Skip to content

Is RCV-IRV "Just Plurality in Sequence"?

A claim you'll hear from STAR/Condorcet advocates. It's partly right and partly overstated — here's the precise version, so you're never caught out by it.

Applies to: the first-choice-elimination methods — Hare and Contingent/Supplementary, which do tally a plurality each round. It does not describe Coombs (eliminates on last place) or Baldwin/Nanson (eliminate on Borda score) — those read the whole ballot, so they aren't "plurality in sequence." See Which RCV-IRV?.

Someone tells you "RCV-IRV is just a series of plurality elections." It's a confusing claim because there's a true idea buried inside an exaggeration. Sorting the two apart is itself good education.

The part that's true (and worth teaching)

IRV does lean on plurality logic in a specific, defensible sense:

  • Each round is decided by first-preference plurality among the candidates still standing. To choose who gets eliminated, IRV looks only at each ballot's current top choice — it does not consult your lower rankings until a higher choice is gone.
  • The final round is a two-way plurality — whoever the remaining two are, the one with more votes wins.
  • A clean way to say it: plurality is just IRV with a ranking limit of 1. IRV generalizes plurality by adding automatic elimination-and-transfer rounds.

This shared DNA has real consequences. Because elimination ignores lower preferences, IRV is not pairwise (it never asks "does A beat B head-to-head?"), which is why it can:

So the honest, defensible sentence is: "IRV inherits plurality's round-by-round blindness to lower preferences — that's why it isn't pairwise and can eliminate a Condorcet-strong moderate."

The shared DNA has a precise name, too: like plurality, IRV fails the Equal Vote (balance) criterion — there's no ballot you can cast to cancel another voter's, which the glossary calls the structural root of vote-splitting that both methods carry.

The part that's overstated (don't say this)

The stronger slogans — "IRV is the exact same as plurality" or "worse than plurality" — don't survive scrutiny, and an informed opponent will dismantle them:

  • IRV is not identical to plurality. It changes outcomes, and it passes several criteria plurality fails: the majority criterion, mutual majority, later-no-harm, Condorcet-loser, and independence of clones. It typically elects more broadly acceptable winners than first-past-the-post and removes the simplest "split the vote and lose" cases.
  • Calling IRV "worse than plurality" is, in general, false. It's the mirror image of the marketing overclaim that IRV fixes everything. Reach for it and you forfeit the credibility the accurate critique earns you.
  • "A majority voted against the winner" is not rigorous. In any crowded field, almost every winner — under any method, including STAR — is "not the first choice" of most voters. "Not first" ≠ "opposed."

A worked illustration (center squeeze, stated carefully)

A constructed 4-candidate example (first-choice shares: Alice 27%, Bob 26%, Charlie 24%, Debora 23%; with Debora ranked 1st or 2nd by ~93% of voters):

Debora is the broadly-liked compromise candidate, yet she has the fewest first choices, so IRV eliminates her in round 1 — before her wide second-choice support can matter. That is the center-squeeze failure, and it's the legitimate core of the "plurality-like" critique: round-one elimination is a first-preference plurality test.

Caveat for honesty: replaying every later transfer needs the full ballot cross-tabs, which a first/second-choice summary doesn't give. The verifiable, sufficient point is the early elimination of a candidate with broad lower-rank support — not a specific downstream winner.

And to keep it balanced

Methods that avoid round-by-round elimination pay their own price. STAR and Score fail Later-No-Harm and the Majority criterion — a candidate a majority scores highest can still lose — and they invite strategic min/max exaggeration. No method escapes Gibbard–Satterthwaite. The fair claim isn't "IRV bad, STAR perfect"; it's that IRV's elimination mechanism has specific, well-documented failure modes that pairwise and scored methods handle differently. (See scores_vs_ranks.md and strict_vs_weak_ranks.md.)

Learn more — in this library