What makes a voting method good? — criteria beyond the winner¶

Picking a good winner (see What makes a good winner?) is only half of choosing a voting method. The other half is everything around the tally: can voters use it, can citizens verify it, does it invite honest voting and real competition, is it practical to run? This page lays out those criteria — and the reason there's no clean answer: a perfect election system will never exist, so a good method is one that balances these against each other.

→ Level: 201 · deep dive — Curriculum 201.6 (deeper theory at 301). Companion: What makes a good winner?.

"A perfect election system will never exist"¶

Arrow's impossibility theorem (ranked methods) and the Gibbard–Satterthwaite theorem (strategy-proofness) prove that no method can satisfy every desirable property at once. So the goal isn't a flawless method — it's the one that balances the advantages against the disadvantages for the election at hand. Which criteria matter most depends on stakes, number of candidates, single- vs multi-winner, and who's counting.

A consequence worth internalizing (Ka-Ping Yee): because every method must violate at least one of the five classic "reasonable behaviour" criteria, "one can always invent situations where a particular method violates one of these criteria — thus, presenting individual cases of strange behaviour proves little." A single anecdote (or even a handful) rarely settles anything; it's why reformers lean on simulations across many electorates (VSE / Bayesian regret) rather than cherry-picked elections — while still using notorious real failures (Burlington 2009, Alaska 2022) to illustrate how a failure mode shows up in the wild.

The criteria¶

1. Winner-selection quality¶

Does it tend to elect a good winner (majority / consensus-Condorcet / high-support-utilitarian)? Measured empirically by Voter Satisfaction Efficiency. Sub-properties are the classic criteria: majority, Condorcet, monotonicity, independence of irrelevant alternatives, later-no-harm, favorite-betrayal.

2. Simplicity & practicality¶

How hard is it to vote and to count?

Ballot simplicity — can a voter mark it correctly without instruction? (choose-one < approval/score < ranking.)
Tabulation simplicity — can the winner be computed (and re-checked) easily? Round-by-round elimination (IRV), Borda, Minimax, Majority Judgment are heavier than a single sum.
Results viewability — can a citizen look at the published totals and see why someone won?
Machine requirement — hand-countable, or does it need software?

Ballot and tabulation simplicity are two axes, not one — and a method can be excellent on one and poor on the other. Most arguments about whether a method is "simple" are really two people each pointing at a different axis, which is why the question is unanswerable as posed. Placing the methods on both at once makes the trade visible:

Method	Ballot simplicity	Tabulation simplicity
Choose-One	maximal — one mark	maximal — one pile per candidate
Approval	maximal — mark any number	maximal — one sum per candidate
RCV-IRV	high — just rank them	low — rounds of eliminate-and-transfer, not summable, needs central tabulation
3-2-1	high — three words (Good/OK/Bad)	medium — three stages with a tuned threshold
STAR	medium — 0–5 per candidate	high — add the scores, then one pairwise runoff
Ranked Robin	high — just rank them	high — one pairwise matrix, which adds across precincts

Three things follow:

"Simple" claims must say which half. "RCV is as easy as 1-2-3" is true of the ballot and false of the count — which is why the repo rates it half-true rather than false.
The two axes are paid for by different people. A voter meets the ballot once, in private, with instructions in front of them. Officials, auditors, journalists, courts and opponents meet the count repeatedly, forever. So ballot complexity is a one-time explanation cost while tabulation complexity is a permanent operational and rhetorical surface — a reason many reformers weight the second axis more heavily, though it's a judgement call, not a theorem.
Simplicity isn't sufficient. Choose-One is maximal on both axes and is still the method everything here exists to replace. Simplicity is a constraint to satisfy, not the objective to maximize.

3. Summability (precinct-level counting)¶

Can each precinct publish a fixed-size tally that adds up to the national result, or must every ballot flow to one place? STAR and Approval are summable (a small matrix / totals per precinct); RCV-IRV and STV are not (transfers need the full ballot set centrally), which delays results and complicates audits. See summability (topic hub).

4. Transparency & auditability¶

Can the public understand the process from ballots to result, and can it be checked? Risk-limiting audits are far easier on summable, single-round methods than on multi-round transfer methods.

5. Honesty (strategy-resistance)¶

Does the method reward sincere voting, or punish it? A method that pressures voters to betray their favorite (or bury a strong rival) corrupts the very data it counts. Favorite-betrayal and later-no-harm pressures, and vulnerability to burial/compromising strategy, all live here. The VSE simulations put a number on it — a strategy-incentive ratio (how often strategy works vs. backfires for a voter): Plurality 17.8 : 1 (strategy almost always pays), IRV ~2.7 : 1, Approval ~2.6 : 1, STAR ~1 : 1 (strategy is as likely to backfire as to help). Low is good: it means sincere voting is safe. This strategy-resistance is arguably STAR's strongest empirical claim — under honest voting the top methods (STAR, Smith/Minimax/Condorcet, Approval) are close; STAR's edge is that it barely moves when voters get strategic.

6. Competition & representation¶

Does it let all candidates — including third parties and independents — get a clear, unpunished reflection of their support? A method with the spoiler effect / vote-splitting suppresses competition (voters abandon sincere favorites) and, in multi-winner form, the question becomes proportionality — does the elected body mirror the electorate? (see the Pets Governance set).

7. Expressivity¶

Does the ballot let voters say what they mean — degrees of support, equal preferences, honest rankings — rather than forcing a single mark? More expressive ballots capture better data (and are what voters say they want), but must be weighed against simplicity.

8. Viability¶

Even an accurate, expressive method can fail politically if it's too hard to explain (Minimax, Majority Judgment, Schulze, EPR). "Analysis by paralysis": a method nobody trusts because nobody can follow the count won't be adopted or defended. Viability = accuracy × explainability × adoptability.

9. Cost & long-term incentives¶

Cost to run and audit; and the long-term effect on campaigning — does it reward broad coalition-building and sincere platforms, or negative campaigning and polarization?

The trade-offs are the point¶

These pull against each other. The most accurate method may be the least simple (Minimax). The most expressive ballot is not the simplest to mark. Summability (STAR/Approval) buys easy audits but a score ballot asks more of voters than choose-one. A good method is a deliberate balance, chosen for the context:

Priority	Leans toward
Simplest possible ballot & count	Plurality / Approval
Consensus (Condorcet) winner, transparent	Ranked Robin
High-support winner + summable + audit-friendly	STAR
Proportional representation (multi-winner)	STV / STAR-PR

Where reasonable people disagree (the live debates)¶

These are genuine, unsettled disputes — presented as the case each side makes, not as a verdict:

Is VSE even the right yardstick? VSE is utilitarian — it assumes summing satisfaction is the goal. Arrow (and others) argue that cross-voter utility isn't comparable ("my 4 vs. your 4"), so the extra information in a scored ballot is illusory and rankings are all that's legitimate. Reply: even a non-utilitarian usually grants "all else equal, more satisfaction for more people is better," and VSE is transparent about its assumption.
Do scored ballots have a stable meaning? Critics note there's no culture-independent mapping from feelings to a 0–5 score (some people rate everything 4–5, others use 0/5), so a score ballot may be "a ranked ballot plus noise or compression." Reply: real preferences do carry intensity (would you rather lose \$30k, gain \$25k, or gain \$30k?), and a runoff (STAR) or 0/1 (Approval) curbs the worst gaming.
Can strategy be modeled at all? VSE assumes a share of voters use simple strategies; skeptics argue real strategy depends on information and coordination and can't be simulated naively, so cross-method rankings deserve caution. Reply: VSE varied many strategy mixes and the qualitative ordering held; and the strategy-resistance result (who gains from strategy) is less model-sensitive than the raw accuracy gap.
Condorcet vs. cardinal. Condorcet advocates: the pairwise-majority winner has the strongest democratic claim, and rankings dodge intensity-gaming. Cardinal advocates: intensity matters, and score methods are simpler to count/audit and monotonic. Under honest voting they're close; the disagreement is really about strategy and ballot philosophy.
Accuracy vs. simplicity. Some conclude the accuracy gaps among good methods are small enough that simplicity, transparency, and auditability should decide (a case for Approval); others weight expressivity more (a case for Score/STAR). Both are defensible.

Where the STAR case lands (stated as advocacy)¶

STAR-Voting advocates argue it balances the list well: an expressive 0–5 ballot that wastes no data, a two-step tally (score, then automatic runoff) that's summable and audit-friendly, strong VSE, and elimination of vote-splitting/spoilers — "Fair, Accurate, Equal." That's a case, not a theorem; opponents weigh ballot/tally complexity, or prefer a Condorcet guarantee (Ranked Robin) or proportionality (STV). This page's job is the criteria; Why STAR and STAR's honest limits argue the trade-off both ways.

Sources — and which way each leans¶

Almost every method comparison you'll read is written by someone with a preferred answer. That doesn't make it wrong — but read each one knowing where it leans. (Fuller rosters: advocacy organizations · who's who in voting methods.)

Source	What it offers	Leans toward
Equal Vote Coalition · STAR Voting	STAR advocacy; the criteria & "Farewell to Pass/Fail" essays	STAR
Center for Election Science · VSE	Approval advocacy; the VSE simulations	Approval (cardinal)
Center for Range Voting	Bayesian-regret simulations; Score advocacy	Score / Range
FairVote	the largest US RCV-IRV campaign	RCV-IRV
Marcus Ogren — Voting in the Abstract	field-wide method analysis	Condorcet-leaning
electowiki	community method encyclopedia	reform-community (mixed)
Stanford Encyclopedia — Voting Methods	rigorous scholarly survey	neutral (academic)
League of Women Voters (WA) — Review of Election Methods	side-by-side comparison	nonpartisan
Wikipedia — Comparison of electoral systems	criteria compliance table	neutral (encyclopedic)
this repository	STAR-first, but tests STAR honestly against every method	STAR (with honest-limits pages)