Center Squeeze¶
One line: a broadly-liked moderate is eliminated early because few voters rank/score them first, so a more polarizing candidate the majority actually opposed goes on to win. It's an RCV-IRV (Hare) failure — a property of the eliminate-the-fewest-first-choices rule specifically, not of ranked ballots in general (and not of every instant-runoff variant); STAR avoids it.
→ Glossary: center squeeze · deeper debate version: Favorite betrayal (301)
Why it happens — and how you know it's Hare¶
IRV only ever looks at each ballot's top remaining choice, and each round it eliminates the candidate with the fewest first choices — that elimination rule is what "Hare" names. A moderate who is almost everyone's strong second choice — but few people's first — has the fewest first-place votes, so the Hare rule drops them before their broad support is ever counted. The two wings survive; the consensus candidate doesn't.
So when a page says "center squeeze," assume Hare. It's a property of the fewest-first-choices elimination rule — the one US "RCV" uses — and it is not a property of ranked ballots in general, nor even of every instant-runoff variant:
- Ranked Robin (a Condorcet count of the same ballot) doesn't eliminate at all, so it has no center squeeze.
- The Condorcet-safe IRV variants — BTR-IRV, Baldwin, Nanson — keep the instant-runoff shape but stop eliminating on first-choices alone, so a Condorcet winner can never be squeezed out. Fittingly, Nanson — quoted at the bottom of this page warning about exactly this flaw in 1882 — designed one of those fixes.
So the precise statement isn't "RCV has center squeeze," or even "IRV" — it's RCV-IRV (Hare). For the full family, see Which RCV-IRV? — Hare and the other variants and the terminology tips.
Minimal test case — run it¶
The smallest clean squeeze, as a matched pair (27 voters, Left / Center / Right):
→ center_squeeze_irv.yaml · center_squeeze_star.yaml
Center is the Condorcet winner (beats Left 15–12, Right 18–9) but has the fewest first-choices (6). The STAR file's output shows all four methods on the same ballots:
Choose-One (Plurality) = Left RCV-IRV = Left Approval = Left
STAR = Center ( = Condorcet winner — also what Ranked Robin would elect )
IRV eliminates Center in round 1; STAR advances Center on strength of support and wins the runoff. (Verified on the engine.) A richer themed version is the Star Wars vote-split demo, 04_star_wars_vote_split.yaml.
A visual example — the voteline 1-D spectrum¶
The squeeze is easiest to see on a left–right spectrum. In the classic voteline demo (zesty.ca/voting/voteline; walkthrough in this video, youtu.be/7btAd1HYvjU?t=1782), three candidates sit on a line — Red (left), Green (center), Yellow (right) — and voters are spread across it. Slide the voters around and you can watch IRV cross a threshold into the squeeze. Here is the configuration where it bites:
red > green > yellow : 33.2% green > yellow > red : 17.5%
green > red > yellow : 13.8% yellow > green > red : 35.3%
First choices: Yellow 35.3%, Red 33.2%, Green 31.3% (fewest). So:
- RCV-IRV eliminates Green first (the center!), its ballots split to the two wings, and Yellow wins 52.9%.
- But Green is the Condorcet winner — it beats Red (66.6 vs 33.2) and Yellow (64.5 vs 35.3) head-to-head. IRV threw out the one candidate a majority preferred over each rival.
- STAR advances Green on its scores (lots of 3s from both wings + 5s from the center) and wins the runoff 65–35.
Run it: center_squeeze_voteline_1d.yaml (scores are a simple 1-D spatial model: own side 5, the adjacent center 3, the far side 0–1; weights are the percentages ×10). The engine's own divergence block:
[Divergence from STAR]
STAR = Green
Choose-One (Plurality) = Yellow (differs from STAR)
RCV-IRV = Yellow (differs from STAR)
Scoring Round Green 3620 (1st) · Yellow 2428 (2nd) · Red 2249
Automatic Runoff Green 645 (65%) vs Yellow 353 (35%) → Green wins
| Method | Winner |
|---|---|
| Plurality | Yellow |
| RCV-IRV / Hare | Yellow ❌ — squeezes out the center |
| Approval · Borda · Condorcet · Ranked Robin | Green |
| STAR | Green ✅ (= the Condorcet winner) |
The threshold, made concrete. This isn't randomness — it's structural. With the voters more centered, Green has enough first-choices to survive elimination and RCV-IRV happens to elect Green too. Shift the electorate so the wings (Red, Yellow) each out-poll the center on first choices, and Green is eliminated first — the squeeze appears. STAR and the Condorcet methods elect Green on both sides of that line, because they read the whole ballot, not just the top choice. (This is also why "RCV ≠ Random" — IRV fails here predictably, every time the geometry repeats.)
Vote splitting vs center squeeze¶
They look alike but aren't. Vote splitting is similar candidates sharing one pool of supporters. Center squeeze is a distinct moderate squeezed by two poles whose voters are different — the moderate can beat each pole head-to-head and still be eliminated for too few first-choices. (Volić, Making Democracy Count, 2024.)
"Core support" doesn't rescue IRV¶
IRV advocates excuse the squeeze by saying the moderate simply lacked "core support" (first-place votes). But the squeezed Condorcet winner can have more first-place votes than the eventual IRV winner and still be eliminated — so the "core support" defense doesn't hold. The worked case: the Ossipoff electorate (BV2158) — candidate C has the most first-choices of anyone (100, vs. the field's 49–53) and beats every rival head-to-head, yet IRV eliminates C and elects D. First-place support was highest and pairwise dominance was total; the elimination order threw both away.
Why it matters: polarization¶
Center squeeze is a depolarization argument, not just a fairness one. A simulation study of candidate incentives (Ogren 2023, Candidate Incentive Distributions, arXiv 2306.07147) finds IRV pushes candidates to court their base far more than opposing voters, while STAR and Condorcet methods reward appealing to opposing-side voters roughly as much as the base — and the gap grows with more candidates. Electing squeezed moderates is how a method lowers the temperature.
Real elections¶
- Burlington, VT 2009 (mayor). Montroll was the Condorcet winner — preferred over Wright 56–44 and over Kiss 54–46 — but had too few first-choices, was eliminated, and Kiss won. IRV was repealed there in 2010.
- Alaska 2022 (US House special). Begich beat both Peltola and Palin head-to-head, but was eliminated first; Peltola won. (Worked through in Favorite betrayal (301).)
How STAR avoids it¶
STAR's scoring round advances the two highest totals, so a broadly-liked candidate (lots of 4s and 5s) reaches the runoff on strength of support, not just first-place counts — exactly the support a moderate has and IRV ignores. STAR is highly Condorcet-efficient: it usually elects the head-to-head winner. Measured on the same sampled electorates, on a 1-D spectrum with seven candidates — the sharpest squeeze conditions — STAR elects the head-to-head winner 74% of the time against RCV-IRV's 52%, and the gap widens as the field grows: Condorcet efficiency, measured.
→ More source notes: RCV-IRV center-squeeze & polarization group in the slide-links index.
Watch it happen (simulation). endolith's elsim has animated 2-D spatial simulations of the worst-case "core collapse" center squeeze. As its write-up puts it:
Under IRV, vote-splitting causes the most-representative candidate to be eliminated first, then the second-best eliminated second, and so on until only the worst two are left in the final round and the second-worst wins… Under TVR [Total Vote Runoff], on the other hand, all voter preferences are included, so the least-representative candidates are eliminated first, transferring their ballots inwards such that support converges to the consensus candidate in the middle of the electorate (who… is the Condorcet winner).
A 140-year-old warning. This isn't new — E. J. Nanson described it in 1882, calling IRV "Ware's method":
Then on the single vote method [plurality] R may win; on Ware's method A, B, C, D, … P, may be excluded one after another on the successive scrutinies, and at the final scrutiny the contest will be between Q and R, and Q, of course, wins… Thus the single vote method may return the worst of all the candidates; and although Ware's method cannot return the worst, it may return the next worst. — Nanson, Methods of Election (1882)
Animations + the full quote: elsim — core-collapse center-squeeze.
Sources: center squeeze (Wikipedia), Ogren 2023, Candidate Incentive Distributions (arXiv), Burlington 2009 (Wikipedia), Alaska 2022 center squeeze (arXiv).