| paper | Guide to Numerical Experiments on Elections in Computational Social Choice |
| authors | Niclas Boehmer, Piotr Faliszewski, Łukasz Janeczko, Andrzej Kaczmarczyk, Grzegorz Lisowski, Grzegorz Pierczyński, Simon Rey, Dariusz Stolicki, Stanisław Szufa, Tomasz Wąs |
| venue | IJCAI 2024 |
| filed under | voting · theory |
| judged by | gpt-5.6-luna / xhigh (triple__luna__xhigh__c2r1) |
| judge confidence | high |
| authors would recognise it | yes |
The paper contains no numbered computational result: its numbered material is descriptive survey and methodological guidance. A distribution \(\mu\) over ranking types is a natural population representation, but the paper supplies no computational problem or result to mirror.
fails bit a — no named computational result to mirror
The proposed representation covers the paper's election profiles, statistical cultures, and size regimes as population-model motivation, but no named computational result.
The strongest honest case is limited: this paper provides an excellent justification for a continuous-population *setting*, but it contains no qualifying computational anchor.
The natural mirror is clear. For an ordinal election with \(m\) candidates, let the voter types be the complete rankings \(T=S_m\), and let \(\mu_t\) be the fraction of voters with ranking \(t\). The finite election sequence \(V\) is replaced by its empirical distribution \(\mu\). Impartial culture, Mallows elections, Euclidean elections, and the other statistical cultures then become ways of generating distributions over such societies. A statistic measured in the experiments—winner share, pairwise preference frequency, swap-distance profile, or distance from a reference election—becomes a functional of \(\mu\).
The most credible regime is the paper’s “politics” regime: \(m\le 20\) candidates and \(n\ge 2000\) voters, including large-scale polling, parliamentary elections, and city-board elections. A more convincing high-multiplicity scenario would be a large survey or recurring institutional electorate in which many respondents share a stable preference type. Euclidean models also give a respectable interpretation: latent voter regions induce positive mass on ranking cells, so the continuous society records the fraction of the population in each ranking type. Participatory-budgeting electorates provide another large-population setting.
The authors would recognize this as the population-level version of their election model: they explicitly define an election through voters’ preference orders and repeatedly study the histogram and geometry of those orders. This is not merely making outcomes fractional. However, the paper does not itself formulate an optimization, decision, manipulation, control, or robustness problem on those profiles.
Crucially, there is no numbered Theorem, Lemma, Proposition, or Corollary in the paper asserting NP-hardness, polynomial-time solvability, fixed-parameter tractability, or any other computational classification. Remark 2.1 and Remark 3.1 are descriptive remarks about data collection and election-size regimes, not computational results. Results cited from other papers cannot serve as anchors for this paper.
Therefore there are zero valid anchors and, under the programme’s rules, zero continuous problems that can honestly be stated “per anchor.” One could invent a follow-up such as optimizing the choice of a statistical culture whose induced distribution best approximates a target \(\mu\), but that would be a new computational problem, not a continuous mirror of a named result in this paper.
The paper’s strongest contribution to ChoCo is consequently indirect: it identifies realistic large-population regimes and supplies natural type distributions—Mallows, Euclidean, urn, and real-election histograms—from which future computational problems could be built. Its weakest point is decisive: the paper studies how experiments are conducted, not a computational social-choice problem whose complexity could be continuized.
The strongest negative case is that the paper contains nothing computational to continuize. This is not a technical objection to distributions over voters: such distributions are entirely natural here. It is a mismatch between the paper and the programme’s required object. The paper is a survey and methodological guide. Its numbered material concerns data collection, election-size regimes, statistical cultures, and experimental practice. It contains no theorem, lemma, proposition, or corollary asserting an algorithmic, decision, optimization, approximation, parameterized, or complexity result. Remarks 2.1 and 3.1 are descriptive, not computational anchors.
The proposed profile mirror therefore has no qualifying target. Replacing \(V\) by a distribution \(\mu\) over \(S_m\) is mathematically clean, but it merely repackages the paper’s data. Winner shares and pairwise frequencies are linear functionals such as
\[
\sum_{t\in S_m}\mu_t\,s(t,c)
\quad\text{and}\quad
\sum_{t\in S_m}\mu_t\,\mathbf{1}[a\succ_t b].
\]
Computing them is aggregation, not a new computational-social-choice problem. Even an average swap-distance statistic is just a weighted sum
\[
\sum_{t,t'}\mu_t\mu_{t'}\kappa(t,t').
\]
Its input may be high-multiplicity, but the paper supplies no result about computing, optimizing, or deciding anything involving it.
The same defeats the proposed “politics” anchor. Large elections, polling populations, recurring institutional electorates, and participatory-budgeting electorates are credible high-multiplicity settings. They establish that the type distribution is sensible. They do not create a computational question present in this paper. The paper reports which sizes and cultures researchers use; it does not solve an election problem whose complexity depends on those voters. The \(n\ge 2000\) regime is an empirical recommendation, not an algorithmic parameter.
Euclidean models do not repair the gap. If one retains only rankings, they again produce a distribution \(\mu\) over ranking types, and the relevant statistics are routine functionals of \(\mu\). If one retains latent voter coordinates, then fitting or integrating over the geometric model becomes a new statistical or geometric problem, not a mirror of a result in the paper. Likewise, participatory budgeting contributes approval-profile data, not a numbered budget-allocation theorem. Moving from those data to fractional budgets or optimization would also shift continuity into the outcome space, which is outside ChoCo’s population-only scope.
The “ground truth” discussion offers the best attempted rescue. One might define a continuous Kemeny-style problem: given \(\mu\), find a ranking minimizing expected Kendall–\(\tau\) distance. That is a legitimate high-multiplicity computational problem, and the authors’ discussion of recovering rankings from multiple sources makes it recognizable as an application. But it is not a mirror of a result in this paper. The paper neither formulates Kemeny optimization nor proves anything about its complexity; it merely lists ground-truth recovery as an experimental motivation. The same is true of optimizing a statistical culture to approximate a target distribution, fitting Mallows parameters, or choosing an embedding for the election map. Each could support a worthwhile future project, but each would be newly introduced computational content.
Thus the proponent’s strongest points establish a useful population-level data model, not a computational mirror. Under the programme’s anchor rule, the correct verdict is red: there are zero admissible anchors, and the paper should not be selected as a source of continuous-complexity problems.
The universal claim is weaker than that verdict. I cannot honestly prove that no worthwhile extension could ever be built: a new Kemeny, distribution-fitting, or robust-aggregation problem might be valuable. What can be defended strongly is the narrower conclusion that such work would be a new ChoCo problem motivated by this survey, not a continuization of one of its computational results.
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.