Tips — Terminology: RCV vs IRV vs RCV-IRV (and friends)¶
Which word to use when. The whole thing untangles once you separate the ballot from the tabulation — the same distinction that matters everywhere else in STAR.
The one idea that fixes it¶
RCV names a BALLOT. IRV names a TABULATION.
- RCV (Ranked-Choice Voting) = a ballot type: the voter ranks candidates (1st, 2nd, 3rd…). It's a family, not a single method.
- IRV (Instant-Runoff Voting) = one way to count a ranked ballot: eliminate the lowest, transfer, repeat. The single-winner method people usually mean.
- The same ranked ballot can be counted other ways:
- Ranked Robin — a Condorcet / "consensus" tabulation (most head-to-head wins). Sometimes written RCV-RR.
- STV (Single Transferable Vote) — the proportional, multi-winner tabulation.
So when someone says "RCV," they've named the ballot but implied a count. In the US, "RCV" has been hijacked to mean IRV specifically — that's the abuse.
The ranked-method family tree¶
The ranked ballot ("RCV") is counted by a whole family of methods, which split into two branches. Knowing which is which keeps you precise — and keeps you from lumping a Condorcet method in with IRV.
flowchart LR
%% LR, not TD: the 9 leaves stack vertically, keeping the diagram ~3 node-columns
%% wide — a TD layout is ~2000px wide and gets shrunk to unreadable on the site.
R["<b>RANKED ballot</b><br/>('RCV' names the BALLOT,<br/>not any one method)"]
R --> COND["<b>Condorcet</b> / round-robin / pairwise<br/>paired comparison / tournament voting<br/><i>elects the head-to-head winner<br/>when one exists · precinct-summable</i>"]
R --> ELIM["<b>Sequential elimination</b><br/><i>eliminate & transfer ·<br/>order-dependent · NOT summable</i>"]
R --> POS["<b>Positional</b><br/><i>points by rank position</i>"]
COND --> RR["<b>Ranked Robin</b> (RCV-RR)<br/>= Copeland + margin tiebreak"]
COND --> RP["Ranked Pairs · aka Tideman"]
COND --> SZ["Schulze · aka Beatpath"]
COND --> MM["Minimax · aka Simpson–Kramer"]
ELIM --> IRV["<b>RCV-IRV</b> (single-winner)<br/>aka IRV · Hare · Alternative Vote ·<br/>'instant runoff' — and 'RCV' in US usage"]
ELIM --> STV["STV (multi-winner, proportional)"]
ELIM --> CB["Coombs"]
POS --> BO["Borda count"]
POS --> BU["Bucklin · aka Grand Junction"](Copeland is the algorithm under Ranked Robin, so it's folded into that node — and it answers to Llull method too, after the man who described it in 1299; Wikipedia's article opens "The Copeland or Llull method." A Condorcet cousin: "Consensus Choice" — Better Choices for Democracy's variant, same family but a different cycle-resolution rule. A second rebrand of the family, Instant Round-Robin Voting (IRRV), is deliberately left off the tree because it doesn't sit at one level — see the alias table. Plain-text version of this tree is in the git log if a viewer can't render Mermaid.)
True statements that follow (and good ways to test your own precision): - "Ranked Robin, Ranked Pairs, Schulze, and Minimax are forms of Condorcet RCV." ✅ - "Condorcet [methods], Borda, Bucklin, and RCV-IRV are forms of RCV." ✅ (All ranked ballots — though Condorcet is a family, the others are specific methods, so they sit at slightly different levels.) - "Borda / Bucklin are Condorcet methods." ❌ — ranked, yes; Condorcet, no.
Spelling/naming watch: it's Bucklin (not "Buckling"); Hare ≈ IRV for single-winner, but "Hare" strictly usually means STV / the Hare quota.
Aliases — same thing, different name¶
The single biggest source of confusion is that one method has many names. This table maps what you'll hear to what it is:
| You'll hear… | What it actually is | Precise name here |
|---|---|---|
| "RCV" (US / FairVote usage) | the eliminate-and-transfer single-winner method | RCV-IRV |
| "Instant runoff", "Alternative Vote", "Hare" (single-winner) | the same eliminate-and-transfer method | RCV-IRV |
| "Round-robin voting", "pairwise voting", "paired comparison", "tournament voting", "Condorcet" (used as a method) | the family that elects the head-to-head winner — four plain-descriptive names for one everyone-plays-everyone format, plus Condorcet's surname | Condorcet methods (a family, not one method) |
| "Ranked Robin", "RCV-RR" | Equal Vote's Copeland-plus-margin-tiebreak | Ranked Robin |
| "Consensus Choice" | Better Choices for Democracy's Condorcet variant (different cycle rule) | a Condorcet cousin — not identical to Ranked Robin |
| "Instant Round-Robin Voting", "IRRV" | depends who's speaking — electowiki defines one specific method, Wikipedia redirects it to the family, the Condorcet Canada Initiative uses it as a banner, and some sources mean Schulze | ask before you answer — then say Copeland or Condorcet methods, whichever they meant |
| "Copeland" | the win-minus-loss algorithm underneath Ranked Robin | Copeland |
| "Llull method", "Llull voting" | the same algorithm as Copeland, credited to its 1299 author instead of its 1951 one | Copeland |
| "Condorcet candidate", "pairwise champion", "beats-all winner" | not a method at all — these name the candidate who beats every other head-to-head | the Condorcet winner (which every Condorcet method elects when one exists) |
| "majority winner", "majority-preferred candidate", "tournament winner" | also aliases for that same candidate, but not safe swaps — the first two lend a pairwise fact the word majority (a beats-all winner may hold no over-half majority), and "majority-preferred" is scope-ambiguous with STAR's runoff sense; "tournament winner" collides with tournament solutions, which return a set | beats-all winner for the field · majority-preferred finalist for a STAR runoff · majority only for over-half → naming decoder |
| "Beatpath" → Schulze · "Tideman" → Ranked Pairs · "Grand Junction" → Bucklin | older / academic names | as named |
Rule of thumb: when you mean the family, say "Condorcet" or "round-robin"; when you mean the specific Equal-Vote method, say "Ranked Robin." Reserve bare "RCV" for the ballot.
Why the precision actually matters¶
Several of the strongest criticisms are IRV-specific, not ranked-ballot-wide:
- Center squeeze, exhausted ballots, non-monotonicity → these are failures of IRV's elimination tabulation.
- Ranked Robin (Condorcet) uses the same ranked ballot and does not have center squeeze.
So if you say "RCV has center squeeze," a sharp opponent can correctly reply "Ranked Robin is RCV and doesn't." If you say "IRV has center squeeze," you're exactly right. Precision protects your credibility — and it's the same reason you insist STAR critics distinguish "the ballot" from "the runoff."
When to use which word¶
| Situation | Use | Why |
|---|---|---|
| US public audience, naming the thing they know | RCV-IRV (or "IRV — what's usually called RCV") | familiar and precise; signals you mean the eliminate-and-transfer method, not the ballot family |
| Technical / comparison / criticism of the method | IRV | the exact, defensible name; criticisms like center squeeze are IRV's, not all ranked ballots' |
| Naming the elimination rule itself | Hare (fewest-first-choices elimination; single-winner = IRV) | precise for the rule — but note "Hare" also names the STV quota, so single-winner "Hare" ≈ IRV |
| Talking about the ballot / the ranked family | ranked ballots / ranked methods / RCV ballot | reserve bare "RCV" for the ballot, and say so |
| A Condorcet count of a ranked ballot | Ranked Robin (RCV-RR / "consensus") | a different RCV tabulation; do NOT lump it with IRV |
| Technical passage about that count — cycles, tiebreaks, clone independence | Copeland (the naming decoder maps the rest) | "Ranked Robin" no longer picks out one procedure — its own coiner calls it a synonym for Condorcet generally — so name the algorithm where precision carries the argument |
| Proportional multi-winner ranked | STV | the proportional RCV tabulation |
House style for this repo¶
- Default to
RCV-IRVin STAR-vs-method comparisons (the engine's[Divergence from STAR]block and the conversation scripts already do this). It's unambiguous to a US reader and makes clear you mean the method. - Use
IRVin tight technical passages where you've already established you're talking about the eliminate-and-transfer count. - Reserve bare
RCVfor "the ranked-ballot family," and when you use it that way, make it explicit (e.g. "RCV ballots can be counted as IRV, Ranked Robin, or STV"). - Name
Ranked RobinandSTVexplicitly — never fold them into "RCV" meaning IRV. - When quoting how others (mis)use "RCV," keep their word but add the correction once.
The mirror image — STAR's own ballot vs tabulation¶
The same split applies to STAR, which is why this framing is so useful:
- The ballot: a 0–5 score ballot (cardinal).
- The tabulation: Score Then Automatic Runoff.
- The same score ballot can also be counted as Approval (threshold), Score (pure total), or Proportional STAR (multi-winner). "STAR" is one tabulation of a score ballot, just as "IRV" is one tabulation of a ranked ballot.
Teaching tip: once your audience holds "ballot vs tabulation," the whole nomenclature snaps into place — and so does why "RCV" alone is ambiguous.
One level up — "preference profile" and its silent default¶
Theory papers draw an election as a preference profile: one column of preferences per voter (≻₁, ≻₂, …). The term is correct — but unqualified it carries a silent default: strict, complete rankings (linear orders — no ties, no truncation, order only). The default is inherited, not argued: in the Arrow tradition, "preference" is a technical term for a complete, transitive relation — which just is a ranking — so social-choice writing uses "preference" and "ranking" interchangeably. Nothing about voters' actual opinions requires that (the economists' baseline preference relation happily allows indifference); it's the same capture as "RCV" collapsing to IRV — a family word swallowed by its most-studied member. It even shares the etymology: IRV's traditional name is "preferential voting," which is how preference came to mean rank in everyday usage.
So keep three layers apart: the voter's opinion (in their head), the ballot the method lets them file (the method's constraint, not the voter's), and the technical "preference" (a ranking, by Arrow-era definition). A ballot of any type is a constrained report of the opinion — never the opinion itself. This care has top-tier precedent: the Stanford Encyclopedia's Voting Methods entry deliberately says ranking rather than preference where only order is meant, precisely because the technical word misleads.
House style, mirroring the RCV rules above:
- Citing or quoting a paper → keep its "preference profile," with a one-time gloss — e.g. "every voter's ranking (a 'preference profile' — here, strict rankings)." The Alaska 201 page does exactly this.
- When the ballot type matters, name what the profile carries: a ranked profile — strict or with ties (weak orders) — a score profile (0–5; theory akas utility / grade profile), an approval profile (0/1).
- Three axes, not one: strict vs weak (are ties allowed?), complete vs truncated (is everyone ranked?), and ordinal vs cardinal (is strength carried?). The first two are literally PrefLib's file extensions — SOC / SOI / TOC / TOI = {Strict, Ties} × {Complete, Incomplete}; the Burlington 2009 ballots in this repo come from its TOI files — and cardinal data fits none of the four, which is the third axis showing up as a format boundary. The theory default sits at (strict, complete, ordinal) — it simultaneously demands resolution voters may not have (ties must be broken) and discards information they do have (intensity). A score profile allows the ties and carries the strength.
- Don't cede "preference" to ranked ballots. A score ballot also reports preferences — a weak order plus strength ("rank your preferences" is marketing, not a definition). STAR's Equal Support voters are precisely the ones whose preference between two finalists is a genuine tie: expressible on a score ballot, rejected by most strict-rank IRV rules.
The distinction is even load-bearing in this repo's tooling: our pref_voting cross-check loads linear orders as Profile but must switch to ProfileWithTies the moment ballots tie or truncate.
And the ladder extends in both directions. Below the strict ranking sits the theory's "order zoo" — total → top-truncated → weak → partial orders (arXiv:1907.00752, Fig. 1) — cataloguing how much less than a complete strict ranking a real preference can be (their split of why is useful too: intrinsic incompleteness = the voter doesn't have a full ranking, vs epistemic = they have one but it wasn't captured). Above the score ballot sits electowiki's rated pairwise preference ballot — a strength for every head-to-head matchup, generalizing every common ballot type (so expressive a single voter can submit their own Condorcet cycle, i.e. it stops being an "order" at all). Every ballot design is a choice of rung on this ladder — and "preference profile," unqualified, silently names just one of them.
See also: GLOSSARY.md (precise definitions) · CURRICULUM.md · the nomenclature episode "Is It RCV or IRV? Why Do You Keep Saying RCV-IRV?".