Local modeling

David Shmoys: Using the power of algorithmic modeling for fairer elections

In any representative democracy, free and fair elections in which voters choose their representatives are the foundation of democratic health. This ensures that every citizen has equal representation and access to the political process, while every individual’s vote is weighted equally.

Unfortunately, there are mechanisms like gerrymandering that hijack this process by harnessing the power of modern computing to redraw districts every 10 years for partisan political purposes. In other words, representatives choose their constituents, rather than the other way around. This process is in full swing across America today.

As a computer science professor and operations researcher, I seek to model electoral fairness and have created algorithms to optimize solutions to achieve this rather than partisan political gain. In other words, nonpartisan researchers can use the power of computing to generate solutions to optimize this political process for equity. The main finding of recent research is that by combining two approaches, ‘fairmandering’ and ‘multi-member districts’ (MMD), both intentional and inherent gerrymandering can be tackled.

Thinking about the statewide constituency, there are two natural parameters to consider – the overall share of the vote Party A received and the overall share of seats Party A received. One of the goals of equity modeling is to establish districts in which there is a proportional outcome – where the expected seat share closely tracks the expected vote share.

Fairmandering uses historical records of recent elections to model, census block by census block, the probability that a specific share of the vote in that block is for Party A. Therefore, for any proposed district or set of blocks, we can characterize the probability that party A wins this district. So, for any district plan, we can also enter the expected discrepancy, or lack of proportional outcome, between seat share and vote share. The goal of fairmandering is to select a district plan that minimizes this gap.

MMDs provide a mechanism to overcome the limitations of fairmandering. Suppose each U.S. congressional district elects multiple representatives replacing the winner-takes-all system with a smaller single-member district. For example, New York could have nine districts where each district elects three representatives, rather than the current 27 districts with one representative. In elections in each constituency, each voter could provide a ranked list of preferences among all candidates and, through a mechanism known as the single transferable vote, could ensure that in that constituency the number of seats won by a party roughly corresponds to the partisan split within this district. This still preserves local representation, while providing two important benefits: expected seats reflect vote share and partisan interests are restricted.

Ultimately, decision-makers and staff members who use algorithms, equations, math, and computer science for this process should already be using fairmandering. Given the sacred duty of drawing congressional districts, creating algorithms for fairness, diversity, equal opportunity, and equal choice should be the primary focus. Without these principles involved in the mapping process, we run the risk of further democratic erosion.