Simon
@analytic.football
Stat nerdery, #lufc and miscellany
Football team ratings model: analytic.football
I chat about FPL @analyticf.pl
Football team ratings model: analytic.football
I chat about FPL @analyticf.pl
Reposted by Simon
No wage regression, the model has a slow moving portion of the ratings (~25-30%) that picks up multi-season performance instead, but if you consistently use the ££ badly then the model won't credit wages independently
November 19, 2025 at 11:48 AM
No wage regression, the model has a slow moving portion of the ratings (~25-30%) that picks up multi-season performance instead, but if you consistently use the ££ badly then the model won't credit wages independently
Another small benefit I should mention is this naturally incorporates weaker priors for promoted teams. Fewer games in the average for the recent errors allows the model to respond quicker where we have less prior evidence -- this is what's driving the change in Sunderland rating
November 18, 2025 at 4:09 PM
Another small benefit I should mention is this naturally incorporates weaker priors for promoted teams. Fewer games in the average for the recent errors allows the model to respond quicker where we have less prior evidence -- this is what's driving the change in Sunderland rating
So how does is impact the ratings? In general, it sees the division with roughly the same shape, though it does moderate positions on Palace, Brentford, Brighton, Bournemouth & Sunderland towards bookie/other model consensus
November 18, 2025 at 4:05 PM
So how does is impact the ratings? In general, it sees the division with roughly the same shape, though it does moderate positions on Palace, Brentford, Brighton, Bournemouth & Sunderland towards bookie/other model consensus
Some other smaller changes include:
- Inclusion of a constant term (representing the missing piece of penalty xG the model chooses not to use)
- This brought down the proportion of a penalty the model wants to use from ~58% (~0.46xG) to ~33% (~0.26xG)
- Inclusion of a constant term (representing the missing piece of penalty xG the model chooses not to use)
- This brought down the proportion of a penalty the model wants to use from ~58% (~0.46xG) to ~33% (~0.26xG)
November 18, 2025 at 4:05 PM
Some other smaller changes include:
- Inclusion of a constant term (representing the missing piece of penalty xG the model chooses not to use)
- This brought down the proportion of a penalty the model wants to use from ~58% (~0.46xG) to ~33% (~0.26xG)
- Inclusion of a constant term (representing the missing piece of penalty xG the model chooses not to use)
- This brought down the proportion of a penalty the model wants to use from ~58% (~0.46xG) to ~33% (~0.26xG)