Preference vs. Support — the two things a score says that a rank can't¶

The single most useful idea for understanding score ballots. A ranking asks one question — "which do you like more?" A score asks two — "which do you like more?" and "how much do you like each?" The first is preference (order). The second is support (level). Ranks keep the first and throw away the second — which is exactly the information that tells a winner everyone loves apart from a winner nobody wanted.

→ The general form of this distinction: scores vs. ranks. This page is the vivid special case — and, near the bottom, the three comebacks a sharp Condorcet advocate will raise, answered.

The example that makes it click¶

Four candidates. Two ballots:

	A	B	C	D
Ballot 1	1	0	1	0
Ballot 2	5	4	5	4

Same preference: both say A and C over B and D (with A = C and B = D). Opposite support: Ballot 1 says "none of these, really — but if you're making me choose, the odd ones"; Ballot 2 says "honestly, all four are fine — the odd ones by a hair."

Now write each as a ranking. Both collapse to the identical ballot:

A = C > B = D

The ranking cannot tell them apart. The reluctant voter and the enthusiastic voter are recorded as saying the exact same thing. Everything that separated "I tolerate them" from "I love them" — the support — is gone.

And here's the part that makes this a clean example, not a lucky one: the gaps are identical too. Every preference on both ballots is a one-point preference (A is exactly one above B on each). So the two ballots share their order and their pairwise margins — even a hypothetical method that read cardinal margins couldn't split them. The only thing that differs is the level: are these 1s-and-0s, or 5s-and-4s? That's the isolation. This example changes nothing but support, and shows that support is precisely what an order can't hold.

The same thing, as a live election¶

The single-ballot version is the idea; here it is as a real 36-voter election you can re-run — the Preference vs. Support case (live on BetterVoting, BV2225 / BV2226). Two elections with byte-identical rankings — the only change is how hard the two wings score the centrist Blair: a grudging 1, or a genuine 4.

Method	Reads…	Blair scored 1 (tolerated)	Blair scored 4 (supported)	Moved?
RCV-IRV	order only	Alex	Alex	❌ can't
Ranked Robin	order only	Blair	Blair	❌ can't
STAR	order + support	Alex	Blair	✅ only method that responds

Because the rankings never change, IRV and Ranked Robin return byte-identical results in both — they cannot tell a tolerated centrist from a beloved one. STAR reads the support, so it alone moves. That's this whole page, as ballots: same preference, different support, and only the score-ballot method can see it.

The two questions, named¶

Preference — which do you like more? A relative / order question: "A over B." Every ballot format captures at least a slice of it — choose-one only your favorite, approval a two-level cut, ranks and scores the full order.
Support — how much do you like each one, on its own? An independent / level question: each candidate is judged against the scale, not against the others ("A is a 5"). Only a rated ballot answers it — approval's yes/no is the crude two-level case; letter-grade ballots (Majority Judgment) are another.

A score is preference and support in a single mark. A rank keeps only the order. To be strict about it: a rank drops two cardinal things — the level of each candidate (support) and the size of the gaps between adjacent choices (A ≫ B vs. A > B-by-a-hair). This page is about the level; just know the gaps go too.

A note on the word. The election-methods literature usually calls this preference intensity, cardinal preference, or a grade. This library says support because that's what the mark means politically — how much you actually back the candidate. (It's also why STAR's runoff "no-preference" bucket is named Equal Support: scoring two finalists the same.)

The race analogy¶

A ranking is a race that records only the finishing order — not the times, and not whether anyone actually ran fast. A photo finish between two world-record sprinters and a photo finish between two people crawling backwards produce the same order column: "1st, 2nd." A track scout handed only "finished 1st" would demand the clock before signing anyone — but a ranked ballot certifies the win without ever asking. Scores keep the clock.

(One honest seam, since a debate opponent will pull it: a stopwatch is objective, and a voter's 0–5 is self-reported. The analogy shows what ranks discard, not that scores measure the clock perfectly — that cost is the last section.)

Why it matters¶

A mandate vs. a resignation. A candidate can win by preference while having almost no support — the "least-bad" pick of a field nobody liked. Ranks can't tell "won a race everyone wanted" from "won a race everyone wanted to leave." Scores can: the scoring round shows a winner averaging 4.5/5 versus 1.2/5 — same order, very different mandates.
Protest and honest indifference. A score ballot lets you say "I have a slight preference but I support none of you" (1,0,1,0) — or "I'm happy with the whole field" (5,4,5,4). A ranking read as pure order gives you at most one crude support signal — rank vs. don't-rank — and nothing finer: it can't distinguish "my grudging best" from "my genuine favorite."
Consensus vs. compromise. Support is how you tell a broadly-loved consensus winner from a barely-tolerated compromise — even when they share the same preference order.
STAR uses both, on purpose. STAR reads support to pick the two finalists (the scoring round) and preference to choose between them (the automatic runoff). Drop the support half and you're back to a ranked method that can't tell the sprinters from the crawlers. (And yes — the runoff deliberately blinds itself to support again, reverting to pure preference between the two finalists. By then support has done its one job: choosing who is in the runoff.)

The comebacks, answered¶

The example wins against a newcomer instantly. A prepared Condorcet advocate will knock on exactly three doors — so here they are, opened first. (In a debate, the side that concedes first and precisely keeps the frame.)

1. "Allow equal ranks and the ballot already shows A = C > B = D — so what's lost?" That collapsed ballot is already the most expressive ranked format there is — a weak order with ties, exactly what Ranked Robin accepts. The loss has nothing to do with whether ties are allowed. No ordinal format, however permissive, records magnitude — a weak-ranked second choice "could be as good as your favorite or almost as bad as your last choice," and the ballot can't say which. Permissiveness buys you ties, not levels.

2. "With truncation, the reluctant voter ranks A = C and leaves B, D unranked; the enthusiast ranks all four — different ballots, so the rank did tell them apart." True — truncation is a real support signal, but a one-bit one: ranked-vs-unranked is a crude personal approval line ("above my threshold"). It's ambiguous (unranked can mean "unknown" as much as "unsupported"), most tabulations read unranked as simply tied-last (restoring the collapse), and — decisively — it still can't say whether the ranked candidates are loved or merely tolerated. Ballot 2's voter backs B and D at a solid 4/5; no truncation choice can record that. One crude bit is not the level.

3. "Voters will min/max, so support is fiction anyway." The sharpest one, and worth owning: strategically flatten Ballot 1 (1,0,1,0) to 5,0,5,0 — and flatten Ballot 2 (5,4,5,4) to the identical 5,0,5,0. Under full strategic voting the reluctant and the enthusiastic voter cast the same ballot; support is exactly the information strategy erases. Two honest answers: (i) real rated elections show most voters don't flatten — intermediate scores are used heavily; and (ii) STAR specifically penalizes flattening — score your two favorites both 5 and you forfeit your vote between them in the runoff — so STAR's strategic optimum sits much closer to sincere than pure Score's does (strategic voting, STAR's honest limits).

The honest cost of support¶

Support isn't free, and this library doesn't pretend it is. Reading level means trusting voters to use the scale sincerely, and it leans on a shared scale — 0–5 with plain verbal anchors ("great … terrible") is a common language of grades, the same social technology as letter grades and star reviews (Balinski & Laraki's case for grading). Comparing one person's "5" to another's is a real assumption — and one that has taken a recent empirical hit: a 2026 experiment found grade distributions shift substantially with the scale voters are handed, which cuts against the claim that grades carry absolute, shared meaning. So state this trade as an open assumption, not a solved problem.

But notice the trade, because this is where the debate actually sits: pure preference isn't assumption-free either. Majority rule makes its own normative call — that every voter's preference counts equally, however faint: a razor-thin "prefer" outweighs another voter's passionate one. That's the utilitarian-vs-majoritarian choice, and the one-person-one-vote question ("why should intensity override a majority?") is answered squarely by STAR's design: every ballot carries the same maximum influence (giving less than full support is the voter's own choice — the equally weighted vote), and the majoritarian runoff is exactly the check that keeps raw intensity from overriding a genuine majority. Support to find the finalists; preference to confirm the winner. That's the cardinal-vs-ordinal tradeoff — made, not dodged.

Scores vs. ranks — the general form of this distinction (relative vs. absolute)
Weak ranks — even a ranking with ties records order only, never level
The score ballot · the fidelity ladder — a score carries both; converting down to a rank loses the support, converting up has to fabricate it
What makes a good winner? — the utilitarian (support) vs. majoritarian (preference) ideals, and why they can disagree
One person, one vote — the legal standard vs. the reformer's "test of balance," and why weighing support doesn't break it
The equally weighted vote — equal scores say "preference: none — support: whatever level you gave both" (5/5 full, 0/0 none: the glossary's Equal Support / Equal Opposition pair)
Glossary: score ballot