Wilson intervals, chi-square homogeneity
One-run games and blowouts
Clubs with great records in close games are said to know how to win. A chi-square test asks whether those records spread any wider than coin flips.
Computed on request from the record book (Oct 13, 2026) and 720 finals in the game log.
The finding
23% of this season's 720 finals were decided by a single run. Club records in those games run from KT's 20–10 to Hanwha's 9–22, which looks like a gap in nerve. A chi-square test says it need not be: if every one-run game were a coin flip, records at least this spread out would turn up 6% of the time. Blowouts tell a different story: records in games decided by 5 or more runs spread far wider than chance (p = 0.001) and track the rest of a club's record at r = 0.88, while one-run records track it at r = 0.46 (95% −0.24–0.84). Good teams are good by beating people badly; close games are where the coin lands.
One structural edge does show: home clubs won 60% of one-run games (95% 52%–67%). Batting last means a walk-off ends the game the moment the winning run scores, so close home wins are almost always by exactly one.
One-run games
23%
168 of 720
One-run spread vs chance
p = 0.06
χ² 16.5, df 9
5+ run games vs chance
p = 0.001
χ² 27.0, df 9
Home club, one-run games
.595
.520–.667
Small sample. About 34 one-run games per club. A .500 team's record over that many has a standard deviation of roughly .086.
Final margins this season
Every final by its margin of victory; one-run games highlighted.
One-run record by club, with 95% intervals
Dots are one-run winning percentages with Wilson intervals; diamonds are each club's record in all its other games.
Close games and blowouts against everything else
Each club's record in the rest of its games (across) against its record in one-run games (left) and in blowouts (right), with a least-squares line.
One-run games · r = 0.46 (−0.24–0.84)
5+ run games · r = 0.88 (0.56–0.97)
Method
Every regular-season final in the game log. Ties (17 this season) are left out of every winning percentage. Intervals on each club's record are Wilson score intervals, which behave better than the textbook ±1.96·SE at small counts. The homogeneity test asks whether the clubs' records are more spread out than the pooled rate p̄ would produce by chance:
χ² = Σ (Wᵢ − nᵢ·p̄)² ÷ (nᵢ·p̄·(1 − p̄)), df = clubs − 1
Subtracting the binomial noise from the observed variance leaves an estimated talent spread of 85 points of winning percentage in one-run games, 96 in blowouts and 80 in all non-one-run games. Correlations use Fisher's z for their intervals.
Ten clubs and a few dozen close games each: a test this small would miss a modest real skill, so “consistent with chance” is not proof that bullpens and late-inning managing do not matter. Three tests are run on this page; at the 5% level, the chance at least one clears by luck is about 14%. Extra innings are capped, so some close games end tied instead of being decided.