Louisiana’s 6th would be a fairly normal-looking congressional district if it weren’t for the large 2nd district that cuts into its middle. Republicans in the Louisiana statehouse designed the 2nd district to separate the high numbers of black voters in inner Baton Rouge from the 6th district, which now contains the mainly-white Baton Rouge suburbs and outlying rural areas. In effect, this created a badly gerrymandered 6th district where voting-age blacks make up only 21.6% of the electorate—helping the Party of Trump to dominate.
Such gerrymandering is common, especially in states with a close Democratic/Republican split (such as Texas and Virginia). District maps are redrawn every decade based on the census results, usually by the party in power in the statehouse. And the process is about to start again. The new census—the first to be conducted online—kicks off March 12.
New district shapes are necessary to reflect demographic shifts revealed by the census data. But the shaping of the districts often has more to do with politics than demographic and geographic correctness.
Large segments of the public have expressed anger about this kind of gerrymandering, especially by Republican majorities. If humans tend to draw district lines in such a biased, political way, what would happen if the job were turned over to a computer to draw lines that are objective and fair? Scientists could potentially feed all the census and geographic data into a neural network and let the algorithm do the rest.
However, as appealing as an apolitical computer program solving gerrymandering sounds, it turns out it’s not that simple. It’s practically impossible to remove the political elements from district mapping, even if technology is involved. However, researchers from Harvard and Boston University have developed a deceptively simple method that keeps the districting process in human hands and gives both parties a hand in shaping the districts, while using sophisticated mapping algorithms to ensure their approach is as fair as possible. Now if only state legislatures will choose to use it.
Why gerrymandering exists
We use voting districts to associate segments of the population—voters—with geographic places. They’re supposed to ensure that segments of the population are adequately represented in everything from school districts to statehouses to the White House.
The shape of these maps has huge implications in election results. That’s why majority parties in some state legislatures have weaponized the drawing of district maps to help themselves win elections and stay in power.
“The ones that continue to pop up . . . [are] states like North Carolina, Texas, and Pennsylvania,” says Harvard Kennedy School political science professor Benjamin Schneer. Largely because of the district lines, it’s common for one party in such states to represent a minority of voters in the state yet hold a majority of seats in the state legislature, Schneer says. “And then [they] can draw maps that really, really advantage that party in future elections.”
The federal government provides some very basic district-drawing rules that apply to federal elections but leaves it mainly up to the states to form their districts. In all but seven U.S. states, the party in power either draws the district lines by itself or oversees a commission that does it. Either way, the minority party often has little say in the process.
Majority parties gerrymander in two basic ways. A party might use a practice called “cracking” to separate concentrations of voting groups (minorities, for example) and spread their voting power across several districts. Or it might do the opposite: Using a practice called “packing,” it might draw district lines to concentrate members of specific voter groups into one district to create majorities of likely supporters in adjacent districts.
Especially in states such as North Carolina, Virginia, and Texas, these tactics have led to the creation of crazily drawn district lines chosen solely to maximize the electoral power of the majority party’s base.
And just this past June, the options for minority parties to seek legal relief narrowed. The U.S. Supreme Court said it would not stop state legislatures in Maryland and North Carolina from drawing maps designed to minimize the political power of minorities, except possibly in extreme cases of gerrymandering. In its majority opinion, the court indicated that it’s up to state legislatures and perhaps Congress to resolve disputes over district maps.
Public outcry over the problem has led to a number of state ballot initiatives to establish independent districting commissions, but even commissions face a tough challenge, and lots of political pressure, in drawing the maps.
A political strategy game could help
A district-drawing tool that people see as fair might allow the parties to work things out together, without having to rely on the court to referee.
Schneer, Harvard Kennedy School grad student Kevin DeLuca, and Boston University political scientist Max Palmer refer to their proposed method of drawing district maps as the “define and combine” procedure. It’s like a political strategy game with only two moves.
Move 1 (define): The majority party goes first. It knows from the new census data the size and density of the population across the state as well as the shapes of the districts that have been used since the last census. Based on that information, the party draws the districts how it wants them. But here’s the trick: it has to draw twice the number of districts as are actually called for, breaking up the districts it actually wants into two pieces. If the population of the state dictated the need for five districts, the majority party would create 10. The only rule is that the party can’t draw “donut” districts that completely encircle other districts.
Move 2 (combine): Now it’s the minority party’s turn. Its job is to simply recombine the “subdistricts” back into the final districts. And only neighboring districts can be combined.
That’s it. Both sides are very aware that their opponent will use its move to get to as advantageous a map as possible. When drawing the original group of subdistricts, the majority party will be thinking about how the minority is likely to combine them. The minority party must anticipate how the majority is likely to draw the subdistricts and have a corresponding strategy for recombining them.
By giving each party one move, the Define and Combine strategy reduces parties’ ability to pack or crack.
“It’s harder to pack when the other party can take what you’ve combined and undo it in the second stage,” Palmer says. “And it’s harder to crack when in the second stage that second-mover party can reunite those cracked groups.”
Districting by algorithm?
To illustrate the procedure’s effectiveness in the real world required some data science. Palmer, Schneer, and DeLuca started with the theoretical question: What would be the partisan composition of districts in a state that implemented Define-Combine? Would the process result in districts that treated voters in a more inclusive and fair way than if only one party drew the district lines?
They chose eight states (based on population and party competitiveness) and used mapping algorithms to generate a representative sample of all the possible ways that the state could be broken into subdistricts as in the “define” step of the procedure. This generated thousands of maps for each state, with each subdistrict having half the population of a congressional district.
Then they examined the ways that the subdistricts could be paired together, as in the “combine” step. They identified the maps that the parties would likely choose when trying to win as many seats as possible within the structure of the Define-Combine procedure. Then they compared those maps to district maps that might have been drawn unilaterally by each party.
Palmer, Schneer, and DeLuca found that implementing Define-Combine significantly affected the number of seats each party won in the legislature. With a single mapmaker, whichever party gets to draw the map retains a big advantage in the number of seats won because that party is able to gerrymander. But in the Define-Combine procedure, the majority party had a much smaller advantage. Ultimately, the process produced more moderate maps.
If algorithms were able to draw every possible district and subdistrict, you may wonder why we can’t just give the whole job of drawing maps to a computer. Why not provide it with all the census, demographic, topographical, and voter data we have and let the computer generate the fairest possible district lines?
The problem is that there are different definitions of “fair,” as Palmer pointed out to me. Even if you had the most powerful neural network and the best data, Palmer says, you’d still have to define your priorities in drawing the map. “Are we going to prioritize keeping counties together, or towns and other communities together?” Palmer says. “Are we going to prioritize what districts should look like and how compact things should be?”
You can imagine those types of decisions being expressed in the parameters for a neural network, which are used to weigh the importance of various data types and point the model toward a desired outcome.
“There’s a lot of decisions there, and those decisions have political impact,” Palmer said.
Inevitably, things get very human again very quickly, and the arguing begins. Even an advanced AI that could understand the nuances of human politics is likely to draw maps that some group of people will hate.
The long road to acceptance
Will states, and their majority parties, ever actually accept alternative methods of map drawing, such as Define-Combine?
They might. State legislatures may see Define-Combine as the least of all evils. Public opinion about state legislatures’ ability to draw fair maps is very low, and pressure is rising in many states to take map drawing out of the hands of politicians and give it to independent commissions. “In comparison, the state legislature might actually prefer some sort of [approach] that brings both parties into the process but still allows it to retain some measure of control,” Schneer says.
When the legislative or commission process ends in deadlock, or when it produces extreme gerrymandering, it’s often state supreme courts that must act as mediator. The court can enlist experts to painstakingly draw a fair map, but this takes a long time. It might be easier and faster for the court to order the state legislature to return to the table and use a simple, prescribed method of creating a map everyone can live with. Palmer, Schneer, and DeLuca believe the courts might see the Define-Combine approach as such a method.
To date, the Define-Combine method has never been tried. Schneer and Palmer are just now preparing the research paper laying out the idea. Once they collect some feedback and perhaps make a few tweaks, they plan to send it to a peer-reviewed political science journal for publication.
If the response is promising, Schneer and Palmer aim to start trying out the approach in the real world. But they want to walk before they run. Instead of using it to help draw congressional maps, they might first suggest that states use it for something smaller, such as drawing new state legislature districts, or for mapping new school districts.
“I’d love to see this approach tried in some smaller places where it still might be partisan and contentious but maybe a simpler problem to start with,” Palmer says. In the near term, Palmer, Schneer, and DeLuca hope their idea might spark some conversation among state legislators.
Their timing seems right. The political climate is only growing more partisan, and finding neutral and trusted people, institutions, and ideas is getting harder. A collaborative tool that takes some of the partisan heat out of redistricting might be just what we need.