Alaska 2022 · 301 — every pathology, and is this fair?¶
You've counted the rounds (201). This page does two harder things: (1) shows that Alaska hit several of IRV's known failures at once, and (2) confronts the honest question — is it fair to keep pointing at IRV's failures, when a critic could construct STAR failures too? The second half matters more than the first.
Sources: Graham-Squire & McCune, arXiv:2301.12075 (the empirical database this case is scaled from) · what makes a good winner · run it: the model.
Part 1 — one election, four failures¶
Alaska is the most-cited real IRV misfire because it triggered several distinct pathologies in a single high-profile race. All four are reproducible in the 200-voter model.
1. Condorcet failure. Begich beats both rivals head-to-head (93–84 and 107–68) but isn't elected. The consensus candidate loses. (This is the headline — see 201.)
2. Spoiler effect. Remove the losing candidate Palin from the ballots entirely, and Begich wins (his voters plus Palin's transfers beat Peltola). Palin could not win — but her presence changed who did. That is the textbook definition of a spoiler, and RCV-IRV is marketed as the method that ends spoilers.
3. Upward monotonicity failure. Take the real result (Peltola wins) and give the winner more support: had ~6,000 Palin-only voters (≈6 in the model) instead ranked Peltola first, Peltola would have lost. Why? Those votes pull Palin below Begich in Round 1, so Palin is eliminated instead of Begich; Begich then beats Peltola head-to-head in the final. Ranking the winner higher made the winner lose. A voting method should never punish a candidate for gaining support — IRV can.
4. Majoritarian failure (ballot exhaustion). The winner's 96 votes are 48% of all 200 ballots — a majority only of the 188 that survived after 12 exhausted. IRV's "majority winner" is a majority of remaining ballots, on a denominator quietly shrunk by truncated ballots.
(The paper documents a fifth, the no-show / truncation paradox: some Palin>Begich voters would have gotten a better result by not voting at all. Same root cause — elimination order is fragile.)
Every one of these traces to the same mechanism: eliminating candidates on first-choice counts. Methods that read the whole ballot — Ranked Robin, any Condorcet method, and STAR through its runoff — avoid all five, because none of them gate survival on first choices.
Another way to see it — breadth of support. Working the same Cast Vote Record, STAR's founder Mark Frohnmayer highlights a measure IRV never looks at: how many voters backed each candidate as either first or second choice. A super-majority of Alaskans put Begich in their top two; by contrast a majority expressed no support at all for Peltola, and a majority none for Palin. IRV eliminated the one candidate with broad support precisely because it counts only first choices each round and ignores every "backup" expression until a ballot's top choice is gone — the preference-vs-support blind spot, live in a real election. (Source: Frohnmayer, What the heck happened in Alaska? — a STAR-founder writeup, so advocacy-leaning, but its factual core matches the Clelland and Graham-Squire & McCune analyses this page is built on.)
Part 1½ — what if Alaska had used STAR or Approval? (Clelland's counterfactual)¶
The reduced model above shows STAR electing Begich. Clelland (2023) proves it, working the full Cast Vote Record by hand — and the comparison with Approval is the interesting part, because it isolates what STAR's automatic runoff actually buys you:
- Under STAR, Begich almost certainly wins. He's the Condorcet winner, and STAR's runoff floor (a second choice earns 1–4 stars, never 0) means his overwhelming second-place support carries him. Clelland's conclusion: "Begich would almost certainly win the election" under STAR, and Palin — the Condorcet loser — "could not win this election under any circumstances" (the runoff guarantees no Condorcet loser).
- Under Approval, the outcome is indeterminate. Her Table 5 shows the winner ranges across all three candidates depending on where voters set their approval threshold — it likely favors Peltola, but Begich is realistic, and it can't even structurally rule out the Condorcet loser, Palin.
Same score ballots; the whole difference is the runoff. STAR converts an indeterminate result into a robust, Condorcet-friendly one. (That's also why this case runs STAR and Ranked Robin but not a single Approval race — a lone Approval instance would fake a determinacy the method doesn't have. It's a virtue worth stating plainly, not a knock on Approval, which earns its keep on simplicity — see choosing among the Equal Vote methods.)
Part 2 — is this fair? (the honest reckoning)¶
A fair critic should be squirming a little by now: couldn't someone build a bad-looking STAR election too? Yes. So here is how this library stays honest instead of just cheerleading.
It is fair because we say how rare it is¶
The empirical finding of the very paper this case is built from: across 182 US RCV elections (2004–2022), exactly two had a Condorcet failure — Burlington 2009 and Alaska 2022. So the honest claim is not "IRV fails all the time." It is:
IRV usually works fine. But when it fails, it fails in one specific, predictable, structural way (center squeeze), and that has now happened in real, high-profile US elections.
Stating the rarity out loud is what separates teaching from propaganda — and it's more persuasive, because it's checkable. Explore the other 180 elections that worked fine on Equal Vote's Real RCV tool; the failures are the exception, and that's exactly the point.
It is fair because the same test applies to both methods¶
The library judges every claimed failure by a four-part test. Apply it symmetrically:
| Test | IRV's center squeeze (Alaska) | STAR's usual "gotcha" (favorite betrayal) |
|---|---|---|
| Structural? (a whole region of electorates, not a knife-edge) | Yes — any center candidate between two wings | No — hand-tuned, knife-edge score margins |
| Sincere? (no strategy needed) | Yes — every ballot here is honest | No — requires voters to coordinate a betrayal |
| Realistic? (ordinary politics) | Yes — two wings + a bridge-builder | Rarely |
| Really happened? (a real election) | Yes — Alaska 2022, Burlington 2009 | Not observed in a real STAR election |
The fair conclusion isn't "STAR never fails." It's: STAR's failures are contrived, strategic, and knife-edge; IRV's center squeeze is structural, sincere, and real. A failure that needs a hand-built ballot and coordinated dishonesty is not the same animal as one that shows up in a US House race on honest ballots.
It is fair because STAR's real cost is conceded, openly¶
STAR is not perfect, and this library says so on its own pages. STAR's genuine honest limit is the runoff reversal: a broadly-liked, high-average candidate can lose the automatic runoff to a polarizing candidate a bare majority ranks higher. That's a real philosophical cost (utility vs. majority), documented plainly in STAR's Honest Limits — STAR is also not Condorcet-compliant, not favorite-betrayal-proof, and gives up Later-No-Harm by design. Teaching Alaska next to those admissions is what makes the comparison trustworthy rather than a sales pitch.
So how do you answer FairVote?¶
Not with "STAR is flawless" — that loses. With three moves: 1. Concede STAR's real limits up front (runoff reversal; not FBC-proof). 2. Insist on the symmetric test — is your STAR example structural and sincere and real, or a knife-edge construction? Alaska is the former; most anti-STAR examples are the latter. 3. Cite the empirics — center squeeze has shown up in real US elections; the STAR failure modes have not. Both methods beat Choose-One by a mile; the honest debate is between good options.
Postscript — the same cast, six months later, and IRV got it right¶
The best evidence that this critique is fair comes from the very next election. In November 2022, the same three candidates ran again in the general — and this time RCV-IRV produced a perfectly good result. IRV again eliminated the centrist Begich early, but per Equal Vote's own Real RCV analysis of the general, Peltola was the Condorcet winner this time — the electorate had shifted, she'd have beaten both rivals head-to-head, and IRV elected her. No Condorcet failure.
That is the honest headline, and it strengthens the case rather than weakening it:
- IRV's failure is conditional, not guaranteed. The center-squeeze mechanism (eliminate the centrist on first choices) fired in both elections — but it only produces a wrong result when the eliminated centrist was actually the Condorcet winner. In August, Begich was; in November, he wasn't.
- So "IRV is broken" is the wrong claim. The right one: IRV has a structural failure mode that triggers in a specific, identifiable configuration (a genuine centrist Condorcet winner with few first choices) — and that configuration really occurred, in Alaska, in August 2022.
- This is exactly why we insist on stating rarity: of these two back-to-back Alaska races, one was a Condorcet failure, not both. A critic who claimed otherwise would be wrong, and we'd lose the argument. Precision is the whole game.
(Equal Vote attributes the November shift to a larger, more left-leaning general electorate and an "electability bias" that moved from Palin — the best-known name in August — to Peltola, the incumbent by November.)
Run the counterpart: the general is a runnable model too — same cast, but all four counts agree on Peltola. Put side by side, the pair is the fairness argument: one election where IRV failed the Condorcet winner, one where it elected them, same mechanism, opposite outcome.
Challenge your understanding¶
Q1. Explain the upward-monotonicity failure in your own words: how can giving Peltola *more* first-place votes make her lose?
Extra first-place votes for Peltola have to come from *somewhere* — here, from Palin-first voters switching. That drops **Palin** below Begich in Round 1, so Palin (not Begich) is eliminated first. Begich then survives to the final and **beats Peltola head-to-head**. The elimination *order* flipped, and the new order is bad for Peltola. IRV's outcome depends on fragile elimination order, so more support can backfire.Q2. A FairVote advocate shows you a STAR election where a voter benefits from dishonest scoring. Give the three-move response.
(1) **Concede** — yes, STAR isn't strategy-proof or favorite-betrayal-proof; no method with 3+ candidates is (Gibbard). (2) **Apply the symmetric test** — is your example *structural and sincere*, or a knife-edge that needs coordinated dishonesty? STAR's failures are typically the latter; Alaska's center squeeze is the former (structural, honest ballots). (3) **Cite the empirics** — center squeeze has occurred in real US elections; STAR's failure modes haven't been observed in one. The honest debate is between two good methods, both far better than Choose-One.Q3. Why is it *more* persuasive — not less — to admit that Condorcet failures are rare (2 in 182 elections)?
Because it's checkable and it targets the real claim. Overstating ("IRV always fails") is easy to refute and reads as propaganda. The true claim — "IRV usually works, but its failures are *structural and predictable*, and have hit real high-profile races" — survives scrutiny, and it reframes the argument as "why adopt a method with a known structural failure mode when STAR/Ranked Robin avoid it?" Rarity + predictability beats exaggeration.Q4. Name STAR's own honest costs. Why include them on a page criticizing IRV?
STAR is not Condorcet-compliant, not favorite-betrayal-proof, gives up Later-No-Harm by design, and has the *runoff reversal* (a high-average consensus candidate can lose to a polarizing bare-majority favorite). They belong here because a comparison that hides one side's costs isn't a comparison — it's advertising. Conceding STAR's limits is exactly what earns the Alaska critique its credibility.Q5. All four methods were run on the same electorate. Which failed the Condorcet winner, which didn't, and what single design choice separates the two groups?
**Failed:** Choose-One and RCV-IRV (both → Peltola). **Passed:** Ranked Robin and STAR (both → Begich). The dividing line is **whether the method eliminates/decides on *first choices only* (Plurality, IRV) or reads the *whole ballot* (Ranked Robin's head-to-heads, STAR's scores + runoff).** First-choice elimination is the step that squeezes out the consensus candidate.Q6. In the November general election, IRV again eliminated the centrist Begich — yet there was no Condorcet failure. Why not, and what does that teach us?
Because in the general the Condorcet winner was **Peltola**, not Begich — the electorate had shifted, and Peltola would beat both rivals head-to-head. IRV eliminating Begich no longer removed the "beats-everyone" candidate, so IRV's winner (Peltola) *was* the Condorcet winner. The lesson: the center-squeeze **mechanism** fires whenever a centrist has few first choices, but it only causes a **failure** when that centrist is the Condorcet winner. IRV's failure is conditional — which is exactly why we state its rarity precisely rather than claiming IRV is always wrong.Back to the model & four counts · 101 · 201 · the prose write-up with real vote totals: RCV-IRV case study.