to determine the number of reps in the lower chamber?

ErisLordFreedom : There is no simple, unbiased algorithm that can allocate a state's population into districts.

To expand on this I liked the OP because I now have a reasonably clear (if doubtless simplistic) idea of how to detect deliberate bias.

Suppose ErisLordFreedom and I both come to you with a redistricting plan. There are the sorts of compromises Eris describes: Say, we both divided San Francisco in half but they rounded off the neighborhoods differently than me and added slightly different areas. Is there a way you can tell Eris was acting in good will and I'm a vile huckster?

What you'd do is a right a computer program to start making small changes. Randomly swap out a pair of precincts on two of my districts. If I haven't paid attention to voting data there should be a 50/50 chance of that helping Republicans or Democrats a little bit. Repeat the random experiment 100 times. If the random swaps help Republicans 95 times out of 100 you might find that awful suspicious! What are the odds I just happened to get into one of the best possible maps for Republicans and any change helps Democrats? Pretty damn low. I'm clearly trying to put my thumb on the scales.

One of the judges mentioned "...preservation of one vote to one citizen." Is now the time to have a conversation about doing away with the electoral college?

It's always that time but unfortunately the Constitution enshrines the electoral college. So the Supreme Court can't overturn it and there are a lot of ignorant junior high civics textbooks (and teachers) who prefer describing how it got to be that way over stating the plain truth that it is at odds with rule of the people.

"One person one vote" in districting (including state legislatures) has been a Supreme Court precedent since the early '60s. Before that some states did just do geographical areas, which obviously helped disempower cities. In one key decision Warren stated "Legislators represent people, not trees or acres. Legislators are elected by voters, not farms or cities or economic interests."

I completely think Gorsuch and others, with their junior high approach to the constitution, completely would be willing to say the Constitution never says one-person-one-vote (and it doesn't) if it weren't so politically untenable. They make this argument to each other and pat themselves on the back for being smarter than everyone else, which is why Gorsuch blurted out loud thinking it was a winner without realizing he was embarrassing himself.
posted by mark k at 9:20 PM on October 9, 2017 [1 favorite]

There's no 100% fair algorithm for redistricting, but one can at least try. Also it's possible to have a fair and non-partisan process. Ratf**ked holds out Iowa as an example of a state that does it fairly. California also recently introduced a non-partisan districting process that's widely respected, although it's new and in a state so heavily Democrat that it's hard to be sure.

Canada is another interesting example. The partisan districting got so bad in the 1960s they got rid of it and replaced it with a non-partisan system that according to the article I linked, works pretty well. They're an interesting story in how you get from point A to point B, albeit in a somewhat different form of government.
posted by Nelson at 10:35 PM on October 9, 2017 [1 favorite]

I now have a reasonably clear (if doubtless simplistic) idea of how to detect deliberate bias.

There are plenty of ways to detect bias (and better ones now, with computer analysis, than we had 20 years ago), and ways to mitigate party influence in the decision-making. But implementing any of those ways requires, sigh, legislative voting, which means convincing whichever party is in power that they should give up their current edge in the future.

SCOTUS can declare that gerrymandering is unconstitutional; they can't pick one of the other options and declare it's the unbiased replacement.
posted by ErisLordFreedom at 4:00 PM on October 10, 2017

"Another aspect of gerrymandering is that it seems to be polarizing politics. Once everyone knows a Congressional district is safely Republican, the election becomes a contest between the Republicans as to who can be the most extremely right wing."

That's not actually well supported by evidence, and there's an at least equally compelling case that competitive districts lead to more extreme representatives.

To start with the "safe district" theory, one explanation for why they tend to have "more extremely" partisan politicians because the voters themselves in that district are more partisan or extreme than the overall distribution, but in general, the representatives in areas where the voters are significantly outside the median tend to be closer to the median themselves. This was challenged by some of the Tea Party, and continues to be something that people like Steve Bannon see an opportunity in.

But in general, it ties into a theory of representation that sees party voters in heterogenous clusters, and in areas where a victory for a party is assured, what tends to happen is that the mass of voters (who, generally, all have extreme views but are unlikely to be specifically motivated by them) who represent the mainstream of the party will have more weight and extremists within the party will feel less motivated and less effective.

That's contrasted with a competitive district, where voters closer to the median between the parties will tend to cancel each other out, which means that extremists who may not otherwise participate will both see the election as having raised stakes (thus necessitating their participation) but also that in order to bring more participation from the extreme wings of the party in, politicians will shift their positions further toward the outer ranges, because they're less likely to alienate engaged "moderate" voters while gaining support from more engaged "extremists."

This means that, in general, competitive races tend to produce representatives that are further away from the median partisanship of their districts than safe races do; competitive districts are more polarizing, compared to safe districts, which are more likely to produce representatives less extreme than their districts.

This is one of the reasons used to justify the "jungle" primary systems, as it increases the pressures to move toward the middle, as voters will approach the general election with less of a partisan orientation. However, to my knowledge, California's jungle primary is too recent to have developed much data about whether that's actually happening, and I don't know enough about Louisiana politics to say whether it's been effective there.
posted by klangklangston at 1:28 PM on October 13, 2017

« Older Dockless bike shares are here. Are cities ready...   |   A cartoon about new motherhood Newer »

This thread has been archived and is closed to new comments