Skip to content

RCV-IRV Misconceptions & False Claims — an index

Voting 201/301 · a claim-by-claim index — each common claim mapped to its precise rebuttal

A single place to look up the claims you'll hear about "RCV" and see, for each, what's true, what's oversold, and where the repo rebuts it in detail. This is an index, not the argument — follow the links for the worked examples and numbers. It cuts both ways: the first table corrects RCV advocates, the mirror corrects reform advocates.

Read this first (so you stay fair and precise)

  • "RCV" names a ranked ballot; "IRV" names one tabulation of it. Almost every failure below is IRV-specific — a Condorcet count of the same ballot (Ranked Robin) avoids center squeeze, and STAR uses a score ballot entirely. So the target is IRV, not ranked ballots in general. (See RCV vs. IRV vs. RCV-IRV — the naming problem, or the conversation version.)
  • Most of these are oversold, not flatly false. Each has a true core — concede it cleanly, then correct the overreach. That candor is what makes the correction land.
  • Don't open with this. It's for when an advocate makes one of these claims; correct once, don't fight the wind.

The claims

Claim you'll hear Verdict The precise correction (one line) Rebutted in
"If your favorite can't win, your next choice will be counted." False Only if your current candidate is eliminated and a later choice still survives; if your favorite reaches the final two and loses, your lower ranks are never read. Depends on elimination order you can't see. Exhausted / inactive ballots §2–3
"It's safe to vote your conscience / you never have to betray your favorite." Oversold That's Later-No-Harm (true of IRV). The favorite-betrayal promise is a different criterion (FBC), which IRV fails — center squeeze, Alaska 2022 / Burlington 2009. Favorite betrayal (301)
"Your vote won't be wasted / no wasted votes / every vote counts." Oversold Inactive (exhausted) ballots leave the count; real IRV elections shed ~10–27% by the final round (Burnett & Kogan 2015). Exhausted / inactive ballots §2,4
"Winners have a true majority." False as stated It's a majority of remaining/active ballots, not of all cast. Alaska 2022: the winner had 48.4% of round-1 ballots. Exhausted / inactive ballots §4
"RCV is as easy as 1-2-3 / simple." Half-true The ballot is simple to mark; the count is rounds of eliminate-and-transfer — not summable, needs a central tally and software. Is RCV "simple"? · Tabulation, step by step
"RCV is non-polarizing / elects the consensus candidate." Oversold IRV center-squeezes the broad-consensus candidate (eliminated early for too few firsts) — Burlington, Alaska. A Condorcet count or STAR elects the consensus; IRV needn't. Center squeeze
"RCV ends the spoiler effect / eliminates vote-splitting." Oversold It tames the ordinary first-round spoiler, but reintroduces a spoiler-shaped failure via center squeeze: a third candidate's presence can flip the winner (Burlington 2009). Center squeeze · What's so good about STAR
"Everyone gets to vote again each round — still one-person-one-vote." Misleading Only ballots whose current candidate is eliminated move; a fully-ranked ballot whose favorite survives never has its lower preferences acted on. "Counts again" ≠ "your full ranking is used." Exhausted / inactive ballots §3
"RCV is easy to count / audit (precinct-summable)." False IRV is not summable — who's eliminated depends on the whole electorate, so ballots must be counted centrally and a recount re-runs the rounds. Lack of summability · Central tabulation
"RCV breaks two-party domination." Disputed / unproven Where IRV has been used longest (Australia's House, a century) two-party dominance persists; center squeeze actively disadvantages the third/center candidate. Treat as an empirical claim, not a given. Center squeeze (polarization notes)

The mirror — claims our own side oversells about IRV

The table above corrects RCV advocates. This one corrects us. Reform arguments have their own overreach, and an opponent who knows the literature will find it in seconds — so audit your own lines first. (Same discipline as STAR's honest limits; same reason Alaska 301 keeps the rarity numbers in view.)

Claim you'll hear reformers make Verdict What survives cross-examination See
"Forcing voters to take sides destroys consensus — so IRV is anti-consensus." Half-right, wrongly aimed The forced ordering is a property of the ranked ballot; Ranked Robin counts the identical ballot and finds the consensus. What discards consensus is sequential elimination + later-no-harm, not ranking. Say "IRV has no mechanism for finding the consensus candidate," not "ranking destroys consensus." §below
"IRV usually elects the wrong winner." False Condorcet failures showed up in 2 of 182 US RCV elections (Graham-Squire & McCune). The failure is predictable in close three-ways, not common. Overstating frequency is the fastest way to lose a technical audience. Alaska case studies
"IRV is worse than what we have now." False Choose-One forces sides far harder — one mark, no later preference at all. IRV is a real improvement on that axis; the argument for STAR is further, not back. RCV-IRV and plurality
"STAR eliminates favorite betrayal / vote-splitting / the spoiler." Oversold Resistant, not proof — STAR fails FBC in rare constructions and keeps a residual chicken-dilemma. Concede it before they raise it. STAR's honest limits §2, §4
"IRV ignores ~100% of the information on the ballots — Approval, Score, Condorcet and Borda ignore none." Right mechanism, broken yardstick The mechanism is real and it's the best thing on this page: IRV reads only each ballot's top surviving choice, so your lower ranks are invisible while your favorite lives. But the theorem is asymptotic in candidate count, and real races have three to ten. And "Approval ignores none" gives the yardstick away — Approval is the least expressive ballot in that list, and "ignores nothing" only because the ballot discarded the information before the count saw it. By that accounting, Choose-One plurality also ignores nothing: it faithfully counts every mark on every ballot. A measure that ranks plurality as maximally information-efficient is measuring the wrong thing. Use the mechanism, drop the metric — and note that the blindness is what buys Later-No-Harm, which STAR lacks. the claim-check · what your 2nd choice does

The flagship: "IRV is anti-consensus"

The valid kernel — and it's stronger than the usual phrasing. IRV is designed to satisfy Later-No-Harm: adding a lower preference can never hurt your favorite. The only way to deliver that guarantee is to refuse to read your 2nd choice until your 1st is eliminated. But "consensus" simply is the aggregate of those lower preferences. So the tension isn't rhetoric, it's a design fork: a method that protects your later ranks from ever helping someone else must also decline to use them to find the candidate most voters agree on. IRV took Later-No-Harm; STAR and Ranked Robin took the agreements. (Woodall's incompatibility makes the sharp version: no monotonic method satisfies both Later-No-Harm and Condorcet — see the Majority Criterion hub, which works the same fork from STAR's side.)

The observable consequence is center squeeze: the candidate acceptable to nearly everyone collects few first preferences — precisely because nobody had to take sides for them — and is eliminated before those preferences are ever read. Burlington 2009, Alaska 2022.

And ordinal ballots genuinely can't say "these are both fine." A ranking expresses A > B; it cannot express equal support, or the gap between "great," "tolerable," and "intolerable." Most IRV implementations forbid equal ranks, so voters must manufacture a preference they may not hold. That's why STAR's ballot has an Equal Support bucket and why the repo separates preference from support.

Where the claim breaks. The corollary does not follow from the premise. "Forcing voters to take sides" indicts the ballot; "anti-consensus" indicts the count. Ranked Robin is the counterexample sitting in this repo — same forced ordering, opposite behavior, because it reads every pairwise agreement on every ballot. And "destroys" overstates frequency (2 of 182). "Anti-consensus" also implies an active push; structurally indifferent to consensus is what's defensible.

The version that survives cross-examination:

Ranked ballots can't express equal support, and IRV's count discards your later preferences until your favorite is out — that's the price of Later-No-Harm. So IRV has no mechanism for finding the candidate most voters agree on, and in close three-way races it predictably squeezes that candidate out. The fix isn't abandoning ranked ballots — Ranked Robin counts the same ballot and finds the consensus. The fix is abandoning sequential elimination.

That phrasing also does the olive-branch work: it hands an RCV supporter a reform that keeps their ballot.

The umbrella term (so the exhausted-ballot claims stay precise)

"Exhausted" is too fuzzy — it bundles unlike cases. Use inactive ballots as the umbrella, with sub-types: voter-side (overvotes, skipped/undervotes, voluntary bullet voting — roughly comparable to Choose-One) vs count-artifact (involuntary truncation from a ranking limit; exhausted-untransferable; non-exhausted-untransferred). The count-artifact cases hit voters who ranked fully and correctly — that's the asymmetry the vague word hides. Full taxonomy: Exhausted / inactive ballots §2. (Coinages like "stalled / stuck ballot" are useful in teaching but aren't standard — flag them as your own label.)

How to deploy

  • Look up, don't dump. Pull the one row that matches what was claimed; give the true core, then the one-line correction, then the link if they want the worked example.
  • Prefer "oversold" to "lie." You're correcting a sales pitch, not accusing a person.
  • Keep their word. If they say "RCV," keep saying "RCV"; clarify once that the behavior you're critiquing is the IRV count specifically.

Cross-references

file: rcv_irv_false_claims.md