The foundational STAR paper — Wolk, Quinn & Ogren (2023), claim-checked¶

The peer-reviewed paper that introduced the two metrics this library leans on — Voter Satisfaction Efficiency (VSE) and Pivotal Voter Strategic Incentive (PVSI) — and made the academic case for STAR. Because the authors are STAR advocates, this page runs it through the same claim-check recipe we apply to every camp: concede the kernel, test what's testable, flag the overreach, disclose the lean — and let the fairness cut against STAR too.

Level: 301 → 401 · deep dive · Full text (open access): DOI 10.1007/s10602-022-09389-3 · local copy wolk_quinn_ogren_2023_star_vse.pdf.

Wolk, S., Quinn, J. & Ogren, M. (2023). "STAR Voting, equality of voice, and voter satisfaction: considerations for voting method reform." Constitutional Political Economy 34(3): 310–334. Accepted 24 Dec 2022, published 20 Mar 2023. Open access (CC BY). Correction notice: 10.1007/s10602-023-09426-9.

Why this paper matters to the library¶

Two numbers used throughout this repo trace back to here:

VSE — the utilitarian-accuracy score behind "STAR tops the accuracy charts" (100% = always elect the utility-maximizing winner, 0% = a random winner). It is the modern form of Bayesian Regret.
PVSI — the Pivotal Voter Strategic Incentive: how much a pivotal voter gains on average by voting strategically rather than honestly. Near 0% means strategy rarely pays; negative means it backfires more often than it works. This is the peer-reviewed source behind the repo's "STAR ≈ 1:1, IRV ≈ 3:1, Plurality ≈ 17:1 backfire ratios."

Until now the library used these figures while citing only a project website. This page anchors them to the peer-reviewed original.

What the paper actually claims¶

STAR = a 0–5 score ballot + an automatic top-two runoff — designed to capture both preference order and preference strength, to be summable (unlike IRV), and to run on existing machines by addition.
Reject pure pass/fail criteria; measure frequency × severity instead. Following Arrow (1950), every method fails some criterion, so binary checklists are "more divisive than constructive." Their alternative is statistical: build a generative electorate model, simulate, and measure how often and how badly each pathology bites.
Favorite Betrayal (FB) and Later-No-Harm (LNH) are incompatible, so a method should maximize both statistically rather than pass one and fail the other badly. STAR keeps FB-resistance and gives up LNH; IRV does the opposite — and, to pass LNH, it ignores down-ballot preferences, which (per Dempsey 2018) is precisely why it cannot eliminate vote-splitting or the center-squeeze spoiler.
The Equality Criterion (their framing of one-person-one-vote): for any ballot there exists an opposite ballot that exactly cancels it. Approval, Score, Smith/Minimax and STAR pass; Plurality and IRV fail. They tie this to Wesberry v. Sanders and show it implies the weak mutual-majority criterion.
Results. On a clustered-spatial model (6 candidates, 101–5001 voters): STAR and Smith/Minimax deliver the highest VSE of all methods tested, Approval close behind — roughly STAR ≈ Smith/Minimax > Approval Top Two > Approval > IRV > Plurality Top Two > Plurality. On strategy (PVSI): STAR's only weak incentive (~2%) is to exaggerate scores while keeping honest preference order; all dishonest STAR strategies — favorite betrayal, burial, bullet voting — are strongly disincentivized. IRV's most-incentivized strategy is favorite betrayal; Plurality incentivizes it at ~14%; Approval's strongest is threshold-setting at ~10%.

The kernel — what's solid (concede it)¶

Genuinely peer-reviewed, in an economics journal, and open access — a real citable anchor, not a campaign PDF.
VSE and PVSI are rigorous, reusable and model-transparent. Appendix B runs robustness checks across models and parameters, and the Python code is supplied. The VSE ordering (cardinal/Condorcet methods over IRV over Plurality) is robust and independently corroborated — see expert consensus & IRV.
Unusually candid about its own bias risk. The authors state they "have done our best to avoid creating a model biased in favor of STAR," and note that Quinn's VSE work predated his STAR activism — he had expected the data to favour Bucklin-style Majority Judgment, not STAR. They deliberately chose a clustered spatial model so that Condorcet cycles occur at a realistic rate: neither the flood produced by impartial culture nor the near-zero of a plain spatial model. That is exactly the methodology this repo argues for.
The vote-splitting / center-squeeze critique of IRV is correct and load-bearing — and it has actually happened: Burlington 2009, Alaska 2022.

Test it — the runnable companions here¶

The paper's claim	Run it in this library
STAR reads strength and order — three "winners" can differ	Three notions of winner
IRV's center-squeeze spoiler, on real ballots	Center squeeze · Alaska 2022 · Burlington 2009
The FB/LNH tradeoff — and that STAR is not FB-proof	Favorite betrayal · FBC simulation
STAR ≈ Smith/Minimax at the top → so where do they disagree?	Ranked Robin · 30 STAR≠RR divergence samples
VSE/PVSI are model-dependent	Election simulation models · the simulations

Read it critically — overreach, and the lean¶

Disclose the lean. Lead author Sara Wolk directs the Equal Vote Coalition, the STAR advocacy organisation; Quinn and Ogren are longtime cardinal-voting researchers. This is advocates presenting their own method. Their candour (above) softens that but does not erase it.
The precision is oversold; the ordering is not. Headline VSE percentages move with the voter model, candidate count and how strategic voters are (Appendix B). The ranking is robust; the exact figures are not constants — the same caution the scorecard already prints.
STAR does not strictly dominate — and the paper's own data says so. On their results a Condorcet method (Smith/Minimax) ties STAR on VSE and beats it on strategy-resistance: Smith/Minimax disincentivises strategic voting across the board, while STAR retains a small (~2%) exaggeration incentive. The honest reading is "STAR is co-best, alongside a good Condorcet method," not "STAR wins outright." That is precisely the STAR-vs-Ranked-Robin tension this library teaches — and here it is a point for Ranked Robin.
The "Equality Criterion" is an advocate-introduced yardstick. Worth the criterion-built-to-fit-the-method check. In fairness it is not a mirror reverse-engineered to single out STAR — it is a reasonable vote-cancellation property, and several methods pass it including a Condorcet one. But it originates with the proponents, so cite it as their framing, not a settled field-standard criterion.
The infographic ratios are not verbatim in the paper. The familiar "1:1 / 2.6:1 / 2.7:1 / 17.8:1" is Equal Vote's works : backfires presentation built on the VSE simulator; the paper itself reports PVSI percentages (STAR ~2%, Approval ~10%, Plurality ~14%, IRV favourite-betrayal-incentivised). Same underlying simulation — but attribute the ratio form to the infographic and the percentages to the paper.

The honest bottom line¶

A strong, genuinely peer-reviewed contribution that hands the field two durable metrics and argues the accuracy-plus-strategy-resistance case for STAR with more candour about its own lean than most advocacy writing. Its firmest results — cardinal and Condorcet methods beat IRV and Plurality on utilitarian accuracy, and STAR strongly disincentivises dishonest voting — hold up and are corroborated elsewhere. Its softest spots are the usual ones: single-decimal VSE precision reads as more certain than a model-bound simulation can be, and the paper's own numbers show a good Condorcet method matching or edging STAR — so the fair takeaway is STAR is co-best, not uniquely best.

The companion article. The next ten pages of the same issue — Constitutional Political Economy 34(3): 335–345 — are the Approval camp's parallel case: Hamlin & Hua, "The case for approval voting." Same journal, same issue, the rival cardinal method, argued by its own advocacy organisation. It gets the same treatment, and its §4.1 example is runnable here.