About this site

Trying to answer a "simple" query on future ranks of teams is helping us tackle other questions

We started out simply wondering how far our teams could slip – "what's the worst?" – or rise by the end of the season. Most teams had three matches left that season, and others, four.

But many media pundits were clearly getting it wrong: not seeing how far some teams could rise, and how far some could fall.

So we had a go, just trying to figure out how this could work for a couple of weeks of matches for one league at the end of a season, but soon getting drawn into how it could work for any league at any time.

And just to be clear, for this sort of analysis we're not asking what's most likely – though we do do that in some other analyses – but what's possible.

One reason pundists were getting their numbers wrong is that match results combinations can get complex in a hurry. A weekend of, say, 10 league matches has around 59,000 (3 to the power of 10) combinations of outcomes – win/draw/lose results – before even considering goals scored.

A week of 20 matches has around 3.5 billion possible outcome-combinations. Looking ahead for 30 days, which is what we've settled on, can involves trillions of trillions of possible combinations. And, of course, win-loss results see three points handed out, but draws, just two. Just to add a bit more complexity.

To solve the puzzle, we've had to develop our own approach of cunning heuristics and self-checking & self-correcting algorithms. Note, if we had just wanted what's most likely we could have used, for example, a Monte Carlo-approach of repeatedly tweaking weighted match outcomes to build probabilties for rank results.

Other analyses

What we've built helps with some other analyses.

For example, if a team moves up the table, it's usually assumed that their own performance has improved, and if they go down, that their performance has dropped. In reality, other factors can combine to make an improving team lose places, and a fading team actually go up the table – if just for a few weeks.

So we break down a team's recent performance by the toughness of its schedule, how much the team improved or underperformed in the period, and how much other, close-ranked teams overperformed or underperformed in their own, separate matches. A different story often emerges than what the change of rank suggests.

Our analysis lets you look back over the last 30 days for a team and "bridge" from its rank 30 days ago to its rank today with the stepping stones being the places, up or down, due to the team itself, to the schedule, or to other teams' good or back luck.

We'll add more analyses, and leagues, over time.

We aim to provide analysis that's updated frequently, and in context of some baseline. We won't be doing after-the-event selection of stats to pretend they are a sound explanation of what's occurred for a team or league.

And finally...

Don't take these analyses too seriously. We try to make them meangingful and accurate, but algorithms can, and do, go wrong.