Justified Representation: From Hare to Droop

TL;DR

This paper systematically studies proportionality axioms in multiwinner approval voting using the Droop quota, extending theoretical understanding and demonstrating increased demands compared to Hare quota-based axioms.

Contribution

It introduces Droop quota-based versions of key proportionality axioms and identifies voting rules satisfying them, expanding the theoretical framework of proportionality in approval voting.

Findings

Droop quota axioms are more demanding than Hare quota axioms.

Certain known rules can be adapted to satisfy Droop-based axioms.

Experimental results show increased stringency of Droop-based axioms in probabilistic models.

Abstract

The study of proportionality in multiwinner voting with approval ballots has received much attention in recent years. Typically, proportionality is captured by variants of the Justified Representation axiom, which say that cohesive groups of at least $\ell\cdot\frac{n}{k}$ voters (where $n$ is the total number of voters and $k$ is the desired number of winners) deserve $\ell$ representatives. The quantity $\frac{n}{k}$ is known as the Hare quota in the social choice literature. Another -- more demanding -- choice of quota is the Droop quota, defined as $\lfloor\frac{n}{k+1}\rfloor+1$. This quota is often used in multiwinner voting with ranked ballots: in algorithms such as Single Transferable Voting, and in proportionality axioms, such as Droop's Proportionality Criterion. A few authors have considered it in the context of approval ballots, but the existing analysis is far from comprehensive. The contribution of our work is a systematic study of JR-style axioms (and voting rules that satisfy them) defined using the Droop quota instead of the Hare quota. For each of the standard JR axioms (namely, JR, PJR, EJR, FPJR, FJR, PJR+ and EJR+), we identify a voting rule that satisfies the Droop version of this axiom. In some cases, it suffices to consider known rules (modifying the corresponding Hare proof, sometimes quite substantially), and in other cases it is necessary to modify the rules from prior work. Each axiom is more difficult to satisfy when defined using the Droop quota, so our results expand the frontier of satisfiable proportionality axioms. We complement our theoretical results with an experimental study, showing that for many probabilistic models of voter approvals, Droop JR/EJR+ are considerably more demanding than standard (Hare) JR/EJR+.