An Empirical Study of the Manipulability of Single Transferable Voting

The proposed continuous STV mass-manipulation problem is a credible high-multiplicity extension, and its hardness would plausibly lie in candidate-side elimination orders. However, the paper contains no qualifying numbered computational result: its hardness and algorithmic bounds are imported from cited work, while its own algorithm and experiments are unnumbered. Under the required anchor rule, bit (a) fails.

The objection that survived

The paper reports the hardness anchor only as cited prior work in prose, rather than proving it in a qualifying numbered result.

What the mirror covers

The mirror covers STV manipulation by a same-type positive-mass bloc and unanimous coalition reports, but not the paper's empirical distributions, runtime experiments, or imported algorithms.

Open questions for a prover

The case FOR (proponent)

The strongest honest case is a genuine continuous-population problem, but it is only conditionally admissible under your anchor rule: this paper contains no numbered computational theorem, lemma, corollary, or proposition. Its complexity claims are prose summaries of cited work. Thus there is no theorem number to quote, and the paper cannot supply a formally compliant anchor. The following is nevertheless the best mirror I would defend.

The strongest provisional anchor is the paper’s statement in Section 2 that STV manipulation by one unweighted agent is NP-hard when the number \(m\) of candidates is unbounded. This is attributed to Bartholdi and Orlin’s cited result \(\cite{stvhard}\), not proved in Walsh’s paper. In full, the underlying problem is: given an unweighted STV election, a designated candidate \(c^\star\), and one manipulator who may submit any complete ranking, decide whether some submitted ranking makes \(c^\star\) win. The hardness is with respect to unbounded \(m\). The paper also reports, but does not prove, the related \(O(n1.62^m)\) algorithm of Conitzer, Sandholm, and Lang and the \(O(m!(n+mk))\) coalition algorithm of Coleman and Teague; I would not use either as an additional anchor.

A natural mirror is Continuous STV Mass Manipulation. Let \(C\) be the candidate set, \(T=S_C\) the complete-ranking types, and let \(\mu\in\mathbb{Q}_{\ge 0}^{T}\) be the normalized distribution of the nonmanipulating population, with \(\sum_t\mu_t=1\). A positive-mass bloc of size \(q\in\mathbb{Q}_{\ge0}\) consists of many voters sharing a true ranking \(\sigma\), with \(c^\star\) preferred to the relevant alternatives. The bloc must coordinate on one reported ranking \(\pi\in T\). The resulting society is

\[ \mu^{q,\pi}=\frac{\mu+q e_\pi}{1+q}. \]

STV is then run exactly as in the paper, using first-place mass, transferring the mass of eliminated rankings, and applying a fixed tie-breaking order. The decision problem asks whether there exists \(\pi\in T\) such that \(c^\star\) wins under \(\mu^{q,\pi}\). Its optimization form asks for the minimum bloc mass \(q^\star\) that can make \(c^\star\) win, together with a witnessing ranking \(\pi\).

This is a credible high-multiplicity regime: a large ranked-choice election may contain thousands or millions of voters but comparatively few recurring party-line, organizational, demographic, or issue-based rankings. The continuous object is the electorate itself; candidates and complete rankings remain discrete, and the winner is still an ordinary STV winner. Nothing is fractionalized at the outcome level, and the bloc’s action is still a strategically chosen ranking. The move is therefore recognizably the paper’s manipulation problem rather than a tractable problem obtained by deleting STV’s difficult structure.

The expected classification is Class B. A finite instance with \(n\) fixed voters and one manipulator embeds exactly by setting \(\mu_t=n_t/n\) and \(q=1/n\). For every report \(\pi\), the resulting masses are proportional to the corresponding integer vote counts in the \(n+1\)-voter election, so all STV comparisons, eliminations, transfers, and majority tests are preserved. The likely source of hardness is therefore the candidate-side elimination order, not the individuality of voters. Continuization should not dissolve it. For fixed \(m\), however, enumerating possible reports and elimination orders gives an algorithm whose exponential dependence is confined to \(m\); under a dense explicit representation of all ranking types, that enumeration may even be polynomial in the input length.

A useful follow-up would replace the common report \(\pi\) by mass-transfer variables \(x_\pi\ge0\) with \(\sum_\pi x_\pi=q\). That would model a coalition whose members may coordinate fractionally across several reported rankings and could lead to LPs indexed by elimination orders. I would keep this as a sequel rather than the lead mirror, because Walsh’s coalition discussion explicitly assumes voting in unison.

The weakest point is decisive: the paper itself does not prove or number the NP-hardness result. Moreover, the literal one-agent version has vanishing mass in a large electorate, so the meaningful mirror necessarily uses the paper’s coalition perspective and introduces a positive-mass bloc. That makes the proposed problem a strong continuous analogue of the paper’s subject, but not a formally eligible mirror under the stated anchor rule. My honest verdict is therefore: substantively strong, likely Class B, but a near-miss rather than a compliant positive case.

The case AGAINST (opponent, writing after the proponent)

The negative case is decisive under the stated anchor rule: this paper contains no qualifying named computational result. It has no numbered theorem, lemma, corollary, or proposition asserting a complexity claim. The NP-hardness of single-agent STV manipulation is only reported in prose and attributed to Bartholdi and Orlin; the \(O(n1.62^m)\) algorithm is likewise imported from Conitzer, Sandholm, and Lang. The paper’s own contributions are an unnumbered algorithmic description, runtime experiments, and empirical observations. A bibliography cannot supply the missing anchor.

That defeats the proponent’s only anchor. Its proposed problem is mathematically sensible, but it is a mirror of an external result, not of a named result in this paper. The paper’s actual central conclusion— that manipulation was usually easy in sampled finite elections—is empirical rather than a computational theorem. A continuous version would require a new question about regions or measures in the profile simplex, and perhaps a probability distribution over such societies. That could be worthwhile, but it would be a new study rather than a continuization of a result established here.

The substantive objection to the proposed model is weaker. A literal single manipulator has vanishing population mass and cannot affect the STV outcome of a nonatomic society. Replacing that agent by a positive-mass bloc, as the proponent does, changes the problem to coalition or weighted manipulation. But this is not a decisive objection: a hybrid model with a continuous background and one positive-mass atom, or a same-type bloc, is a legitimate high-multiplicity formulation. It preserves rankings, STV transfers, and the strategic report. The proposed mirror therefore survives as a plausible new research problem if the anchor requirement is relaxed.

So the strongest honest verdict is procedural, not substantive: no compliant anchor exists, and the case for a mirror cannot be grounded in this paper’s own named computational results. I cannot honestly claim that every such continuous STV question is worthless; the proponent’s mass-manipulation model is a credible Class B candidate once imported complexity results are allowed.

The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.