At 6’5″, Aaron Dennis towers over the whiteboard beside him. Blue marker in hand, the 22-year-old hunches slightly to jot down suggestions being shouted by a group of people deep into a brainstorming session. Dressed mostly in nerdy T-shirts (one reads Science! with a test tube in place of the letter i), they’re trying to come up with names for a tech tool they plan to build during a two-day hackathon at Tufts University’s data lab.
The group includes computer science PhD candidates, mathematicians, political operatives, and experts in so-called geographic information systems, or GIS. That’s the mapping technology that underlies many apps and software tools that run our lives, from Google Maps to logistics software.
It also comes in handy when you’re carving the American electorate into voting districts that favor your political party, a time-honored—and reviled—tradition known as gerrymandering.
That’s what’s brought the group here to Tufts. They’re participants in a weeklong summer camp of sorts for adults focused on how math and technology can be used to make electoral maps more fair, and to convince judges and juries when they’re not. Gerrymandering, they believe, allows politicians to choose their voters, not the other way around. This event is the first of many planned by the unfortunately named Metric Geometry and Gerrymandering Group at Tufts. You can think of the hackathon as the arts and crafts part of the week—a chance for the geeks to get their hands dirty. Attendees had to apply to this session; just 14 made the cut.
On the whiteboard Dennis has scribbled “Gerrymandr,” “Gerrymetrics,” and “Politishape.”
“What about Salamander?” offers 33-year-old Ariel M’ndange-Pfupfu, a data scientist from Washington, DC. Gerrymandering got its name, after all, in 1812, when then-Massachusetts Governor Elbridge Gerry ratified a political map in which one district looked like a salamander.
The group quickly settles on the name Mander. The consumer-friendliness pretty much ends there. Try to stay awake for this part: Mander is a set of code written in the Python programming language that calculates how compact a district is. Compactness is often used as a kind of shorthand for fairness in legal debates over gerrymandered districts. The thinking assumes, not always correctly, that compact districts are better for democracy. But there’s no uniform way to measure compactness. In some places officials just eyeball it. Dennis and his team want to build a simple tool to ensure “everyone’s using the same code.”
It’s not exactly the sexiest idea, particularly when some believe tech could eradicate gerrymandering. Just feed a machine data about the electorate and a few legal parameters, and surely it will draw fairer maps than the partisan hacks who do most redistricting today. Spend a week at gerrymandering camp and you’ll quickly see how naive that is.
“It’s not enough to have a system that learns how to do this,” says Justin Solomon, a professor at MIT’s computer science and artificial intelligence lab who is running the hackathon. “You need a system that learns and explains what it learned.” Since no human can definitively explain to a judge why a machine drew a map the way it did, Solomon says, “That’s a critical problem.”
Judges increasingly need such explanations. Courts from Texas to North Carolina are challenging the legality of gerrymandered maps. In October, the Supreme Court will hear arguments in a case on partisan gerrymandering in Wisconsin, which has the potential to rewrite the rules on district-drawing. After the 2020 census, every state will begin the once-a-decade process of redrawing political maps. From algorithmic tools such as Mander that aim to help with the complicated math to educational efforts designed to teach citizens about gerrymandering, there’s ample room for innovation.
If that sounds like your idea of summer fun, well, your country may need you.
A Not-So-Simple Solution
Moon Duchin has been overwhelmed with criticism from her fellow math geeks lately. Since founding the Metric Geometry and Gerrymandering Group at Tufts earlier this year, the geometry professor has been inundated with condescending emails, calls, and handwritten letters. “There’s a kind of arrogance you see from the math and tech side, which is, ‘These idiots just haven’t asked us. We can solve that problem,’ ” Duchin says, the sleeve of her black-and-white blouse rolled up to reveal a heart tattoo on her right forearm. “It’s a special flavor of arrogance about this problem that’s really destructive.”
Truth be told, Duchin used to be one of those people. She first became interested in gerrymandering after attending an American Political Science Association event several years ago “as a tourist,” she says. That’s where she learned about the various ways mapmakers assess a district’s compactness. You could, for instance, take the distance from a district’s center to its outermost point, or measure the length of its perimeter. If districts are supposed to group people based on where they live, the thinking goes, then districts that zig and zag across the map are suspect. Such designs may be trying to avoid voters from one party in that district or water down the opposition’s votes in another. In the world of gerrymandering, the practice of stuffing the opposition into a few districts and spreading their remaining votes thinly across multiple districts is called “packing and cracking.”
There are often good reasons for oddly shaped districts. A mountain range may get in the way, or mapmakers may want to keep intact a so-called “community of interest,” such as native Spanish speakers. But examining a district’s shape is often only a starting point. As a geometrist, Duchin believed it might be something she could introduce to her students. “I thought, ‘I’ll just find the best book or article about all of those metrics and incorporate it into our class,’ ” she says. After some digging, she realized that there are plenty of ways to evaluate a political map and little agreement about the “best” method. Different techniques will yield different answers to this question: Is this map gerrymandered?
So Duchin resolved to bring together math and technical colleagues to forge a consensus around the measuring tools, and to assemble a network of experts to help people engaged in gerrymandering disputes. People like Megan Gall.
As the chief social scientist at the Lawyers Committee for Civil Rights Under Law, Gall crunches the numbers that underpin the committee’s voting-rights cases. Since 2013, when the Supreme Court rolled back a critical piece of the Voting Rights Act that required some states and jurisdictions to clear changes to voting practices with the federal government, the committee has been flooded with alerts about potential instances of race-based gerrymandering. The racial impact is crucial because the Supreme Court only considers a district improperly gerrymandered if it discriminates against a racial or ethnic group.
Gall typically works on five to 10 cases at a time, digging into the data to look for signs of racially polarized voting, crucial to showing a map is discriminatory. That often requires compiling a decade’s worth of data on a district’s changing borders, the demographics of the voting-age population, voter turnout, and election results. She then uses a mathematical model to estimate how many votes a candidate received from different racial groups. If, say, candidates overwhelmingly preferred by black voters have never been elected over a decade, Gall says, that’s a red flag for the courts.
Finding those red flags isn’t easy, however. For one thing, individual votes are secret, so understanding a racial group’s preferences rests largely on demographic data about the residents of a neighborhood. Then the necessary data is often not easily available or available in one place. “That sort of data hunt can be so difficult and still turn up no data,” she says.
Which is why Gall’s eyes widen when she talks about how technology could make her job a little easier. “In a magical world, where I had all of that information by state,” she says, holding her hands in front of her as if hovering over a crystal ball, “it would give me an analytic edge.”
In other arenas of gerrymandering research, computing power isn’t just helpful—it’s required. The computing demands are so intense that they’ve drawn the attention of Wendy Cho, a research scientist at the National Center for Supercomputing. During the Tufts conference, Cho described her Parallel Evolutionary Algorithm for Redistricting, or PEAR. The algorithm aims to analyze every possible way to divide a state into districts, then assesses the degree to which a specific map is like the others, or an outlier. This statistical technique is used in many specialties, from breaking cryptographic codes to DNA sequencing.
The number of possible maps is so astronomical that Cho runs PEAR on the Blue Waters supercomputer at the University of Illinois, which can perform 1 quadrillion calculations per second. Bringing such computing firepower to bear is warranted, Cho says. “This ability we now have to compile, synthesize, and analyze massive amounts of information can be used not only for your GPS to get you a good restaurant recommendation,” Cho told the conference. “We can use it to improve society.”
A Non-Partisan Approach
At this point you may be wondering, what side are these guys on? Suffice it to say, not many of the 600 attendees rode the Trump Train to gerrymandering camp. One hackathon participant, Vanessa Archambault, 33, worked on the tech team for Hillary Clinton’s presidential campaign. She and other attendees say Democrats are particularly motivated to redraw political maps because they hold so little power right now: Republicans control the White House, both houses of Congress, 34 governorships, and 32 state legislatures.
That said, Duchin designed the group to be deliberately nonpartisan, and the conference was about as free of overt politics as a meeting about gerrymandering can be. During a week in which Donald Trump threatened to rain “fire and fury” on North Korea, there was scarcely any talk of his administration. “We’ve tried to elevate the conversation from being upset about the election to a nonpartisan conversation about fairness,” says Solomon, the MIT professor.
There was one exception. During a happy hour in a neighborhood bar typically filled with Tufts students, author David Daley appeared to sign books and fire up the crowd. A minicelebrity in the field of gerrymandering, Daley is the author of the irreverently named Ratfucked, a meticulously reported account of the Republican effort to redraw the political map in their favor following the 2010 census. Nicknamed REDMAP, the plan entailed pouring millions of dollars into congressional and state elections so that Republicans could control the mapmaking in most states. Using a tool called Maptitude, they “packed and cracked” Democratic votes—and it worked. In 2012, Democratic House candidates received more total votes, but Republicans won the majority of seats. “The problem with democracy is what happened behind closed doors after the 2010 election in states like Ohio, where the Republican leaders who were drawing lines barricaded themselves into a suite at the Doubletree they called ‘the bunker,’ ” Daley told the happy hour attendees. “Because whenever politicians barricade themselves into a bunker, something good must be going on.”
After several minutes, Duchin gave Daley the signal to cede the floor. It’s not that anyone in the room particularly disagreed with Daley, but no one wanted the night to devolve into a Republican-bashing session. After all, when given the opportunity, Democrats gerrymander too. Former attorney general Eric Holder, with the help of Barack Obama, is raising gobs of money for his National Democratic Redistricting Committee in hopes of painting maps blue after the 2020 census.
Which brings up a critical question: If mathematicians agree on how to draw fairer maps, or if the hackathon attendees build tools to make that process easier, who will use them? If both Democrats and Republicans take to their bunkers in 2021 to redraw maps in their favor, who will listen to the do-gooders with the fancy equations and sophisticated algorithms?
Gall has a simple answer: “The courts. The courts are the ones that are going to care.” That would take time, however, for new maps to be drawn and lawsuits to be filed.
Duchin is more optimistic. As courts increasingly invalidate gerrymandered maps, as technicians build better tools to evaluate those maps, and as the public wises up to the problem, nefarious actors on both sides of the aisle may be dissuaded from trying to rig the system in the first place. “I think parties are on notice that the age of the most egregious gerrymanders is over,” Duchin says. “You’re never going to suck the partisanship out of it, but the bounds of what people think they can get away with will narrow with all this attention.”
No Magic Bullet
Early in the afternoon on the first day of the hackathon, the tang of onion lingers in the air from lunch and empty iced coffee cups clutter the desks. The 14 participants are facedown in their computer screens, working quietly on their projects. Within hours, Nick Doiron, a 28-year-old software consultant for Silicon Valley Bank and self-proclaimed mapping fanatic, is the first with a functioning product.
It’s called District Genius, a takeoff on the song lyric app Genius. On that app, users highlight a lyric and can add comments in the margins. The information is stored, so subsequent users can see explanations for the lyrics to Despacito, for example. District Genius is similar. On Doiron’s screen is the map of Pennsylvania’s first congressional district, a long and jagged shape that looks a little like an upside down AK-47, extending from Northeast Philadelphia to wealthy western suburbs such as Swarthmore. District Genius allows anyone to highlight a section of the map and suggest why a boundary may have been drawn that way.
Dorion thinks the tool could help make ordinary citizens more attuned to the process. “One thing we learned this week is people don’t even know if they live in a gerrymandered district,” he says.
By the following day, District Genius is one of eight projects to emerge from the hackathon. One team sought to address concerns about data availability by scraping and compiling voting data in Tennessee and Louisiana. Another built an educational tool to visualize the mathematical equations Gall and others use to probe for racial gerrymandering. Dennis and his team hope to integrate features of Mander, the program that evaluates compactness, into products at their employer, a mapping-software company.
None of the projects are particularly glamorous. They are not magic bullets. But they could help chip away at one of the most intractable problems our democracy faces. It may not be as fun as archery class, but it’s not bad for a summer vacation.