Abstract
This paper provides the first comprehensive game-theoretic analysis of liquid democracy, modeling the choice to delegate a vote as a rational trade-off between the effort required to vote accurately and the benefit of "borrowing" a peer's expertise. We prove that stable delegation patterns (Nash equilibria) exist under several conditions and demonstrate how network structures—such as scale-free or small-world networks—drastically change the collective accuracy of the group.
Introduction
Liquid democracy is a hybrid voting system where voters can either vote directly or delegate their vote to a proxy. Unlike traditional proxy voting, these delegations are transitive: if Alice delegates to Bob, and Bob delegates to Charlie, Charlie casts Alice’s vote. While liquid democracy is praised for its flexibility, it faces a "rationality" gap. Most previous research assumes voters delegate simply to improve accuracy, ignoring the cost of effort involved in becoming an informed voter. This paper asks whether individual rationality leads to a group-level "Nash Equilibrium" and how these rational choices affect the group's ability to reach the correct decision.
Methods
The authors introduce the Delegation Game, a formal model where $n$ agents must decide on a binary issue. Each agent has an individual accuracy ($q_i$) and a cost of effort ($e_i$). If they vote directly, their utility is $q_i - e_i$. If they delegate, they pay no effort but inherit the accuracy of their "guru" (the final person in the delegation chain).
The study utilizes synthetic social network benchmarks to evaluate how different community structures influence delegation: Erdős-Rényi (Random), Regular, Watts-Strogatz (Small-World), and Barabási-Albert (Scale-Free). Simulations on 250-node networks compared "Iterated Best Response" dynamics against "One-Shot" dynamics to explore convergence and collective accuracy.
Key Insights and Results
The Emergence of Gurus
Rationality naturally leads to the emergence of "gurus"—highly accurate individuals who accumulate significant voting power. In effortless settings, the population eventually delegates to the single most accurate person they can reach. As shown in Figure 1, the percentage of gurus generally decreases as the network degree increases, indicating that denser connectivity allows for more efficient delegation to a smaller pool of experts.
Network Effects on Accuracy
While delegation improves the accuracy of individual voters by providing a better "guru" for less effort, it can decrease the group's overall accuracy. This occurs because delegation concentrates too much weight on a few individuals, potentially violating the statistical benefits of the Condorcet Jury Theorem.
As illustrated in Figure 2, network topology significantly impacts the probability of a correct majority. Scale-free networks (represented in blue) often perform the worst, especially as the degree increases, as they are prone to local optima and high weight concentration on a few hubs.
Discussion and Limitations
The authors successfully proved that Pure Strategy Nash Equilibria (PSNE) always exist for deterministic types and effortless voting games. However, a formal proof for the existence of Nash Equilibria in the most general case (probabilistic types with non-zero effort) remains an open challenge. Furthermore, the model assumes voters' types are independent, whereas real-world voters are often polarized into echo chambers. Future research should explore how different voting rules and non-binary issues interact with these rational delegation dynamics.
Summary updated