An Unbiased Algorithm Could Help Put an End to Partisan Gerrymandering
Three psychology researchers have taken it upon themselves to solve the nationwide problem of partisan gerrymandering using a computer algorithm.
The US Supreme Court recently heard oral arguments in the case of Gill v. Whitford to decide if re-mapping state voting districts can result in gerrymandering that is so unfair to the minority political party, it violates the US Constitution. The question is whether the process of re-drawing voting lines to benefit the party in power effectively prohibits voters from being treated equally and blocks their freedom of expression, necessitating judicial intervention.
A redistricting plan drawn up by Wisconsin Republicans after the 2010 election gave them an advantage in subsequent contests. In 2012, Democrats won 51.4 percent of the statewide vote and Republicans won 48.6 percent, but Democrats got just 39 seats out of 99 while Republicans secured 60. The gerrymandered map had the apparent effect of diluting the influence of Democratic votes.
The Supreme Court is expected to decide the case by June 2018, but in the meanwhile three psychology researchers have taken it upon themselves to solve the nationwide problem of partisan gerrymandering using a computer algorithm.
“In recent years, the sophistication of gerrymandering has increased in line with the data science revolutions sweeping across all sectors,” Olivia Guest, experimental psychology researcher at University College London and one of the creators of the algorithm, told Seeker. “Our goal was to suggest an alternative in which democratic principles for districting are openly debated and formalized.”
In most states, state legislatures control the redistricting process for both state legislative and congressional districts. Computer algorithms are already used to draw or re-draw district maps in many states, but often in secret. According to the Herald Times Reporter, when Republican state legislators in Wisconsin redrew the voting districts in 2010, they did so in the offices of a Madison law firm without revealing anything to the public.
Guest and her colleagues want the redistricting process to be transparent.
“What we propose is to free redistricting from party political biases,” she said. Their algorithm was created using an open-source code that’s available for anyone to reproduce the results, which Guest believes is crucial to the transparency of the model. “The use of open-source software and transparent, easy to understand code, would help keep the process unbiased, and allow people to verify and trust the results.”
The model they tested was based on the simplest possible conditions: voters within the same district should be geographically close. They believe politicians shouldn’t be involved in district-mapping at all, relying instead on computers to do the work.
“Districting should be no different than multiplying two large numbers together using a calculator,” Guest remarked. “One knows the numbers, but relies on the computer to do the calculation properly.”
The researchers were inspired to devise a solution to partisan gerrymandering because they see it as a threat to democracy.
“Gerrymandering is corrosive to basic democratic values,” Bradley Love, another experimental psychology researcher at University College London and a co-creator of the algorithm, told Seeker.
“In extreme cases, it disenfranchises citizens by creating districts in which results are virtually preordained,” he said. “Government can become less representative and responsive to the will of the people, and instead [become] captured by special and powerful interests.”
In developing the formula, the research team aimed to test the difference in population concentration between computed districts and districts that currently exist. They theorized that the density would be largest for big states, which would be challenging for any human to fairly and accurately separate into districts. To evaluate this, they created a clustering algorithm to redraw lines for all 435 congressional districts in the US, adhering to a federal law that requires all districts to have roughly the same population size.
“The model starts with one cluster for each district at random locations within the state,” Guest explained. “At the start of each round, the model assigns people to the nearest cluster. At the end of the round, each cluster updates its location to be in the center of its people. Then, using these updated positions, a new round begins with everyone reassigned to the nearest cluster once again. This goes on and on until the clusters stabilize, defining the voting districts. The one nuance is that clusters with more people are penalized so that each cluster ends up with roughly the same number of people.”
One aspect of redistricting that many states require is the preservation of communities of interest — people who are demographically similar based on race, class, and/or culture. Because these groups are likely to have similar political concerns, they benefit from having unified representation in the legislature. But grouping voters based on geography could split these communities apart, which is why geography alone is not sufficient criteria for redistricting.
The research team’s algorithm can take more than geographic location into account, but it requires more human involvement than simply inputting census blocks. If the algorithm were to consider communities of interest, it would require humans to define what constitutes a community of interest, something that can be difficult to do. But the algorithm is capable of mapping districts based on different criteria.
“Basically, people determine the goal of districting and the computer solves it,” Guest said.
Though other algorithms have been proposed to identify partisan gerrymandering, Guest and Love are confident that their redistricting tool brings multiple benefits to the process. Increased transparency would reduce the chance that a politician would purposefully draw districts to benefit their own party, and having computers carry out the task would produce an equitable map.
“Our work suggests that beyond political bias, a lot of what we view as gerrymandering may instead result from the complexity of the districting task,” Guest said. “It’s beyond human abilities to perfectly group millions of people into fair districts, so it’s best to formulate the goals and let the machine labor away to return the correct answer.”
Other technological tools are already aiding gerrymandering behind closed doors, and the researchers acknowledge that politicians could potentially use this new algorithm for similar purposes, but they hope that increasing pressure to combat partisan redistricting will help prevent that.
“Rather than use algorithms to disenfranchise, machines could just as easily be used openly to promote fairness and democratic values,” Love said.
Nothing stands in the way of states to prevent them from adopting a solution like this today if they choose to do so.
“There’s no need to wait,” Guest said. “We think any system is better than the current one, given how opaque the redistricting process is and how biased and convoluted districts currently are.”
But the public would have to hold politicians accountable in order for this vision to become a reality.
“Voters need to demand a change and hold politicians’ feet to the fire,” Love said. “Voters across the spectrum would have to unite [and] democratic values need to be elevated over short-term partisan gains.”
WATCH: The Mind Games Politicians Use to Win Votes