7,116 shootouts, recounted

The Coin Toss That Wasn't Worth Ten Points

A famous study found that the team kicking first won 60.5% of 129 football penalty shootouts. Recount 7,116 archived shootouts, operate seeded subsets, test a chosen equivalence margin, and keep the post-2003 choice of kicking order visible as a causal limit.

2010 selected sample 78 / 129

60.5% first-kicker wins, as reported by Apesteguia and Palacios-Huerta.

2025 public archive recounting...

The browser is reading the compact match rows now.

How a ten-point lead appears

Each dot is one seeded sample without replacement from the archived rows. At 129, chance has room to make a 60:40-looking split. At 7,116, every recount is the whole archive.

1297,116

Preparing 128 recounts...

old result 60.5%

Choose a dot. Its numerator, denominator, seed, and exact match IDs will appear here.

No recount selected.

Not different is not the same as close enough

The two-sided exact test gives p = .... That does not establish a fair 50:50 mechanism. The next instrument asks a narrower question.

±0.5 points±25 points
Which rules era?
Computing...

This is Pipke's two one-sided t-test at alpha 0.01. Passing means the 98% t interval fits inside the margin you chose. It does not mean the true share is exactly 50%.

Two intervals, kept separate

All 7,116 rows

35%50%65%

Computing intervals...

The exact 95% Clopper-Pearson interval describes estimation at a conventional 95% confidence level. The equivalence verdict uses a 98% t interval because each one-sided test is set to 1%. They are not the same check.

Association

What happened?

The first kicker won 3,570 of 7,116 archived shootouts. This is a row count.

Equivalence

What is excluded?

With a margin of 1.6 points and alpha 0.01, the full sample rejects shares outside 48.4% to 51.6%.

Causality

What would switching order do?

Before July 2003 the toss assigned order. After that, the toss winner chose order. Most archive rows are from the choice era, so the marginal share alone does not identify a switching effect.

The check

The green lines are assertions, not decoration. The second count uses byte-level population counts rather than the row reducer used by the page. The poison line changes known losses into wins and must be rejected by the published anchor.

Free choices and limits

  • The sampling view uses the first 128 values in a fixed xorshift32 seed sequence and draws without replacement. It visualizes sampling variability inside these rows. It is not a bootstrap confidence interval for a wider population.
  • The exact interval is set to 95%. The equivalence instrument follows the paper at alpha 0.01 and lets you choose the practical margin.
  • Pipke separately reports a 1.8-point equivalence result after adding Kocher, Lenz, and Sutter data for a 7,627-row sample. Those supplemental rows are not embedded here, so this instrument does not apply that margin to the 7,116-row archive.
  • The era boundary is July 2003, following the rule history reported in Pipke's online appendix. This archive contains only 40 earlier rows, so that slice is visibly weak.
  • The archive is a researcher-curated scrape from Flashscore, not an official governing-body dataset. Pipke's validity filters are already applied. The compact payload retains only match ID, year, starter side, and first-kicker outcome.
  • The 2022 World Cup final row, ID rBfk6fO7, records France kicking first and Argentina winning. The Euro 2020 final row, ID 4U8yxaPL, records Italy kicking first and winning. These were spot-checked against official FIFA and UEFA match reports, but two checks cannot certify all 7,116 scraped rows.

Sources: Pipke (2025), OSF replication archive, Apesteguia and Palacios-Huerta (2010), and IFAB Law 10. Reproduction notes and the executable verifier are in research/the-coin-toss-that-wasnt-worth-ten-points/.