How we get the numbers
Everything here comes from public data, through open code, with the rule for each verdict written before looking at the result.
Where the data comes from
- World Cups 1986 to 2022: Fjelstul World Cup Database (CC-BY-SA 4.0).
- 2026 World Cup: OpenFootball (public domain).
- Team strength: each team's Elo rating before every match, computed from every international match since 1872 (a public-domain results database).
- Each source is downloaded from a fixed version, so the same data always gives the same numbers.
How a claim becomes a verdict
Claims are tested in batches. Each batch's list, the measure for each claim and the effect size it would need are written before running. Every claim on the list is published below, including the ones that came to nothing.
Each claim gets one effect and a 95% interval, from resampling matches. When the question is "is this beyond chance?", the reference is a range of World Cups drawn at random from the history.
- True
- the interval sits above zero and both halves of the history (1986-2002 and 2006-2026) point the same way.
- True, borderline
- as True, but the 99% interval reaches zero.
- Myth
- the whole interval sits below the smallest effect the claim implies.
- Can't tell
- neither: the data is not enough.
Every claim tested
Two batches, of 12 and 11 claims. Those on the site link to their page; the rest say why they were left out.
Batch 1
- MythThe first goal is worth more in the knockout.Not on the site: a small gap and a confusing reading (winning at 90 or going through).
- Myth2-0 is the most dangerous lead.
- Can't tellThe worst time to concede is just before half-time.Not on the site: inconclusive.
- Can't tellA late equalizer wins you extra time.Not on the site: inconclusive, with only 37 cases.
- MythA team is vulnerable right after scoring.Not on the site: a weak effect, and the two halves of the history disagree.
- MythA team leading by one sits back.
- MythGoals come in clusters.Not on the site: the result hinges on explaining the restart after a goal.
- Myth, borderlineKnockout teams play for the draw.Not on the site: an ambiguous message.
- Partly true2026 was an outlier World Cup.On the site as "What was new in 2026".
- MythFootball has become tighter.Not on the site: it depended on the roller-coaster claim, which was not published.
- MythSome teams are specialists in deciding late.
- True, borderlineLose the opener and you are out (and the 48-team format eased it).
Batch 2
- Can't tellPenalties are a lottery.Not on the site: the favorite wins just over half of shootouts, too close to 50% to call.
- Can't tellKicking first in a shootout is an advantage.Not on the site: inconclusive.
- MythAt the World Cup, the favorite struggles in the knockout.
- Can't tellThe World Cup has become less predictable.Not on the site: inconclusive.
- True, borderline2026's big wins came from weaker teams.Not on the site: strength explains little of the jump; it appears as a caveat in "What was new in 2026".
- Can't tellPedigree counts: past champions beat expectations.Not on the site: inconclusive.
- TruePlaying at home helps at the World Cup.
- MythA team leading by one sits back (controlling for strength).Not on the site: it revises the reading of the batch-1 myth, whose page now says only that there is no sign of sitting back.
- TrueInternational football has become more of a roller-coaster.Not on the site: a small effect, measured in matches outside World Cups.
- True, borderlineThe kick that must go in trembles.
- Can't tellRest decides knockout matches.Not on the site: inconclusive.
Download the data
The tables the site uses, ready to download. Every rate carries how many cases back it up (k of n).
- summaryMatch metrics by scope and stage (groups, knockout). (480 rows)jsoncsv
- knockout_roundsMatch metrics of the knockout by scope and round. (992 rows)jsoncsv
- group_matchdaysGroup-stage draws and goals by scope and matchday (1-3). (90 rows)jsoncsv
- group_zonesTeam results on group matchdays 2-3 by zone before the match. (60 rows)jsoncsv
- marginsShare of matches by final margin (0, 1, 2, 3+) by scope and stage. (120 rows)jsoncsv
- goal_bandsShare of regulation goals by minute band, by scope and stage. (180 rows)jsoncsv
- stoppage_goalsSecond-half stoppage-time goals (k) per match (n), by scope. (15 rows)jsoncsv
- single_testsThe single externally defined tests (ADR 0002). (1 rows)jsoncsv
- mythsVerdict of each published myth, from a rule fixed before running (ADR 0008). (8 rows)jsoncsv
- myth_two_nilMyth '2-0 is the most dangerous score': not won at 90' after the first 2-0 and after the 2-1 that follows, against teams at 1-0 in the same minute. (2 rows)jsoncsv
- myth_leader_sits_backMyth 'the team leading by one sits back': goals per 90 from minute 76 at 1-0, 0-0 and 0-1. (3 rows)jsoncsv
- myth_late_specialistsMyth 'some teams are specialists in deciding late': share of each team's goals scored and conceded from minute 76, with funnel limits. (96 rows)jsoncsv
- myth_opener_lossMyth 'lose the opener and you are out': share reaching the knockout by era and opener result. (9 rows)jsoncsv
- myth_season_outlierMyth '2026 was an outlier World Cup': each headline of 2026 against a band of same-size draws from all editions. (5 rows)jsoncsv
- myth_home_advantageMyth 'playing at home helps at the World Cup': hosts' result at 90' against the neutral Elo expectation, by edition and host. (14 rows)jsoncsv
- myth_favorite_knockoutMyth 'the favorite suffers more in the knockout': favorite's result at 90' against its Elo expectation, groups and knockout. (2 rows)jsoncsv
- myth_must_score_kickMyth 'the kick that must go in trembles': shootout conversion when a miss eliminates, against the other kicks (1986-2022). (2 rows)jsoncsv
- matchesFull processed table (matches). (708 rows)csvparquet
- goalsFull processed table (goals). (1832 rows)csvparquet
- team_matchesFull processed table (team_matches). (1416 rows)csvparquet
Data for 1986-2022 derived from The Fjelstul World Cup Database © 2023 Joshua C. Fjelstul, Ph.D., licensed under CC-BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), available at https://www.github.com/jfjelstul/worldcup. Modifications: only men's World Cups from 1986, reshaped and aggregated by CopaData (https://github.com/stuaninauts/copadata). Data for 2026 from OpenFootball (https://github.com/openfootball/worldcup.json), public domain (CC0). Team strength (Elo) computed from the international results of https://github.com/martj42/international_results, public domain (CC0).
Corrections
- 2026-10-11: a goal in the England × Argentina semi-final was recorded as 92' (it was 90+2). The 2026 knockout figures were corrected.
- 2026-10-12: the United States and Bosnia appeared under two names, depending on the source. They now have one.
Code and decisions
The code, the tests and the record of each decision are in the repository. See on GitHub →