Read on Twitter
I've built a model that uses match performance data to infer the relative strength of #PremierLeague teams; it updates itself after each match based on how much better or worse a team performed vs the expectation.
This thread looks at team's Defensive strength following GW10.
This thread looks at team's Defensive strength following GW10.
For those who haven't seen this before, the dots here represent the current view of the number of goals a team is expected to concede against the average PL opponent.
The lines show the movement in this estimate over the last 6 matches illustrating 'speed & direction of travel'.
The lines show the movement in this estimate over the last 6 matches illustrating 'speed & direction of travel'.
Teams are sorted in order from best (top) to worst (bottom), with the number of goals I expect a team to concede increasing from left to right.
The colour of the dots represents whether the team has improved of late (blue), or deteriorated (red); the 'tail' shows by how much.
The colour of the dots represents whether the team has improved of late (blue), or deteriorated (red); the 'tail' shows by how much.
(Those familiar with the charts might notice I've re-formatted them slightly; the aim is to put greater emphasis on the actual current values & rank of the teams, and a little less emphasis on the movement.
Feedback welcomed, as ever.)
Feedback welcomed, as ever.)
Examining the chart, we could say that there are currently four 'islands' (1, 2-5, 6-18, 19-20), between which teams are separated by rather large 'jumps' in quality.
MCI are alone at the top, having become more reserved, whilst LIV have suffered from injuries to key personnel.
MCI are alone at the top, having become more reserved, whilst LIV have suffered from injuries to key personnel.
CHE have also been more solid recently, with Lampard experimenting with the team balance & Mendy helping.
MUN & WOL remain stodgy, elite defensive teams.
Now we jump to the large 'middle class' of defences; TOT are far from vintage Mourinho, but have leapfrogged Ndidi-less LEI.
MUN & WOL remain stodgy, elite defensive teams.
Now we jump to the large 'middle class' of defences; TOT are far from vintage Mourinho, but have leapfrogged Ndidi-less LEI.
I have ARS a little further back; improving defensively as Arteta continues to attempt to slow down & suffocate opponents.
BHA have improved a lot despite not getting the breaks & are rated only a shade below average now. Only a matter of time before the results come?
BHA have improved a lot despite not getting the breaks & are rated only a shade below average now. Only a matter of time before the results come?
Very little separates LEE, WHU, SHU, BUR, SOU and EVE, but it's notable how quickly things can change over six matches, with Everton plummeting from 7th to 15th!
AVL showed great improvement from the restart, but have settled now beside CRY & NEW & are below average defensively.
AVL showed great improvement from the restart, but have settled now beside CRY & NEW & are below average defensively.
Then, there is a very large jump separating them from WBA & FUL, who are somewhat worse than any of their rivals.
As intuition also suggests, these are the teams to particularly target when choosing #FPL attackers & captains - even if it doesn't always work out as we planned!
As intuition also suggests, these are the teams to particularly target when choosing #FPL attackers & captains - even if it doesn't always work out as we planned!
In addition to the Defensive ratings, the model produces the equivalent Attacking ratings, which I will delve into another time.
From these, we can derive Overall strength, style imbalances ('Tilt') & also predict final league placings by forecasting the remaining matches.
From these, we can derive Overall strength, style imbalances ('Tilt') & also predict final league placings by forecasting the remaining matches.
If anyone has any thoughts, questions, suggestions on any of this, please don't hesitate to reach out.
For more detail on how you can build models like this, as well as other high-browed #FPL musings, check out the @uncertainty_pod, which I record with my friend @analytic_fpl.
For more detail on how you can build models like this, as well as other high-browed #FPL musings, check out the @uncertainty_pod, which I record with my friend @analytic_fpl.