Solutions · Sectors
Research & teaching
Documented datasets and models, to work on.
Football produces long, clean, public series — rare ground for teaching statistical modelling or putting it to the test. But starting from scratch on collection and cleaning costs time a course or a project does not always have.
What we supply
Results since 2019 across six leagues, attack and defence strengths estimated by season, and the probabilities the model actually published on each matchday. That last point is the most useful and the rarest: having predictions as they stood when they were made, rather than recomputed after the fact, is the only honest way to evaluate a model.
For teaching
The models we use — Maher (1982), Dixon-Coles (1997) — are published and can be reimplemented in a single practical session. Our methodology page describes the full chain: constrained estimation, calibrated time weighting, temporal validation, uncertainty propagation by resampling. Enough to build a course without starting from nothing.
Our terms
Free for academic or teaching use, on a simple reasoned request. All we ask is to be cited, and to see the published work — that is how we find out where our model is wrong.
Why us
We use what we make available. This is not data published once and abandoned: it is our daily working tool, and it gets corrected when it turns out to be wrong.
A few concrete examples
- Results from six leagues since April 2019
- Estimated strengths by team and by season
- The probabilities published on each matchday, not recomputed