After looking at how the defensive side influences the expected goal value of a shot in part one, part two will look at how the attacking side impacts a shots value by looking at what action preceded it.
The plot that I used in part one to show how the probability of a goal decreases as the number of defensive players behind the ball increased doesn’t work so well with the variables for what preceded the shot however, these variables are binary so either it was true that a pass preceded the shot or it wasn’t, the gradual decline in probability doesn’t really work the same.
With this in mind this piece will just highlight how some players and teams create rather than how best to create. With that said I did quickly get the xG per shot for each method just to show how some methods of creation, on average, are more valuable than others. The types of creation and average xG for each can be seen in the table below:
Rebounded shots are the most effective way of scoring, which shouldn’t come as a huge surprise given that rebounded shots will usually have little to no defenders between the ball and goal while there’s also a good chance the ‘keeper hasn’t recovered from the first shot.
It’s a similar situation to shot deflection being in 3rd, it’s likely the defence and ‘keeper aren’t set for a shot so when the deflected shot falls to an attacking player there’s a greater chance that they’re able to score.
The problem with these two methods is that you can’t really play for them. While you can send your team out to be active off the ball, trying to claim every loose or 2nd ball for themselves, you can’t intentionally play for rebounds – unless it’s a specific situation such as crowding the ‘keeper on set pieces for instance.
It may be interesting to see a team try to create rebound opportunities in open play – possibly using an accurate long-range shooter – but it’s not realistic, it feels like it’d be a real life ‘Football Manager’ experiment.
Rebounds and throw ins won’t have their own section as there wasn’t much to mention given how infrequently they occur (Leicester’s Christian Fuchs created the most chances from throws with 9 while Bayern’s Robert Lewandowski had the most rebounds with 8) but the first section will be:
Dangerous Moments happened even less frequently than throw-ins and rebounds but given they’re something different I thought it’d be worth given a brief overview of them.
Dangerous Moments are another measure from StrataBet that is more subjective than the event data from the likes of Opta, but like the defensive pressure ranking it’s very understandable. StrataBet define a Dangerous Moment as an occasion when a shot is not always taken but the opportunity is there to shoot.
A good example of this is Hector Bellerin’s goal against Everton on the final day of 2016/17, Ozil plays the ball across the face of goal but Welbeck isn’t able to convert meaning Welbeck is credited with being the player who assists the shot under the qualifier of Dangerous Moment. Take a look at the goal below:
While I like the having this quantified, given how infrequently a Dangerous Moment occurs before a shot (126 times across the Top 5 Leagues) there isn’t much to draw from this.
It could be a good way to find players who are moving the ball into dangerous goal scoring areas but not getting properly credited due to teammates not being able to get a shot off, however the player with the most Dangerous Moments preceding a shot (Ivaylo Chochev from relegated Palermo) only managed 4 across the whole season.
Hopefully Dangerous Moments will have a bigger part to play when looking at secondary assists in part 3.
Crossing – Play your cards right
One of the most interesting observations from the xG per shot table is that the xG value following a low cross – on average – is almost twice as big as that of a high cross. While this may seem obvious given headed goals are also harder to score than regular shots it feels like something that isn’t being fully utilised.
Low crosses may not always be available, if the opposition drop back it’s much easier to send an early high cross than it is to work your way to the byline and cut it back, but given the increased probability of a low cross it may be worth teams being more patient and seeking out the opportunity to send a low ball in.
With this in mind I thought it’d be interesting to see which players and teams favour low crosses and the quality of chances they’re creating from these crosses. It’s also important to remember that these numbers are for chances created from crosses not crosses played meaning it’s likely that if teams play less crosses per game it’s because they’re creating less chances per game.
Only 8 teams from Europe’s top 5 leagues created with more low crosses than high crosses last season these were – Barcelona, Atletico Madrid, Sevilla, RB Leipzig, Wolfsburg, Werder Bremen, Bastia and Darmstadt, with RB Leipzig playing the highest proportion of their crosses low with 60%.
The question is though, does this have any link to performance?
And the answer… well not really. The correlation between the percentage of crosses a team plays being low and xGD is a measly 0.216.
What’s interesting though is the correlation between the number of chances created via low crosses per game and xGD is 0.681 which is a more respectable figure.
Obviously looking at the number of low crosses is going to produce a better correlation than percentage as there’s going to be a good correlation between chances created and xGD in general (0.823) but what makes it interesting is the correlation of chances created via high crosses and xGD is 0.470, so there’s a better correlation between xGD and the number of low crosses a team plays than there is xGD and the number of high crosses a team plays.
This is repeated by looking at the correlation regarding xGD and percentage of chances created that were high or low crosses, but with much worse numbers. Looking at the percentage of chances created that were high crosses the correlation is -0.008, it has basically no effect on a teams xGD. However, for low crosses this correlation is 0.270.
This still isn’t a good correlation but it’s a lot more than that of high crosses, given this minor correlation difference and the difference in the resulting xG per shot should teams be putting more attention in trying to create low crossing opportunities?
The average number of chances created via high crosses per game last season was 1.095, but only 0.577 for low crosses. On average teams are creating twice as many chances via high crosses despite chances from low crosses having twice the value, should this be something teams should be trying to change in the future?
Moving on from teams to players, who were some of the best crossers last season?
The top 5 for xA from crosses p90 (>= 900 minutes) were Lucas Vasquez (Real Madrid), Ousmane Dembele (Dortmund), Ivan Perisic (Inter), Yannick Bolasie (Everton) and Alejandro Gomez (Atalanta).
It’s worth noting we’re dealing with a small sample size here, the most chances a player made using crosses last season was 32 so when looking at things such as xA per cross a player may have only created 10 chances with crosses all season.
What would be a good idea is to combine a players cross completion rate (or even some kind of expected cross value) with their xA per cross in order to see how likely it is a cross from a player will result in a goal.
Unfortunately my data for all crosses isn’t from StrataBet and doesn’t separate high and low crosses, this plus different naming/ID’s makes it hard to cross between the two.
That being said for players who created 15 or more chances via crosses last year the one with the highest xA per cross was Barcelona new boy Ousmane Dembele with an xA per cross of 0.303. Dembele also has a good mixture of high and low crosses, creating 12 chances through high crosses and 9 through low.
Given that 19.4% of Dembele’s crosses led to a shot last season you could use this to say the probability of a Dembele cross resulting in a goal is 0.194 (chance of cross leading to shot) * 0.303 (chance a shot from his cross being a goal) which means on average a Dembele cross has a 0.059 chance of resulting in a goal.
I can’t do this for every player (I’m going to keep looking into it) but it could be an interesting way to see who has the most valuable crosses – rather than just the most valuable crosses the lead to shots.
The players to round off the top 5 for xA per cross were Ivan Perisic, Ladislav Krejcí (Bologna), Marko Arnautovic (Stoke) and Hector Bellerin (Arsenal).
If we then go on to look at high crosses only for those with >=10 high crosses the highest xA per cross once again belongs to Dembele with his 0.282 per cross being roughly 33% more than next best Alejandro Gomez. While for those with >= 8 low crosses the highest xA per cross falls to Fin Bartels from Werder Bremen with 0.397.
Overall low crosses seem to be a more effective way of scoring but teams don’t seem to be putting their efforts into creating low cross opportunities, while in terms of players Ousmane Dembele seems to be one of the best crossers in the world and should be a great fit for Barca who create the most low cross opportunities per game in Europe.
Set Pieces and Gylfi Sigurdsson
Set pieces are a huge part of football, making effective use of them can give a team a huge competitive edge over other teams in the division. This piece on StatsBomb is a much better look at the value of set pieces than I could give so rather than looking at the effectiveness of set pieces as a whole I thought I’d focus on one man – Gylfi Sigurdsson.
Sigurdsson had one the longest transfer ‘sagas’ of the summer with his move to Everton being a matter of when rather than if from the day the window opened. His eventual move to Everton ending up being the 7th most expensive in the Premier League this summer with Everton splashing almost £45m on him (via Transfermarkt), but did he warrant this fee and 5 year contract having recently turned 28?
On the surface his stats look impressive, his xA of 10.807 could only be bettered by Kevin De Bruyne, Christian Eriksen, Eden Hazard and David Silva in the Premier League last season. This sounds great, only elite creative players created better chances than Sigurdsson, however, only 30% of this came from open play.
Of those four above Sigurdsson the next lowest was Eriksen with 67.1% from open play, Sigurdsson even has the lowest percentage of all players with an xA greater than 5 in the Premier League. The next lowest is still almost double that of the Icelandic international as Dmitri Payet created 54.2% of his chances from open play.
After establishing the importance of set pieces it’s worth pointing out wanting an elite set piece taker isn’t a bad thing, but the fee, age of the player, contract length and just how reliant he was on set pieces for creativity make this seem like a bad transfer for Everton to make. His open play statistics are below average and 3 years of his 5 year contract will most likely be after his peak years.
The graph below shows xA from Set Pieces p90 and xA from Open Play p90 (filtered for >= 900 mins, xA >= 5 and xA from Set Pieces p90 >= 0.05 as the bottom was cluttered with players who don’t take set pieces), you can see Sigurdsson excels in set pieces but is among the worst for open play.
This doesn’t tell the full story however, someone who takes more set pieces is likely to create more from set pieces so something to look at is the xA per set piece they take. For this I took players who created over 10 chances from set pieces last season and plotted the number of chances created from set pieces against the xA per set piece taken. Warning, this looks like a mess:
Again there’s positives for Sigurdsson, he’s only behind Ryad Boudebouz for number of chances created from set pieces but still manages to be among the top few for xA per set piece.
The Premier League average for xA per chance created from a set piece is 0.129, Sigurdsson’s is 0.164, so the difference between an average set piece taker and an elite one is just 0.035 per set piece. So while Sigurdsson can expect a goal from every 6.10 chances created from set pieces that lead to a shot the average would be a goal every 7.75 chances from set pieces that lead to a shot.
Given that Sigurdsson took 226 set pieces last year and created 43 chances from them the chance a Sigurdsson set piece results in a chance is 0.190, couple this with the fact a Sigurdsson set piece has a 0.164 chance of resulting in a goal means the chance a Sigurdsson set piece turns into a goal is 0.164 * 0.190 which is 0.031.
Last season Ross Barkley took the most set pieces for Everton last season with 103, leading to 23 chances, so there’s a 0.223 chance a Barkley set piece results in a chance. Barkley’s xA per set piece is slightly above league average at 0.136 meaning the chance a Barkley set piece turns into a goal is 0.223 * 0.136 which is 0.030.
This means it should take Sigurdsson 32.258 set pieces per goal – it actually took 28.25 last season – and Barkley 33.33 set pieces per goal which was hugely under performed last season taking 51.5 per goal.
If they both performed to their probability last season the difference in assists per set piece would be minimal, is it worth the big money upgrade for a player so reliant on set pieces or would the money be better spent elsewhere? Particularly when the end locations of their chances created in open play look like this:
Barkley’s map isn’t perfect but there’s both a higher quantity of chances and a lot more of those chances coming centrally than there is for Sigurdsson.
The question ultimately comes down to: Who should Everton have bought instead of Sigurdsson?
The alternatives that stand out from the stats are:
Vincenzo Grifo – 24 – Gladbach – I talked about Grifo as an alternative to Sigurdsson in my 5 transfers you may have missed piece and having looked at the StrataBet data he still seems as though he could have been a good option – though it’s too late now as he was bought by Gladbach for ~£5.4m.
Comparing their stats with StrataBet data and there isn’t much difference. Grifo had an xG p90 of 0.211, slightly more than Sigurdsson’s 0.156 while his xA p90 was also slightly more, 0.334 to 0.292.
His actual open play goal + assist contribution from was also better with 0.469 p90 compared with Sigurdsson’s 0.449.
Moving on to set pieces Grifo had an xA per set piece of 0.141, slightly less than Sigurdsson’s 0.164 but still more than the Premier League average of 0.129. 47% of Grifo’s xA came from open play where his 0.127 p90 is slightly better than Sigurdsson’s 0.082.
Given Grifo would have cost a ninth of Sigurdsson and is 4 years younger it seems like it could have been a much better use of money from Everton and probably my personal favourite alternative.
Yunus Malli – 25 – Wolfsburg – Similarly to Grifo, Malli hasn’t been at his current club long having joined Wolfsburg in January 2017. With that being said, after Wolfsburg had a disappointing season it could have been worth Everton testing the water with a speculative bid.
Malli’s xA per set piece was higher than Sigurdsson’s at both Mainz and Wolfsburg last season, having just over 0.17 for both sides, however his open play contribution dropped right off for Wolfsburg. At Mainz 40% of his xA was from open play, this dropped to 18.6% at Wolfsburg, just over half of Sigurdsson’s 30%.
Over the course of the season for both sides though his xG + xA was marginally more than Sigurdsson having 0.479 p90.
Being 3 years younger than Sigurdsson and possibly costing around half the price this could have been an interesting option to pursue. Though Wolfsburg may not have wanted to sell a player they only signed in January, European football and a move to the Premier League (presumably including a nice rise) could have been tempting for Malli.
Pablo Sarabia – 25 – Sevilla – Seemingly moving backwards through transfer windows Sarabia joined current club Sevilla for £900k last summer and had a productive first season amassing an xG + xA of 0.495 p90.
Sarabia looks a good alternative to Sigurdsson given that his xA per set piece was only just behind the Everton player at 0.155, he created a lot less from set pieces (0.633 to 1.163) but it’s still an impressive number.
His open play numbers were also a lot more impressive than Sigurdsson’s with 67.2% of his xA coming from open play, leading to 0.2 xA from open plan p90, almost triple Sigurdsson’s 0.082 p90.
His crosses were also almost twice as threatening than Sigurdsson’s, albeit from a very small sample size as Sigurdsson had an xA of 0.170 per his 6 crosses while Sarabia had 0.313 for his 7.
Sevilla may not want to sell Sarabia – and he may not want to leave given their Champions League football – but with him joining for such a minimal fee, even after director of football Monchi has moved on it may be hard for them to turn down a potentially big profit
Notable mention – Rodrigo de Paul, Udinese. Only 23 his xA per set piece is marginally above league average at 0.130, but his xG + xA of 0.354 is a bit behind Sigurdsson.
To conclude, Sigurdsson is an elite set piece taker and set pieces are a huge part of the game, but is he worth £45m and a 5 year contract at 28 given almost all of his creativity stemmed from set pieces last season and the probability of his set pieces resulting in goal is practically equal with Barkley?
EDIT: xA per Set Piece may be slightly misleading, it’s xA per set piece that leads to a chance rather than the xA per every set piece the player takes.
Passes are probably the event that has the most variation for those that precede a shot, an open play pass can mean a players knocked the ball 5 yards inward for his team mate to hit a 30-yarder or a player has played a defence splitting pass laying the ball on a plate for his teammate.
This is where more advanced metrics such as expected goals and assists start to become useful as these passes are no longer treated equally but rather those players creating more dangerous chances are fairly rewarded.
With this in mind there’s two things I’ll be focusing on for open play passes – Borussia Dortmund and Cesc Fabregas.
Borussia Dortmund’s passing numbers don’t immediately jump off the page – they only created the 22nd most chances via passes per game in the top five leagues last season – but when you start digging into their passing stats some impressive and interesting numbers start popping up.
For teams that created more than 5 chances per game from open play passes, on average, Dortmund had the most dangerous passes with the xG per shot following their passes being the highest in the top five leagues. Dortmund’s 0.193 xG per shot after an open play pass is around 50% more than the top 5 league’s average and this number may continue to rise after their new signings.
Looking at the average xG value for shots their passes created, (>= 15 chances created from passes) 3 of the top 5 now play for Dortmund – new signings Mahmoud Dahoud, Maximilian Philipp and youngster Christian Pulisic – with each of their key passes resulting in an average xG value of greater than 0.2.
Adding to this Dortmund even had 2 of their players in the top 5 for those with the highest xA p90 from open play passes with Shinji Kagawa and the now departed Ousmane Dembele.
Even after losing Dembele Dortmund’s have great attacking options this season, should the new signings be able to repeat their dangerous passing in a Dortmund shirt then chances shouldn’t be hard to come by for Pierre-Emerick Aubameyang this season.
Dortmund have a lot of dangerous passers in their squad but the most prolific creative passer last season was Cesc Fabregas. Fabregas had an xA p90 of 0.337 from passes, edging the top spot from Lionel Messi by 0.005.
On average his passes weren’t as dangerous as others, marginally ahead of the average with an xG per shot of 0.143 but the volume he creates at is incredible – his xA p90 last season was the 2nd highest in Europe with 0.564 but his actual A p90 of 0.949 was significantly ahead of other players.
With that being said it’ll be interesting to see how the Spaniard fits into Antonio Conte’s plans this season. Fabregas only played 38.8% of available minutes in the Premier League last year and Chelsea have invested in both Tiemoue Bakayoko and Danny Drinkwater this season.
While Fabregas’ passing may be unrivaled in the Premier League he could be classed as too attacking for the deeper midfield roles while he may not stretch the play enough to be part of the front 3. It seems as though for another year Fabregas will play a big part in Chelsea’s season albeit not from a somewhat small amount of minutes.
Pressing has become a huge part of the game over recent seasons, more teams are demanding their players to be active off the ball and look to quickly regain possession from the opposition high up the pitch.
It’s worth noting however, turnovers that directly precede the shot means a player has won the ball back and gone on to shoot with no event in between so the results here won’t necessarily be an index of the counter pressing teams. It could be that teams who play it out from the back have made an error or a failed clearance falls to someone who shoots, rather than an organised press leading to a turnover and then shot.
If anything looking at 2nd assists next week may be a better look at counter pressing teams – though I haven’t looked the data over properly yet.
With that being said the top 18 teams for chances created via turnovers were the 18 Bundesliga teams, with Augsburg leading the pack creating 16 chances via turnovers. After the Bundesliga 6 of the next 10 clubs then come from La Liga – Sporting Gijon, Malaga, Celta Vigo, Real Madrid, Atletico Madrid and Alaves. The only non-German and non-Spanish teams to make it into the top 30 were Lazio, West Ham, Montpellier, Lille, Southampton and Bournemouth, though by this point we’re down to 3 chances created via turnovers.
While Germany may dominate in the number of chances created via turnovers, for teams with more than 3 chances created via turnovers Spain have 4 of the top 5 for xG per shot following a turnover with Atletico Madrid leading the way having an xG of 0.368 per shot following a turnover.
There’s only small margins for turnovers – Augsburg’s 16 turnovers only led to an xG of 2.455 – but it is still interesting to look at. The player who benefited the most from turnovers was Julian Baumgartlinger who’s 3 turnovers led to an xG of 0.853 for the Austrian international, accounting for just over a third of his total xG. Meanwhile the player who made the most of their turnovers (>= 2 turnovers) was Halil Altintop who’s shots following turnovers had an average xG of 0.345.
On the other end it was Freiburg who had the worst record for turnovers against with opposition teams able to create 22 chances following turnovers in 2016-17, though it was Sporting Gijon who’s turnovers proved most valuable for the opposition as their 4 turnovers resulted in an xG of 0.378 per shot.
Given the small sample size of turnovers resulting in shots there aren’t many conclusions to draw from the data – apart from they like to press in the Bundesliga of course.
Looking into how chances are created can bring up some interesting observations, particularly when breaking it down into different categories. One idea that I might do once we’re a few more games into the season is see if/how this data can be used to improve a teams performance by looking at how they create or how opponents create against them, looking for patterns and any kind of inefficiencies in their game.
Before then though will be part 3 of looking at StrataBet data which should be much shorter than this piece. It’ll look at second assists, highlighting players who may not get just praise when only looking at goals and assists.
This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations.