Abstract
We investigate "Reverse Gerrymandering," a game-theoretic framework where autonomous agents strategically relocate between fixed districts to influence collective decisions. Unlike traditional top-down gerrymandering, this bottom-up approach relies on decentralized movement. Our analysis reveals that while infinite cycles are theoretically possible, these systems almost always converge rapidly to stable Nash equilibria in practice, leading to significant improvements in overall social welfare.
Introduction
Traditional gerrymandering involves a central authority redrawing district boundaries to favor specific political outcomes. This paper introduces "Reverse Gerrymandering," flipping the perspective to a "bottom-up" process where district boundaries are fixed, but the agents (voters) are mobile. This model applies to various real-world scenarios, such as employees choosing corporate committees, activists moving to swing states, or autonomous robots allocating themselves to specific tasks.
The Reverse Gerrymandering Model
The authors model this interaction as an iterative game. The environment consists of a set of candidates and voters distributed across fixed districts, each employing a plurality voting rule. Agents move one at a time, following a "best-response" dynamic: an agent only moves if the relocation improves their personal utility.
Two primary utility functions are explored:
- Globalists: Agents who only care about the final overall winner of the election.
- Lexicographic: Agents who seek to maximize the total representation (number of seats) their preferred candidate wins.
To maintain realism, the model includes district capacity constraints ($b_-$ and $b_+$), ensuring districts are neither too empty nor overcrowded.
Stability and Convergence
A central question is whether such decentralized movement leads to a stable Nash Equilibrium or results in endless "musical chairs."
While Figure 2 illustrates a theoretical cycle where agents 'a' and 'b' move indefinitely, the authors' simulations show that such cases are extremely rare in practice (occurring in less than 0.66% of scenarios). In the vast majority of settings, the system stabilizes very quickly, typically in fewer than 13 steps.
Impact on Social Welfare
The act of moving serves as the primary driver of influence. The simulations demonstrate a consistent "Social Welfare Boost," where the percentage of voters satisfied with the final outcome increases as agents relocate.
As shown in Figure 5, social welfare remains high across various district counts (3, 5, 10) and preference models (Single-Peaked vs. Uniform). The difference between the minimum and maximum allowed district size (the "gap") is a crucial variable; loosening these constraints consistently leads to higher social welfare. Lexicographic agents, in particular, see utility gains up to 250% higher than globalists, as they have more opportunities to make pivotal moves that improve their candidate's standing without necessarily flipping the final winner.
Conclusion and Limitations
The study proves that decentralized "reverse gerrymandering" is a powerful and generally stable mechanism for increasing agent utility. However, the model currently assumes homogeneous agent types and perfect information. Future work should address the costs of physical movement and the impact of limited or noisy data on agent decision-making.
Summary updated