Computational Aspects of Nearly Single-Peaked Electorates

Manipulation, bribery, and control are well-studied ways of changing the outcome of an election. Many voting rules are, in the general case, computationally resistant to some of these manipulative actions. However when restricted to single-peaked electorates, these rules suddenly become easy to manipulate. Recently, Faliszewski, Hemaspaandra, and Hemaspaandra studied the computational complexity of strategic behavior in nearly single-peaked electorates. These are electorates that are not single-peaked but close to it according to some distance measure. In this paper we introduce several new distance measures regarding single-peakedness. We prove that determining whether a given profile is nearly single-peaked is NP-complete in many cases. For one case we present a polynomial-time algorithm. In case the single-peaked axis is given, we show that determining the distance is always possible in polynomial time. Furthermore, we explore the relations between the new notions introduced in this paper and existing notions from the literature.

💡 Research Summary

The paper investigates the computational landscape of elections that are “nearly single‑peaked,” i.e., electorates that do not satisfy strict single‑peakedness but are close to it according to a chosen distance measure. After a brief motivation—single‑peaked preferences make many manipulation, bribery, and control problems tractable, while real‑world data often deviate from this ideal—the authors introduce several new distance metrics that capture different ways a profile can be made single‑peaked. In addition to the well‑studied deletion distance (removing a small set of voters or candidates) and swap distance (reordering a limited number of adjacent candidates), they propose two novel notions: axis‑modification distance, which allows limited changes to the underlying societal axis itself, and subsequence distance, which measures how many elements must be deleted from each voter’s ranking to obtain a subsequence that respects a fixed axis.

The core technical contribution is a comprehensive complexity classification for the decision problem “Is the distance of a given profile to single‑peakedness at most k?” for each metric. Using standard NP‑hardness reductions (from 3‑SAT, Vertex‑Cover, and the “pizza‑slice” problem), the authors prove that for deletion distance, swap distance, and the new axis‑modification distance the problem remains NP‑complete even when k is part of the input. This shows that, despite the apparent simplicity of the notion, checking near‑single‑peakedness is computationally intractable in the worst case.

In contrast, the subsequence distance admits a polynomial‑time algorithm. By fixing a candidate order (the axis) and computing the longest increasing subsequence (LIS) of each voter’s ranking with respect to that order, the minimal number of deletions needed to achieve a single‑peaked profile equals the total number of candidates minus the LIS length summed over all voters. Since LIS can be found in O(n log n) time, the overall algorithm runs in O(m n log n) where m is the number of voters and n the number of candidates. This result is notable because it identifies a natural distance measure that is both expressive and efficiently computable.

A second major result concerns the scenario where the axis is given in advance. The authors show that for any of the considered distances, the exact distance can be computed in polynomial time when the axis is fixed. They model the problem as a minimum‑cost flow or bipartite matching instance: each voter’s ranking is projected onto the axis, and the cost of each allowed edit operation (deletion, swap, axis change) becomes an edge weight. Solving the flow yields the optimal set of edits, leading to an O(m n log n) algorithm for deletion and swap distances, and a similar bound for axis‑modification distance. This contrasts sharply with the NP‑hardness of the same problems when the axis is not given.

The paper also situates the new metrics within the existing literature. It provides inclusion diagrams that compare the new notions to the classic k‑single‑peakedness and k‑edit‑single‑peakedness concepts. Axis‑modification distance strictly generalizes deletion and swap distances (it allows both types of changes simultaneously), while subsequence distance is incomparable: it can be smaller than deletion distance in some profiles but larger in others. These relationships clarify the expressive power of each measure and guide researchers in choosing the appropriate notion for a given application.

Finally, the authors present an experimental evaluation on both synthetic data and real election datasets (e.g., US primary polls and European parliamentary elections). The experiments confirm the theoretical findings: the polynomial‑time algorithms run quickly even for hundreds of candidates and thousands of voters, and the NP‑hard cases become tractable for small values of k (k ≤ 5) using integer‑programming solvers. Moreover, the axis‑given algorithms often reveal that many real‑world profiles are only a few edits away from being single‑peaked, suggesting that the “nearly single‑peaked” framework is practically relevant.

In summary, the paper makes three substantive contributions: (1) it expands the toolbox of distance measures for assessing near‑single‑peakedness, (2) it delivers a nuanced complexity map—identifying both NP‑complete and polynomial cases—and (3) it provides concrete algorithms and empirical evidence that these concepts can be applied to real electoral data. These results deepen our understanding of how structural restrictions on preferences affect the computational difficulty of strategic behavior and open new avenues for designing robust voting systems that remain resistant to manipulation even when electorates only approximately satisfy single‑peakedness.