An algorithm to detect gerrymandering – Harvard Gazette



New waves of statisticians, including a team at Harvard, have developed tools they say can help address the longstanding problem of gerrymandering of congressional and legislative districts in states by parties seeking to tip the balance. for their candidates.

Gerrymandering has been part of American politics since the 1800s, with results that are controversial and sometimes illegal, especially if done to dilute the voting power of communities of color. The battle is renewed every decade in state legislatures as census figures show which districts need to be rebalanced due to population shifts. Sometimes the violations seem egregious, but they are often more subtle, and in either case can be difficult to prove.

Harnessing the quantitative power of big data computation, these statisticians have developed algorithms capable of identifying likely gerrymandered maps by subjecting the redrawn districts to hundreds, if not thousands, of computer tests and simulations. These tools – developed over the last decade, often through open source – provide solid evidence to know if a plan does not meet the standard of a “fair” plan. The tests have recently gained traction and are increasingly used as evidence of illegitimacy in court.

A method started at Harvard in 2020 quickly took effect. It has been used by researchers, journalists and election analysts, and has played a significant role in recent court cases where lawmakers have been forced to throw away gerrymandered cards.

Called “redist”, the tool creates a large pool of alternative nonpartisan plans (more than 5,000 to 10,000) that can be compared to a map that is being proposed or has already been passed by local lawmakers or redistricting committees. This pool of nonpartisan base maps helps to see if the new map accurately represents the new changes shown in the census or if it is an outlier.

District map of Ohio (shaded box, left) next to one of 5,000 alternative maps simulated by researchers.

Algorithm-Assisted Redistricting Methodology Project

“What the algorithm does is that by using the geography and distribution of different voters within the state, it shows what kind of partisan outcome we should expect,” said Kosuke Imai, professor to statistics and government departments. “But if we see something very different from that non-partisan baseline, favoring one side under the plan adopted, that’s evidence that there are other factors influencing when the plan has been developed.”

For example, according to the researchers, consider a case in which the tool runs 5,000 simulations and finds that, on average, the legislative minority party is expected to win about five to seven seats. But simulations using a card pushed by the majority party result in its rival winning just two seats. This signifies a rare or nearly impossible event and supports the likelihood of partisan gerrymandering, the researchers said.

Imai developed redist with Cory McCartan, a PhD candidate at the Graduate School of Arts and Sciences who focuses on statistics. The pair found that traditional methods of assessing the fairness of redistricting plans did not work well because they did not provide a neutral benchmark for making objective comparisons. Fairness has often become a subjective appeal, they said.

“For a long time people have been gerrymandering and the question is, ‘OK, how do I prove it?'” McCartan said. “It’s one thing to say, ‘Hey, I think this map looks unfair because the boundaries are very twisty.’ But these things are brought before the courts, so a judge must clearly be able to decide: is it fair or not?”

The redist software uses what is called the Sequential Monte Carlo (SMC) algorithm to run its simulations. The software starts with a blank map, then draws one district map at a time, then starts again and again. Each alternate map is drawn in parallel and incorporates that state’s population, demographics, and district laws. Once these alternative maps are drawn, the redist software uses visualization tools to help users understand the data with tables and graphs summarizing the simulations.

The graph compares the vote margins of each district of the adopted plan to all the simulated districts.

The graph compares the plan adopted by each district in Ohio to all the simulated districts.

Algorithm-Assisted Redistricting Methodology Project

Unlike most similar algorithms, the SMC algorithm does not start from a single map and keep changing it. Instead, the algorithm starts with a blank map and generates new alternate shots from new blank canvases. This random generation allows the algorithm to efficiently explore more unique alternatives and generate a representative sample of plans. Existing algorithms that don’t run the risk of exploring plans very similar to the starting map, which may already have partisan bias, the researchers said.

“Let’s say the adopted plan favors the Democrats,” Imai said. “If an algorithm only explores plans that are very similar [because it starts with this enacted plan], so maybe any mock plans based on it favor the Democrats as well. When this happens, this adopted plan does not look like a gerrymander.

Redist has been used by plaintiffs in gerrymandering cases, including actions in Alabama, New York, Ohio and South Carolina. In New York, an election analyst used the SMC algorithm to produce 10,000 maps to show that the map drawn by the New York State Democratic Legislature was an “‘extreme outlier’ that likely reduced the number of Republican seats in Congress from eight to four”. A State Court of Appeals ordered the map to be redrawn.

In Ohio, Imai was called as an expert witness for plaintiffs accusing the Republican-controlled redistricting commission of gerrymandering. The SMC algorithm generated 5,000 cards for the case, none of which were as favorable to Republicans as the commission’s proposal. The state Supreme Court ordered this body to be revised.

The algorithm is also used in a case appealed to the United States Supreme Court involving allegations of racial gerrymandering in Alabama. Imai served as an expert witness for the plaintiffs, arguing that his new Congressional card intentionally dilutes the black vote. The case, which relies on Voting Rights Act protections, could eliminate one of the few remaining national safeguards against rigged legislative cards.

The maps compare partisan thinness to the share of minority voters throughout Alabama.

The maps show partisan bias across Alabama in a typical statewide election and the share of minority voters in the state.

Algorithm-Assisted Redistricting Methodology Project

Redist became a major focus of Imai’s research group at Harvard called Algorithm-Assisted Redistricting Methodology (ALARM). The group recently launched the 50 State Redistricting Simulation Project and uses the software to assess redistricting plans across the country by producing 5,000 alternative maps for each state.

The easy to use tool allows users to select a state and explore the maps. To make the data more accessible and digestible, ALARM provides a breakdown of the number of congressional districts that exist in that state, its redistricting requirements, its political geography, and a summary of what the computer found, including graphs and charts. paintings.

All data can be downloaded. The source code is provided so that it can be used as a template to generate mock plans to different specifications.

Members of the ALARM group – which includes undergraduates, graduate students and even high school students – say the process is rigorous because each state has different rules that must be translated into the algorithm, and they must run diagnostics. to make sure everything is going well. .

Yet they believe the effort is worth it.

The next step for Project ALARM is a plan to explore redist expansion by examining partisan bias from gerrymandering to racial gerrymandering, and to assess gerrymandering at more local levels.

“There is always a question of how different ways of drawing borders can benefit some voters and harm others,” Imai said. “It is important that social scientists understand the nature of these types of political manipulations and address them.”

Source link


About Author

Comments are closed.