Proportional Fairness for Harmful Decisions
2026-07-28 • Computer Science and Game Theory
Computer Science and Game Theory
AI summaryⓘ
The authors look at how to fairly share 'public bads'—things everyone dislikes or suffers from, like pollution—when choosing among options. They find that usual fairness ideas don't work well for these public bads, so they create new fairness rules specific to this problem. They show that under certain conditions, their approach leads to fair and stable outcomes, generalizing known methods from related problems. However, without these conditions, fairness is harder to achieve and some negative results occur. Their work opens new questions for finding fair ways to allocate public bads.
public badsallocationcoreLindahl equilibriumcompetitive equilibriumCEEIpublic goodsprivate badsfairnessimpossibility results
Authors
Benjamin Cookson, Soroush Ebadian, Dominik Peters, Nisarg Shah
Abstract
We study allocation of (divisible) public bads, where agents incur costs for alternatives and the goal is to pick a lottery over the alternatives. We show that the traditional definitions of the core, a central criterion of proportional representation for allocation of public goods, private goods, and private bads (chores), do not make sense for allocation of public bads. We introduce two formalizations of the core tailored to public bads. Under a structural condition which subsumes allocation of private bads, we show that zero-respecting Lindahl equilibria satisfy both formalizations, exhibit additional fairness guarantees, and strictly generalize competitive equilibria from equal incomes (CEEI) for allocation of private bads. Without this structural condition, we show that Lindahl equilibria exhibit undesirable behaviors, prove sharp impossibility results separating public bads from public goods, but show that a rule using a reduction to public goods recovers one of our formalizations of the core. Our results lay the groundwork for studying fair allocation of public bads, an overlooked yet fundamental problem, and highlight several structural and algorithmic directions that remain open.