Skip to content

Filed as Equal-Vote/bettervoting#1407.

Equal Support ballots (incl. an all-5s vote) are being counted as "abstentions" and dropped from the score totals

One-line summary

In a real STAR race, BetterVoting reports 6 abstentions, but those 6 ballots are not abstentions — they are Equal Support ballots (every candidate scored the same), and two of them carry non-zero scores: one voter gave all seven candidates 5 stars, another gave all seven 4 stars. Treating an all-5s ballot as an "abstention" is confusing, and dropping its stars changes the score totals. The winner is unaffected, but the labeling and the score-round bookkeeping are wrong.

The election

A real BetterVoting STAR race, id pet ("What Makes the Best Pet?"), 7 candidates, 461 ballots. Public ballots + both reports:

Evidence

From this election's own export, summaryData reports:

{ "nAbstentions": 6, "nTallyVotes": 455 }

Pulling those ballots out of the 461, all six are "flat" — every candidate gets the same score — and they are:

[0,0,0,0,0,0,0]   ×3    all-zero (cast: "I rate everyone zero")
[ blank      ]    ×1    truly blank — no score recorded (a real abstention)
[5,5,5,5,5,5,5]   ×1    all FIVES — a maximally engaged voter
[4,4,4,4,4,4,4]   ×1    all FOURS

So of BetterVoting's "6 abstentions," only the 1 blank ballot is actually an abstention. The other five are cast ballots, and two of them are about as engaged as a ballot can be.

This single classification explains every difference between BetterVoting and an independent STAR tabulator on the same ballots:

BetterVoting Full count (all 461)
Ballots tallied 455 461
"Abstentions" 6 1 (the blank only)
Equal Support in runoff 92 98
Per-candidate score totals 9 lower (e.g. Dog 1798) (e.g. Dog 1807)
Runoff: Dog vs Cat 190 / 173 190 / 173
Winner Dog Dog

Note the arithmetic closes exactly: 92 + 6 = 98, and the all-5s + all-4s ballots contribute 5 + 4 = 9 stars to every candidate — precisely the uniform 9-point gap in the score totals. The flat ballots' stars are being removed from the scoring round, not just set aside in the runoff.

Why this is wrong (and confusing)

A ballot that scores every candidate the same is Equal Support / no preference, which is a real, deliberate vote — not an abstention:

  1. An all-5s voter clearly participated. Calling that an "abstention" tells the voter (and any auditor) that the ballot was blank. It wasn't.
  2. Those stars should count in the score round. In STAR, the score round is "add every star." An all-5s ballot adds 5 to every candidate; dropping it silently lowers every total. (It doesn't change who finishes top-two here, but it makes BetterVoting's published totals fail to match a hand count of the ballots.)
  3. It only looks harmless because the example is symmetric. Equal Support is correctly neutral in the runoff (a flat ballot has no preference between the two finalists, so it's rightly excluded from the runoff denominator). That is the legitimate idea — but it belongs in an Equal Support bucket, counted in the score round and excluded only from the runoff percentage. Folding it into "abstention" and dropping it from the tally conflates two different things.

The correct STAR treatment (and what the other tabulator does): a flat ballot is counted in the score round and lands in Equal Support in the runoff; only a blank ballot (no scores at all) is an abstention.

Minimal reproduction

Two candidates, and a single 5,5 ballot shows it (the tabulatable YAML):

A,B
5,0    # prefers A
4,0    # prefers A
0,5    # prefers B
5,5    # EQUAL SUPPORT — loves both equally (NOT an abstention)
0,0    # all-zero — cast, supports neither
-,-    # blank — the only true abstention

Expected, correct STAR reading: 6 ballots cast, 1 abstention (the blank). The 5,5 and 0,0 are counted in the score round and sit in Equal Support in the runoff:

Automatic Runoff:  A 2,  B 1,  Equal Support 3
Voters with a preference: 3 of 6 (3 Equal Support). A 2 (67%) vs B 1 (33%).

Confirmed on two controlled BetterVoting elections. Both pick the same winner as an independent STAR tabulator; both show BetterVoting filing flat ballots as abstentions.

(a) Two candidates — the tightest statement (id 3w6v4b, 5 ballots: 0,5 4,0 5,5 5,0, blank). BetterVoting reported nAbstentions: 2, nTallyVotes: 3 — it counted the 5,5 and the blank as abstentions. With two candidates a 5,5 is flat, so it gets flagged directly.

(b) Three candidates — shows it's a systematic flat-ballot rule, not just a 5,5 quirk (id dq2dmm, 8 ballots incl. 0,0,0, 3,3,3, 5,5,0, blank). BetterVoting reported nAbstentions: 3, nTallyVotes: 5 — it counted the blank, 0,0,0, and an engaged 3,3,3 (a voter who scored every candidate 3) as abstentions, but correctly kept 5,5,0 (tied on the two finalists but not flat). So the rule drops a fully-marked ballot while a real "no preference between the finalists" ballot is counted — the two ideas are different, and only the flat one is mislabeled.

Worked write-up + the engine's full report for each (the _tabulated.txt shows the ballots, the runoff, the "1 abstention" note, and the "Runoff math" funnel — i.e. exactly the counts to compare against BetterVoting):

The ask

  1. Don't label Equal Support as "abstention." A flat / no-preference ballot is a cast vote; reserve "abstention" for ballots with no score recorded.
  2. Count flat ballots' stars in the score round, so published totals reconcile with the raw ballots. Exclude no-preference ballots only from the runoff percentage denominator (as "Equal Support"), where exclusion is correct.
  3. If the current behavior is intentional, document it on the result card and in the docs, with the exact definition, so the numbers can be reconciled.

Happy to share the full JSON export, the other tabulator's complete report, or run any case you'd like. The runoff result matched perfectly — this is about labeling and the score-round count, not the outcome. Thanks for BetterVoting.

Cross-reference (our learning docs): frozen evidence BV_result_snapshot.md · synthetic minimal case abstention_reconciliation_min_c2_b6.yaml · small real-BV reproduction recipe SMALL_CASE_reproduce_on_BV.md · full race best_pet_c7_b461.yaml + README.md · concept lessons bettervoting_and_the_engine.md and runoff_percentages.md. This issue: Equal-Vote/bettervoting#1407.