Where are the really hard manipulation problems? The phase transition in manipulating the veto rule

Voting is a simple mechanism to aggregate the preferences of agents. Many voting rules have been shown to be NP-hard to manipulate. However, a number of recent theoretical results suggest that this complexity may only be in the worst-case since manipulation is often easy in practice. In this paper, we show that empirical studies are useful in improving our understanding of this issue. We demonstrate that there is a smooth transition in the probability that a coalition can elect a desired candidate using the veto rule as the size of the manipulating coalition increases. We show that a rescaled probability curve displays a simple and universal form independent of the size of the problem. We argue that manipulation of the veto rule is asymptotically easy for many independent and identically distributed votes even when the coalition of manipulators is critical in size. Based on this argument, we identify a situation in which manipulation is computationally hard. This is when votes are highly correlated and the election is “hung”. We show, however, that even a single uncorrelated voter is enough to make manipulation easy again.

💡 Research Summary

The paper investigates the computational difficulty of strategic manipulation under the veto voting rule, a simple yet widely studied aggregation mechanism. While many voting rules are known to be NP‑hard to manipulate in the worst case, recent theoretical and empirical work suggests that such hardness may be rare in practice. The authors adopt an empirical approach to explore how the probability that a coalition of manipulators can make a desired candidate win changes as the coalition size grows, and they complement this with a theoretical analysis of special cases that are provably hard.

Methodologically, the authors generate large numbers of random election profiles under the assumption that each voter’s ranking is drawn independently and identically from a uniform distribution (i.i.d. uniform). For each profile they consider the veto rule, where each voter “vetoes” his or her least‑preferred candidate and the candidate with the fewest vetoes wins. A coalition of size (m) attempts to change the outcome in favor of a target candidate by re‑ordering its own votes. The authors compute the minimal number of vetoes that must be shifted (the “manipulation cost”) using exact dynamic‑programming techniques, and they record whether the coalition succeeds for each value of (m).

The experimental results reveal a smooth, sigmoidal transition in the success probability as (m) increases. Specifically, when the coalition size reaches roughly (\Theta(\sqrt{n})) (where (n) is the total number of voters), the probability of successful manipulation jumps from near‑zero to near‑one. Moreover, when the horizontal axis is rescaled by (\sqrt{n}) (i.e., plotting success probability against (m/\sqrt{n})), the curves for all tested values of (n) collapse onto a single universal S‑shaped function. This “universal scaling law” indicates that, for i.i.d. votes, manipulation under the veto rule is asymptotically easy even when the coalition is at the critical size predicted by worst‑case complexity analyses.

To identify circumstances where manipulation might be genuinely hard, the authors construct highly correlated election instances. They consider a “hung” election in which all voters share the same ranking, making the election perfectly tied before manipulation. In this setting, a coalition must precisely split the vetoes to overturn the tie, a problem that is equivalent to the classic partition/subset‑sum problem. Consequently, deciding whether manipulation is possible becomes NP‑complete. The authors verify experimentally that, under such correlated profiles, the success probability remains low and the runtime of exact algorithms grows sharply, confirming the theoretical hardness.

A striking observation is that the introduction of even a single uncorrelated voter into a hung election dramatically reduces the difficulty. The random voter breaks the perfect symmetry, destroying the exact partition structure and restoring the smooth scaling behavior observed for i.i.d. profiles. Thus, in realistic elections where perfect correlation is unlikely, the hard instances identified by the worst‑case theory are expected to be exceedingly rare.

The paper concludes that while the veto rule can be NP‑hard to manipulate in pathological, highly correlated scenarios, the average‑case behavior for typical, independently generated votes is far more benign. The existence of a universal scaling law suggests that manipulation probability can be predicted from simple parameters (coalition size and number of voters) without exhaustive simulation. For mechanism designers, the findings imply that protecting against manipulation may require deliberately introducing randomness or diversity into voter preferences, rather than relying on computational hardness alone. The authors also advocate for further research that blends empirical phase‑transition studies with rigorous worst‑case analysis to better understand the practical security of voting rules.