- Prediction Modelling
- Data Analysis
- World Cup 2026
France vs Spain: why better data means a better prediction
Two models, the same semifinal, two different datasets — and how the fuller picture flipped the favourite.
By Jerin Peter Paul
I ran the same models twice on the 2026 World Cup semifinal between France and Spain. The first forecast used only the six matches each side had played at this World Cup. The second added what the form table left out — the real history between these teams and their world rankings — on top of the same tournament data.
The models never changed. Only the data did. And the answer flipped.
The two models
Poisson distribution model. This is the core model. Goals in football are low-frequency count events — exactly what a Poisson process is built to describe. It converts each team's scoring and conceding rates into an expected number of goals, then works out the probability of every scoreline.
Elo rating system. Used as a second, independent check on the outcome. It scores each team's strength purely from results and goal difference, then turns the gap between two ratings into a straight win probability. It does not predict scorelines, but it corroborates or challenges what Poisson says — which is why it was run alongside it, not instead of it.
The six games so far
Each team's six World Cup matches on the road to the semifinal. Goals scored and conceded are the main drivers; the rest is context.
France
| Opponent | GF | GA | Poss % | Shots | On target | Corners | Yellows |
|---|---|---|---|---|---|---|---|
| Senegal | 3 | 1 | 54 | 11 | 8 | 6 | 0 |
| Iraq | 3 | 0 | 56 | 19 | 5 | 4 | 0 |
| Norway | 4 | 1 | 57 | 19 | 10 | 5 | 1 |
| Sweden | 3 | 0 | 61 | 25 | 13 | 9 | 0 |
| Paraguay | 1 | 0 | 76 | 15 | 5 | 12 | 3 |
| Morocco | 2 | 0 | 48 | 22 | 9 | 5 | 0 |
| Totals / avg | 16 | 2 | 58.7 | 111 | 50 | 41 | 4 |
Spain
| Opponent | GF | GA | Poss % | Shots | On target | Corners | Yellows |
|---|---|---|---|---|---|---|---|
| Cape Verde | 0 | 0 | 74 | 23 | 8 | 11 | 1 |
| Saudi Arabia | 4 | 0 | 65 | 21 | 9 | 6 | 0 |
| Uruguay | 1 | 0 | 68 | 5 | 1 | 6 | 1 |
| Austria | 3 | 0 | 65 | 22 | 10 | 9 | 0 |
| Portugal | 1 | 0 | 56 | 15 | 6 | 7 | 1 |
| Belgium | 2 | 1 | 68 | 18 | 8 | 5 | 2 |
| Totals / avg | 11 | 1 | 66 | 104 | 42 | 44 | 5 |
Prediction one: tournament data only
Feeding only these six games into the models, France come out as the marginal favourite. Expected goals land at 1.42 for France and 1.08 for Spain. Over 90 minutes France win 45 percent, draw 26, Spain win 29. Add extra time and penalties and France go through 59 percent of the time.
Prediction one — tournament data only
Result of a single 90-minute match, in per cent
The catch: this rests on six games in which the two teams lost none and drew once. With almost no defeats and zero France-versus-Spain history in the data, the model has little to learn from — so the answer is thin, and easily skewed.
What the form table left out
Now add the real record between these two teams, and their world rankings.
| Head-to-head | Meetings | Spain wins | France wins | Draws |
|---|---|---|---|---|
| All time (since 1922) | 38 | 18 | 13 | 7 |
| Since 2006 | 11 | 7 | 3 | 1 |
| Last 5 meetings | 5 | 3 | 1 | 1 |
| Recent meetings | Competition | Result |
|---|---|---|
| Jun 2025 | Nations League SF | Spain 5–4 France |
| Jul 2024 | Euro 2024 SF | Spain 2–1 France |
| Oct 2021 | Nations League final | France 2–1 Spain |
| Mar 2013 | WC qualifier | Spain 1–0 France |
| Oct 2012 | WC qualifier | Spain 1–1 France |
Spain owned the recent rivalry — winning three of the last four meetings, and the last two in a row. On rankings the sides were level: coming into the semifinal France sat top of the world on 1948.97 points, Spain just behind on 1934.79 — a gap of about 14 points.
Head-to-head since 1922
38 meetings — the history the form table ignored
Prediction two: with the full picture
Blending three signals — each weighted by how much it really tells us — the picture changes. Form leaned France, but that was built against weaker opponents. The head-to-head leaned firmly to Spain. The rankings were a coin flip.
| Signal | What it says | Leans | Weight | France win % |
|---|---|---|---|---|
| Tournament form | France scored more, vs weaker sides | France | 30% | 59 |
| Head-to-head | Spain won 3 of last 4, last 2 straight | Spain | 40% | 38 |
| World ranking | France 1948.97 vs Spain 1934.79 | Even | 30% | 52 |
| Blended | Spain | 48 |
Chance of a France win, by signal
Weighted 30 / 40 / 30 → a blended 48% for France
The result is a near-even call that now tips to Spain: roughly 48 percent France, 52 percent Spain. Adding real data erased France's phantom edge.
Prediction two — with the full picture
Blended win probability
The actual result
Full time — Spain 2–0 France. Dallas, 14 July 2026. A Mikel Oyarzabal penalty and a Pedro Porro strike sent Spain to their first World Cup final since 2010.
The thin-data model backed the wrong team. The fuller data pointed to Spain, and Spain won. If anything, the blend still understated them: Spain had conceded only one goal in the whole tournament before the final.
The models never changed. Only the data did. Feed a model a thin slice of history and it will hand you a confident answer that is quietly wrong. Feed it the full picture and it tells the truth.