Ranked Choice Voting Effects

What the evidence says

Ranked choice voting (RCV) elections can exhibit mathematical voting-rule failures, and real elections have produced them. Fox (2024) documents that instant-runoff voting (IRV) can exhibit upward and downward monotonicity paradoxes, where a candidate loses after gaining voter support or wins after losing it, and can exhibit a no-show paradox through ballot truncation. A prior review by Graham-Squire and McCune found these failures "uncommon" but present across the elections they examined. Concrete instances exist: the 2022 Alaska U.S. House special election exhibited five distinct failures at once (an upward monotonicity paradox, a no-show paradox, a Condorcet winner failure, a spoiler effect, and a compromise failure), the 2021 Minneapolis election exhibited four (both monotonicity paradoxes, a spoiler effect, and a compromise failure), and a downward monotonicity paradox occurred in 2 of 421 IRV elections examined (0.5%, including the 2020 San Francisco Board of Supervisors District 7 race). Weaver's (2023) theoretical analysis confirms IRV is vulnerable to both types of monotonicity failure but limits its proof to 3-candidate elections and states the frequency of monotonicity failure more broadly remains "a very open question."

Set against those documented failures, large-sample analysis of real RCV elections points the other way on frequency. Katzman (2025) finds RCV almost always selects the Condorcet winner regardless of which of three elimination procedures is used — over 99.4% of 359 elections with a Condorcet winner, with FFP failing in 2 (2009 Burlington Mayoral, 2022 Alaska House), MLP failing in 1, and LBC failing in 0 — and finds RCV rarely violates the independence-of-eliminated-alternatives (spoiler-effect) criterion in real US elections, at rates of 0.008 to 0.051 across all three procedures. In simulation, the same study finds the LBC elimination procedure produces fewer Condorcet-criterion violations than FFP (0.000 vs 0.026 with complete ballots), fewer monotonicity violations (0.107–0.443 vs 0.129–0.548 with complete ballots), and fewer spoiler-effect violations than FFP and MLP (0.102–0.263 vs 0.121–0.374 and 0.135–0.364 with complete ballots). Separately, Heilman's (2022) simulations find Borda count more noise-stable than IRV among methods satisfying the Condorcet Loser Criterion, though this remains an unproven conjecture verified only in limited cases.

In multiwinner settings, sequential RCV and single transferable vote (STV) diverge in ways single-winner comparisons don't capture. McCune (2023), analyzing 1,070 real multiwinner elections, finds sequential RCV chooses a different winner set than STV in 45.3% of cases (485 elections), and where they disagree, the sets differ by exactly one candidate 97.7% of the time (474 of 485). Sequential RCV does not produce results that agree with proportional representation; instead it aligns with an excellence-based paradigm, selecting the Condorcet committee in 97.3% of elections where one exists versus 56.1% for STV, and producing winner sets with more candidate similarity. STV represents more parties in the winner set than sequential RCV in 293 of 1,070 elections, with no elections running the opposite direction. Sequential RCV also raises the share of voters who get all their top-ranked candidates in the winner set (maximal representation) while generally also raising the share of voters with no representation at all (misrepresentation) — a trade-off rather than a uniform gain.

On voter experience, Deshpande (2024), citing Clark's study of Maine, reports RCV decreases voter confidence, satisfaction, and ease of use, with the negative effects outweighing potential positives, and produces no measured improvement in campaign civility. Strategic voting concerns occur in RCV elections but are found empirically rare, not absent, per Graham-Squire and McCune as cited in that same source.

Older transferable-vote mechanics, distinct from modern single-winner IRV, show their own coordination effects. Santucci (2021) describes a bottoms-up multiwinner example where one seat remained unfilled because neither party spread its preferences across candidates, and describes Bucklin voting where independent voters' lower-ranked choices build up major-party vote totals without a reciprocal flow back to independent candidates from major-party voters. Separately, Santucci (2018) states that RCV manufactures an electoral majority in a fragmented candidate field — framed as a design property rather than a quantity measured in that study.

Where the evidence splits

Sources disagree on how much spoiler effects shape RCV outcomes. Deshpande (2024) states the spoiler effect "is shown to be significant" in RCV, citing a separate reference. Katzman (2025), measuring real US RCV elections directly, finds RCV rarely violates the spoiler-effect (independence-of-eliminated-alternatives) criterion, at rates of 0.008 to 0.051 across three elimination procedures, and finds simulated elections show the LBC procedure reduces such violations further relative to FFP and MLP. The claims on this page do not state what accounts for the disagreement — Deshpande's claim rests on a cited secondary source without a described method, while Katzman's is grounded in directly measured election data across candidate counts and elimination procedures — but the sources themselves do not reconcile the two findings, so the split is unexplained by the sources on this page.

Thresholds and dose-response

Reducing the number of available seats to two increases the disagreement rate between sequential RCV and STV winner sets from 45.3% (485 of 1,070 elections, actual seat counts) to 55.4% (593 of 1,070 elections), per McCune (2023).

Candidate count changes whether procedural differences reach conventional statistical thresholds in Katzman's (2025) simulations. With partial ballots, LBC's advantage over FFP on IEA (spoiler-effect) violations is not distinguishable from FFP at 3 candidates but reaches the .05 level at 5 and 9 candidates. On monotonicity violations with partial ballots, no procedure difference is distinguishable at 3 or 5 candidates, but LBC's lower rate than FFP (0.311 vs 0.380) reaches the .01 level at 9 candidates.

Open empirical questions

• What accounts for the disagreement between a claimed spoiler-effect influence in RCV outcomes generally and Katzman's finding that real RCV elections rarely violate the spoiler-effect (IEA) criterion? No source here measures the gap directly.
• Does the excellence-based, Condorcet-committee tendency McCune finds in sequential RCV (measured on real Scottish multiwinner elections) hold in other electoral systems and jurisdictions?
• Does Borda count's simulated stability advantage over IRV hold in real elections, or only in the simulated and conjectural setting Heilman describes?
• Do the voter-confidence and satisfaction effects measured in Maine generalize to other jurisdictions that have adopted RCV?