The recent explosion of data in football has allowed for a greater detail of understanding of several areas. The primary one I’ll be focusing on is the evolution of Assists.
Assists in it’s own right is quite a poor dataset. Whereby Goals at least tell you something of the quality of the player in question, Assists are completely reliant on another player to be registered. Imagine Lionel Messi playing alongside Emile Heskey and laying chances on a plate for him – would he have as many Assists in this case as he does when playing alongside Suarez? It’s likely he wouldn’t.
The evolution of Expected Goals (xG) into the mainstream – even seeing it included on Match of the Day, albeit without any explanation on the show itself – means that Expected Assists (xA) are now a semi-accepted part of Football Data as well.
These are a much better look at how we capture those creative players. If Messi now sets up Heskey and he misses, it matters not, as Messi would get the credit for the expectation of Heskey scoring on a repeatable basis from a similar chance. Likewise, if Messi – great player though he is – is aided by some top notch finishing from Suarez, we’d still be able to see that perhaps he was over performing in this regard, compared to how many Assists we’d expect him to have.
My question is this – are current methods of capturing Expected Assists working as well as they could be? I think the answer is no – and I’ll explain why StrataBet do things differently.
What does StrataBet do differently?
The previous method used in collecting StrataData involves a slightly different way to the norm of logging assists. The objective way of doing this credits only those cases where the player who shoots receives the ball directly from a player on his own team. So a slight deflection? Assist doesn’t count. Great run past 4 players before being tackled with ball rebounding to his own teammate? Assist doesn’t count. While I can see the point behind this, if we’re looking at pure repeatability then surely the player who has created the opportunity to shoot should be given some credit.
It’s clear footballers take notice of their assists – and can often have performance bonuses linked to them – as this article and video showing Cesc Fàbregas having one taken away from him shows.
That is why StrataData focuses on Creation, rather than Assists – so we would reward players for creating opportunities on a repeatable basis rather than just those who make the perfect pass.
So, that means straight away the StrataData model for xA is likely to be different from those using objective data. An Expected Assist could then be calculated as the Assister getting the credit for creating a chance based on the expected conversion rate of the chance created. This can factor in several aspects such as location data, method of creation, player positioning etc. – StrataData uses a subjective rating to roll all this into one, giving an expected goal rating for each chance. We can then calculate the Expected Assist rating based on the number of Chances and their Ratings created by each player.
The Fàbregas example
But we weren’t happy with that. One clear example stuck in my head and it really made me want to look at how Creation was captured. The game was Chelsea vs. West Brom from 11/12/16. A tight game was decided by a single goal. I didn’t watch the match live but saw on Twitter Costa had scored and been assisted by Fàbregas – cue a barrage of tweets about how Fàbregas is so good at assists and has some kind of Premier League record (at least 10 assists in 6 different Premier League seasons) – then I saw the goal. Fàbregas plays an aimless ball into the corner, Costa chases it, bullies his man off the ball, beats another man when cutting inside and smashes the ball into the net from a much more central position, though he still has work to do – the goal is shown below.
So what’s wrong with that? Fàbregas provided the assist Costa scored the goal? 1 goal for Costa, 1 assist for Fàbregas.
Let’s look at the Chance Rating for this goal – judging this by the StrataData scale we have this as a Very Good Chance, which has an xG value of 0.245 – essentially from that position in those circumstances we’d expect Costa to score that chance roughly 1 in 4 times.
But what about the assist? It’s essentially just a long ball into the corner. How much credit does Fàbregas deserve? Using our new method of rating Primary Creation (the assist to the shot) and Secondary Creation (the pass before the Assist) we can gather expected values for these as well, in this case we’d rate Fàbregas’ Assist as ‘Poor Creation Quality’ and he would be awarded 0.033 xA.
This is useful in 2 ways. It assigns a value to those assists, rating them closer to the quality and expectation of being scored. It also allows us to look into the data and see which players are doing a better job of making chances for themselves.
This can then be used to judge how much a team will miss a striker – for example a situation where a striker has an xG score very close to the xA of those chances is likely to be a player who relies on others to set him up. While it can still depend on the type of chances created, these teams would miss the players creating the chances more than the attackers having the final shot.
A player with an xG score much higher than the xA score for those chances is capable of turning a ‘bad’ pass into a much higher quality chance for himself. These players are missed much more when they don’t play, with the team often reliant on individual brilliance.
Top 20 Premier League xA leading contributors 2017-18 (Old vs. New Method)
Data correct up to 22/09/17
While the early stage of the season means that there aren’t too many differences still there are some interesting points to look at straight away.
Eriksen drops by 0.652xA, which, for a player so lauded by so many, means that Tottenham, are likely still relying on more than just an end product by Kane and Alli to finish their chances. In fact Ben Davies surprisingly moves above him marginally as the chances he’s created stayed at the same level between the old method and the new method.
The top of the xA table is still populated by Manchester United, Manchester City and Tottenham players, making up 7 of the top 10 so at least as a sense check the new method seems to be working.
The Team difference also effects the top sides most, largely due to the number of chances they have created rather than a reduction in quality, but it does mean that these sides may be more reliant on top quality strikers turning Good Chances into Great ones. The top 5 seeing a reduction in xA from the old method to the new method are Manchester City, Liverpool, Newcastle, Arsenal & Chelsea.
I’ll follow this up later in the season to see if the theory still holds true but I’m hopeful it will be a promising new way to look at the quality of assist produced.
This is just one part of the variety of data we collect at StrataBet. If you are interested in purchasing StrataData please contact us on firstname.lastname@example.org to discuss your needs