Skip to content

felsenthal_paradoxes — Felsenthal's paradox examples, run for real

Worked examples from Dan S. Felsenthal, "Review of Paradoxes Afflicting Various Voting Procedures Where One Out of m Candidates (m ≥ 2) Must Be Elected" (University of Haifa / LSE, revised 26 May 2010; Leverhulme Trust Voting Power in Practice workshop, Château du Baffy, Normandy). Each example becomes a live BetterVoting election plus LH-engine YAML case(s), tagged by paradox (paradoxes: field) and indexed in the paradox registry. Teaching pages: the voting-paradoxes teaching pages.

Cases

Case What it shows Races Live YAML
BV2144 — Felsenthal Ex.1 Four plurality paradoxes at once (Condorcet winner, Condorcet loser, absolute loser, SCC/spoiler) in a 7-voter election; STAR resolves all four Choose-One → Ana · STAR → Bo results ↗ plurality · star
BV2145 — Felsenthal Ex.2 (1 of 2) Plurality-with-runoff (= IRV for 3 candidates) eliminates the Condorcet winner Ada; + the SCC note (Cleo's exit would elect Ada outright) IRV → Ben · RR → Ada · STAR → Ada results ↗ irv · rr · star
BV2146 — Felsenthal Ex.2 (2 of 2) Non-monotonicity, live: two voters RAISE Ben and Ben loses (elimination flips Ada→Cleo); RR & STAR unmoved IRV → Ada · RR → Ada · STAR → Ada results ↗ irv · rr · star
BV2147 — Felsenthal Ex.3 (I of III) District I (17 voters): runoff deletes Alma, elects Bruno; the district is a Condorcet cycle IRV → Bruno · STAR → Bruno results ↗ irv · star
BV2148 — Felsenthal Ex.3 (II of III) District II (15 voters): Bruno wins round one with a majority (8 of 15); also the Condorcet winner IRV → Bruno · STAR → Bruno results ↗ irv · star
BV2149 — Felsenthal Ex.3 (III of III) REINFORCEMENT paradox, live: Bruno won both districts, the amalgamated 32 voters elect Alma (17–15); STAR stays with Bruno IRV → Alma · STAR → Bruno results ↗ irv · star
BV2150 — Felsenthal Ex.4 (1 of 2) Full 11-voter electorate: the runoff deletes the Condorcet winner Beth and elects Carl — the Andy voters' worst choice IRV → Carl · RR → Beth · STAR → Beth results ↗ irv · rr · star
BV2151 — Felsenthal Ex.4 (2 of 2) NO-SHOW paradox, live: two Andy voters stay home and their side gets Beth instead of Carl; read in reverse = the weak TWIN paradox; RR & STAR unmoved IRV → Beth · RR → Beth · STAR → Beth results ↗ irv · rr · star
BV2152 — Felsenthal Ex.5 Approval misses the Condorcet winner (Felsenthal & Maoz 1988): mixed approval cutoffs elect Anna over Bert, who beats everyone head-to-head Approval → Anna · RR → Bert results ↗ approval · rr
Felsenthal Ex.6 (LH-only) Approval can elect a PARETO-DOMINATED candidate: top-three cutoffs tie Aria/Beau 3–3 though every voter prefers Aria; adversarial lot exhibits the Beau branch (random on BV → not freezable) Approval → Beau (by lot) · RR → Aria approval · rr
BV2153 — Felsenthal Ex.7 ABSOLUTE MAJORITY paradox: Amos is 51 of 100 voters' first choice, everyone approves top two, Bella (100 approvals) wins; IRV & RR elect Amos Approval → Bella · IRV → Amos · RR → Amos results ↗ approval · irv · rr
BV2154 — Felsenthal Ex.8 The absolute loser wins Approval on ONE strategic ballot — and the three races give THREE winners (April / Clara / Bruce) from one electorate Approval → April · IRV → Clara · RR → Bruce results ↗ approval · irv · rr
BV2160 — Fishburn Ex.14 Borda's TRUNCATION paradox on paper (C 20 → A/B 16 tie when three voters truncate); live races split three ways with Borda: STAR → B, Choose-One → A, Borda → C STAR → B · Plurality → A results ↗ star · plurality
BV2161 — Felsenthal Ex.15 Borda's SCC paradox on paper (C wins; B exits → A) — arithmetic-corrected profile; STAR & Choose-One agree on C STAR → C · Plurality → C results ↗ star · plurality
BV2162 — Nurmi Ex.16 (1 of 2) Full sincere rankings: RCV-IRV elects A over the Condorcet winner B; STAR & RR elect B STAR → B · IRV → A · RR → B results ↗ star · irv · rr
BV2163 — Nurmi Ex.16 (2 of 2) TRUNCATION paradox, live: 17 voters rank only D, their ballots exhaust, and B (their preference over A) wins; STAR & RR unmoved STAR → B · IRV → B · RR → B results ↗ star · irv · rr
BV2164 — Coombs Ex.17 Coombs (paper) deletes the CONDORCET WINNER Arlo first — last-ranked by the most voters — and elects Bree; STAR & RR elect Arlo STAR → Arlo · Plurality → Bree · RR → Arlo results ↗ star · plurality · rr
BV2165 — Coombs Ex.19 (1 of 2) Full 15-voter electorate: Coombs (paper) → Boone; STAR → Boone (runoff 8–7), Choose-One → Cass; cyclic profile (no RR/IRV races) STAR → Boone · Plurality → Cass results ↗ star · plurality
BV2166 — Coombs Ex.19 (2 of 2) NO-SHOW paradox, live — and STAR FLIPS TOO: two abstainers get their favorite Cass under Coombs (paper) and under STAR's runoff (39/37/41, 9–4); first live score-family no-show case, shown honestly STAR → Cass · Plurality → Cass results ↗ star · plurality
BV2167 — Minimax Ex.29 Minimax (paper) elects D — the Condorcet AND absolute loser, who loses every matchup 5–6 but never badly; Choose-One agrees (5 first choices); STAR elects the cycle member B STAR → B · Plurality → D results ↗ star · plurality

The appendix is now fully covered: §A1 (Plurality, Ex.1), §A2 (plurality-with-runoff, Ex.2–4), §A3 (Approval, Ex.5–8), §A4 (successive elimination, Ex.9–12 — worked tables, no elections: the procedure exists in neither BetterVoting nor the LH engine), §A5 (Borda, Ex.13–15 — Ex.13 is BV2153's electorate and lives as a Borda section on that page; Borda counts are paper-only for the same reason), §A6 (RCV-IRV, which Felsenthal calls the 'Alternative Vote'; Ex.2–4 reused + Nurmi's Ex.16), §A7 (Coombs, Ex.17–22 — worked tables, with Ex.17 and the Ex.19 pair live as BV2164–66), §A8 (Range Voting, Ex.23–24 — paper only: the 1–10 grades don't fit BV's 0–5 ballot), §A9 (Majority Judgment, Ex.25–28 — paper only, per Felsenthal & Machover 2008), and §A10 (Condorcet/Minimax, Ex.29–33 — worked tables, Ex.29 live as BV2167). New examples from other sources continue through the standard BV-backed case workflow.

The procedures nothing here could count — now runnable

§A4, §A7, §A8, §A9 and §A10 examine five procedures no tabulator in this repo could count, so for a long time their examples existed only as ballot profiles written out in a sentence. They are now runnable case files, counted by four tools in pref_voting_tabulation_engine/coombs_report.py, minimax_report.py, successive_elimination_report.py and grade_methods_report.py — each cross-checked against Eric Pacuit's pref_voting where the library has a counterpart. All 27 counts reproduce Felsenthal's published results, with three documented exceptions (Minimax Ex.31, successive elimination Ex.10, and MJ Ex.26, which has no data here to run).

None are BV-backed: BetterVoting offers none of these methods, and several are deliberately indeterminate (Minimax Ex.32, Coombs Ex.21 and successive elimination Ex.10 all end in genuine ties, which is the lesson). Ranked cases are tabulated under a supported method for their _tabulated mirror — Minimax and successive-elimination cases as Ranked Robin, because the pairwise matrix is exactly what both consume, and Coombs cases as RCV-IRV, the mirror-image count Coombs inverts. The five grade-ballot cases carry a grades: block instead of ballots:, keeping Felsenthal's 1–10 and A–J scales, which fit neither the engine's 0–5 validation nor BetterVoting; they have no mirror and no generated page, and their counts live on the concept pages.

Case What it shows Under its own rule LH count
Coombs Ex.18 NON-MONOTONICITY: four voters raise Bree and Bree loses — Dana is deleted first instead of Arlo Coombs → Arlo (was Bree) IRV → Arlo
Coombs Ex.20 — District I 34 voters; A deleted on 14 last places, B takes 18 of 34 Coombs → B IRV → B
Coombs Ex.20 — District II 7 voters, B has an outright majority (source prints "6", lists 7 — typo flagged) Coombs → B IRV → B
Coombs Ex.20 — amalgamated REINFORCEMENT: both districts chose B, the union deletes C and elects A Coombs → A IRV → B
Coombs Ex.21 — before 20 voters; A deleted on 7 last places, B wins 11 of 20 Coombs → B IRV → B
Coombs Ex.21 — after TWIN: two more B-first voters arrive and B's certain win becomes an A/B tie at the final deletion Coombs → tie IRV → B
Coombs Ex.22 SCC: C drops out and A wins on a first-round majority instead of B Coombs → B, then A IRV → A both ways
Minimax Ex.30 — before 19 voters, cyclic; worst losses A 13, B 11, C 12, D 14 Minimax → B RR → A
Minimax Ex.30 — after NO-SHOW / TWIN: three C>A>B>D voters stay home and get A, which they prefer to B Minimax → A RR → D
Minimax Ex.31 TRUNCATION — convention-dependent: Felsenthal's equal-probability reading elects C, this repo's reading keeps B Minimax → C or B RR → A
Minimax Ex.32 — District II 3 voters, D is the outright majority winner Minimax → D RR → D
Minimax Ex.32 — amalgamated REINFORCEMENT: both districts chose D; combined, four of six pairs draw 7–7 and B/D tie Minimax → tie RR → B
Minimax Ex.33 SCC: B drops out and A becomes the Condorcet winner instead of D Minimax → D, then A RR → D
Succ. elim. Ex.9 PARETO + PATH INDEPENDENCE + SCC: agenda D,A,C,B elects B, whom all 11 voters rank below A; agenda A,B,C,D elects D; dropping D elects A SE → B / D / A by agenda RR → A
Succ. elim. Ex.9 — no-show Two D-first voters stay home, the cycle breaks, A wins from any agenda SE → A RR → A
Succ. elim. Ex.10 — District I Three voters, agenda B,D,A,C elects C SE → C RR → B
Succ. elim. Ex.10 — District II One voter; C is trivially the Condorcet winner SE → C RR → C
Succ. elim. Ex.10 — amalgamated REINFORCEMENT, conditionally: all three rounds tie 2:2, so C or B depending on the tie-break SE → C or B RR → C
Succ. elim. Ex.11 — before Six voters; round 1 ties, A advances, C wins SE → C RR → B
Succ. elim. Ex.11 — after TWIN: an exact copy of an existing C-first ballot arrives and B wins instead SE → B RR → B
Succ. elim. Ex.12 — sincere Agenda B,C,D,A elects D — the A>B>C>D voter's last choice SE → D RR → C
Succ. elim. Ex.12 — truncated TRUNCATION: that voter names only A, sits out round 1, and gets A SE → A RR → C

The five grade-ballot cases have no generated page — they are counted by grade_methods_report.py and read on their concept pages:

Case What it shows Under its own rule
Range Ex.23 Four paradoxes, two candidates: one 10 outvotes three majorities RV → y (Condorcet and absolute loser); MJ → x
Range Ex.24 TRUNCATION: V4 leaves y ungraded and x wins (--ungrade y/V4) RV → y, then x
MJ Ex.25 The median elects y though 2 of 3 grade x higher MJ → y
MJ Ex.27 NO-SHOW / TWIN: V1 and V2 abstain and get y (--abstain V1,V2) MJ → x, then y
MJ Ex.28 TRUNCATION: V6 grades only y, x's median falls D → A (--ungrade x/V6) MJ → x, then y