The Impossibility of Strategyproof Rank Aggregation

In rank aggregation, the goal is to combine multiple input rankings into a single output ranking. In this paper, we analyze rank aggregation methods, so-called social welfare functions (SWFs), with respect to strategyproofness, which requires that no agent can misreport his ranking to obtain an output ranking that is closer to his true ranking in terms of the Kemeny distance. As our main result, we show that no anonymous SWF satisfies unanimity and strategyproofness when there are at least four alternatives. This result is proven by SAT solving, a computer-aided theorem proving technique, and verified by Isabelle, a highly trustworthy interactive proof assistant. Further, we prove by hand that strategyproofness is incompatible with majority consistency, a variant of Condorcet-consistency for SWFs. Lastly, we show that all SWFs in two natural classes have a large incentive ratio and are thus highly manipulable.

💡 Research Summary

The paper investigates the possibility of designing social welfare functions (SWFs) for rank aggregation that are strategy‑proof with respect to the Kemeny distance, while also satisfying basic fairness axioms such as anonymity, unanimity, and a Condorcet‑style consistency called majority consistency. Rank aggregation is relevant in many multi‑agent settings (hiring committees, recommender systems, ensemble learning, etc.), and a strategy‑proof SWF would prevent participants from misreporting their preferences to obtain a more favorable collective ranking.

The authors adopt the Kemeny distance (the number of pairwise disagreements between two rankings) as the metric that induces voters’ preferences over output rankings. An SWF is called strategy‑proof if no voter can, by changing only his own reported ranking, obtain an output ranking that is strictly closer (in Kemeny distance) to his true ranking than the ranking produced when he votes truthfully.

The main impossibility result (Theorem 2) states that for at least four alternatives (m ≥ 4) and an even number of voters, there is no anonymous SWF that simultaneously satisfies unanimity and Kemeny‑strategy‑proofness. More precisely, the theorem covers the cases m ≥ 5 with any even n, and the case m = 4 when n is a multiple of four. The proof proceeds in two stages:

1. Base cases via SAT solving – The authors encode the existence of an SWF meeting the three axioms as a propositional formula. Variables represent the output ranking for each possible input profile; constraints encode anonymity (invariance under voter permutations), unanimity (if all voters rank x above y then the output does too), and strategy‑proofness (no beneficial unilateral deviation). Using a state‑of‑the‑art SAT solver they show the formula is unsatisfiable for (m = 5, n = 2) and (m = 4, n = 4).

2. Inductive lifting – From these unsatisfiable base cases they construct inductive arguments that preserve the impossibility when the number of alternatives or voters is increased, thereby covering all required (m, n) pairs.

To increase confidence, the authors extract a human‑readable proof of the (m = 4, n = 4) case (about 20 pages) and formally verify the entire theorem in Isabelle/HOL, a proof assistant known for high trustworthiness. This combination of automated and interactive verification addresses common concerns about computer‑aided proofs.

In addition to anonymity and unanimity, the paper examines majority consistency: if the majority relation (pairwise majority preferences) forms a complete ranking, the SWF must output that ranking. The authors prove (Theorem 1) that any SWF satisfying Kemeny‑strategy‑proofness cannot be majority‑consistent when m ≥ 4, providing a direct analogue of the classic Gibbard‑Satterthwaite impossibility for Condorcet‑consistent single‑winner rules.

The final part of the paper studies incentive ratios, a quantitative measure of manipulability. For a given SWF, the incentive ratio is the worst‑case ratio between a voter’s utility (measured as negative Kemeny distance) when manipulating versus when voting truthfully. The authors show:

• The Kemeny rule and all distance‑scoring rules (where each voter assigns a convex, increasing score to a ranking based on its Kemeny distance) have an incentive ratio of at least ⌊m/2⌋ − m, which grows linearly with the number of alternatives.
• Positional scoring rules (including Borda) have an unbounded incentive ratio; as m grows the potential gain from manipulation can become arbitrarily large.

These quantitative results complement the qualitative impossibility theorems, indicating that not only do strategy‑proof SWFs not exist under the given axioms, but the commonly used SWFs are highly manipulable in practice.

The related‑work section situates the contribution within a rich literature on social choice, Gibbard‑Satterthwaite extensions, and previous attempts to achieve strategy‑proofness via randomization, set‑valued outcomes, or domain restrictions. The paper also clarifies the relationship to earlier work on Kemeny‑strategy‑proofness, betweenness strategy‑proofness, and judgment aggregation, emphasizing that the Kemeny‑distance based definition used here is stricter than many alternatives.

In conclusion, the authors deliver a robust, computer‑verified impossibility theorem for rank aggregation: with four or more alternatives, any SWF that is anonymous, unanimous, and respects majority preferences cannot be strategy‑proof under the natural Kemeny distance model. Moreover, the standard SWFs (Kemeny, distance‑scoring, positional scoring) exhibit large incentive ratios, confirming that manipulation is not merely possible but potentially very rewarding. Practitioners designing ranking systems must therefore resort to alternative approaches—such as randomization, restricted preference domains, or mechanisms that bound the incentive ratio—if they wish to mitigate strategic behavior.