Findings / Home advantage
Home advantage in international football, and why the gap grew
Venue-control decomposition: matches are split on the real `neutral` flag into true-home (neutral=False) vs neutral-site (neutral=True), then goal difference and win/draw/away rates are compared within era and tournament type. The neutral split is the natural control -- not a random assignment. Window: 1872-11-30 to 2026-06-16, n=49,425 matches played -- the full committed international-results corpus (~150 years, 1872-2026), not a rolling cache
Home advantage is real; the true-home/neutral gap grew not because true-home rose (it is flat) but because neutral-venue advantage collapsed over time.
Venue split
| Venue | n | Home win | Draw | Away win | Goal diff |
|---|---|---|---|---|---|
| True home | 36,350 | 0.5074 | 0.2286 | 0.264 | 0.6744 |
| Neutral | 13,075 | 0.4418 | 0.2242 | 0.3339 | 0.3007 |
Scroll horizontally for all columns.
Effect — true-home minus neutral: goal-diff effect 0.3737, home-win-rate effect 0.0656.
By era
| Era | True-home gd | Neutral gd | Effect |
|---|---|---|---|
| pre-2000 (n=18,235/5,827) | 0.6743 | 0.4556 | 0.2187 |
| 2000+ (n=18,115/7,248) | 0.6745 | 0.176 | 0.4985 |
Scroll horizontally for all columns.
The effect grew because the neutral-venue difference collapsed, not because true-home advantage rose -- true- home goal diff is flat across both eras (0.6743 to 0.6745), while the neutral-venue goal diff fell from 0.4556 to 0.1760.
By tournament type
| Bucket | n | True-home gd | Neutral gd | Effect |
|---|---|---|---|---|
| Friendlies | 18,388 | 0.5648 | 0.1779 | 0.3869 |
| Qualifiers | 15,928 | 0.7592 | 0.1358 | 0.6234 |
| Finals & continental | 15,109 | 0.7553 | 0.3694 | 0.3859 |
Scroll horizontally for all columns.
The true-home edge is biggest in Qualifiers (0.6234). Neutral games concentrate in Finals & continental play (8,747 of 13,075 neutral matches).
Travel leg — results.parquet carries city/country strings but no coordinates; a travel-distance residual leg is NOT_TESTABLE from committed data and is deliberately not estimated.
Confounds / notes
- neutral games are non-random (concentrate in tournament finals -- 8747 of 13075)
- crowd is never directly observed (NOT_TESTABLE)
- no strength-of-schedule / team-quality adjustment (opponent-raw)
- draws included in win-rate base
- Results describe venue, era, and tournament splits in the published observation window.
This exhibit graduates the 3 mechanisms already confirmed local in the committed mechanism ledger (scripts/platformkit/analytics_showcase/out/mechanism_ledger_export.json) into a published page -- the numbers above are not recomputed here, only re-presented with their receipts.