I've had a go at modelling this year's All Ireland football championship

StatsAndSolos

This uses a model I created's view of team ratings and plays through the championship 5000 times to get the %s each team makes each round.

It handles things like All Ireland Series qualification based on a similar model for the league; the draw; home advantage (where known already); seeding from the group stages.

It struggles with extremes particularly with heavy favourites (e.g. it might have 99.9 when perhaps it should have 96-98 because at a certain stage it's tough to get the maths to reflect reality!), and it can be too slow to change its opinion when a team is rapidly showing improvement (e.g. Cork).

If you find it interesting I'll be tweeting periodic updates along with similar content on Twitter at @StatsandSolos (I'm not able to post links)

PARlance

You should be able to automatically post links after 5 posts S&S. Think it's to help prevent bots/spam. Only discovered your account recently, very enjoyable. I'll post a link for anyone else.

https://twitter.com/StatsAndSolos/status/1635288473086341121?t=5B0BJUDI6H2CfjilobdHMQ&s=19

WesternZulu

Mayo finish second...was that an assumption your model had to meet before being validated?

dxhound2005

History means nothing. Derry who have not been in the Final for 30 years, are almost matching Mayo, and are ahead of Dublin. Before the Ulster Championship last year, they were no hopers.

elefant

https://www.boards.ie/discussion/comment/120358057#Comment_120358057

Yes, Derry's near 1/3 chance of making an All-Ireland final seems extremely high.

StatsAndSolos

Yea there are a few things there.

The model is relatively (when I say relatively I'm comparing to bookies' odds) very low on Galway, so that frees up probably too large a % for other teams.

The model considers uncertainty in the range of outcomes given two teams have a certain rating (hence why the best team doesn't win 100% of the time), but it doesn't allow for uncertainty that the rating itself is wrong, which is possibly worthwhile given (a) we're talking months out (b) it's just a model and (c) the sport doesn't lend itself to establishing how teams truly stack up against each other - league not taken fully seriously and championship short, compared to e.g. a 38 game premier league season. So when it thinks one team is better than another it's possibly too sure of it.

It does make an allowance for teams who are currently rated significantly lower (Dublin) or higher (Derry) than their long term average occasionally returning to their longer term position (one way of trying to handle that rating uncertainty), albeit this is a nod to the idea more than a full commital to it. It's not easy in a model that rates off performances to keep Dublin high when they've made a habit of losing games!

Now all that said I post comparisons to bookies' handicaps both on the Twitter page and the website, and it does okay - so far in the league is is wrong vs the actual margin by an avg of 4.8 pts pg vs handicap avg of 4.7. But that is at match level where it has been back tested, not an extrapolation across a full championship like I've tried to do here!