Closest Races in America

Question: What were the closest local races in the 2022 general election?

Method: Aggregate precinct-level results to the contest level, compute margins between the last winner and first loser, rank by margin ascending.

With jq

Aggregate votes by (state, county, contest, candidate), then compute margins. This is easier in Python — jq handles filtering but not multi-key aggregation well.

Quick filter for contests where any candidate has very few votes separating them:

# Find all contests in L4 flat export, group by contest
jq -r '"\(.state)\t\(.county)\t\(.contest_name)\t\(.candidate_canonical)\t\(.votes_total)"' \
  flat_export.jsonl \
  | sort -t$'\t' -k1,3 -k5 -rn \
  > contest_candidates.tsv

With Python

import json
from collections import defaultdict

# Aggregate precinct results to contest level
contests = defaultdict(lambda: defaultdict(int))

with open("flat_export.jsonl") as f:
    for line in f:
        r = json.loads(line)
        key = (r["state"], r.get("county", ""), r["contest_name"])
        contests[key][r["candidate_canonical"]] += r["votes_total"]

# Compute margins
results = []
for (state, county, contest), candidates in contests.items():
    if len(candidates) < 2:
        continue  # uncontested
    ranked = sorted(candidates.items(), key=lambda x: -x[1])
    winner_votes = ranked[0][1]
    runner_up_votes = ranked[1][1]
    margin = winner_votes - runner_up_votes
    results.append({
        "state": state,
        "county": county,
        "contest": contest,
        "winner": ranked[0][0],
        "winner_votes": winner_votes,
        "runner_up": ranked[1][0],
        "runner_up_votes": runner_up_votes,
        "margin": margin,
    })

# Sort by margin ascending
results.sort(key=lambda x: x["margin"])

# Print closest 20
for r in results[:20]:
    print(f"{r['margin']:>6}  {r['state']} {r['county']}: {r['contest']}")
    print(f"        {r['winner']} ({r['winner_votes']:,}) vs {r['runner_up']} ({r['runner_up_votes']:,})")

What We Found

Exact Ties

19 contests nationally ended in an exact tie in 2022. The most striking:

The Dawson County, Georgia school board race is the highest-vote exact tie in the dataset: 25,186 to 25,186. In a multi-seat “vote for 3” contest, this tie occurred between the top two winners — both were elected, so no recount was triggered. But the margin between 3rd place (24,901) and 4th place (24,844) — the actual win/lose boundary — was 57 votes.

Single-Vote Decisions

43 contests were decided by exactly one vote. These are the races where a single additional voter would have changed the outcome. Examples:

Races Within 5%

3,284 contests (approximately 7.2% of all contested races) were decided by a margin of 5% or less. These are competitive races where campaign effort, turnout operations, or ballot design could plausibly have changed the outcome.

The Multi-Seat Complication

For multi-seat contests (school boards with “vote for 3”, city councils with “vote for 2”), the naive margin between 1st and 2nd place is misleading — both candidates may have won. The meaningful margin is between the last winner (Nth place, where N = vote_for) and the first loser (N+1th place).

The Python recipe above computes the 1st-vs-2nd margin. For correct multi-seat analysis, modify the margin computation:

vote_for = r.get("contest", {}).get("vote_for", 1)
if len(ranked) > vote_for:
    margin = ranked[vote_for - 1][1] - ranked[vote_for][1]

The Dawson County tie (25,186 each) is between co-winners. The real margin at the cutoff is 57 votes.

Prerequisites

This recipe requires L4 flat export JSONL with entity-resolved candidate IDs. Without entity resolution, precinct-level records cannot be aggregated to contest-level totals — and ties cannot be detected.