Exploring Expected Goals With StrataBet Data: Part Three – Creativity II

Following on from part two which looked at the action that preceded a shot, part three will take a step back and look at the secondary assist – the action before the action that preceded the shot.

The secondary assist isn’t something that is regularly recorded in football but provides an interesting look at players who may be involved in the teams build-up, but not be the player to either take or set up the chance.

There are other measures which perform a similar job – StatsBomb’s xG Chain is much more in depth and looks at possession chains rather than just the player before the creator – however looking at the player before can still yield some interesting results.

A couple notes before starting:

I had planned on making some kind of expected second assist model but the results didn’t turn out great and given the number of variables and fairly low sample size for certain situations it might be best that I didn’t.

Instead of doing this I decided to look at the xG value of the resulting shot – similar to the xG Chain – as a measure of how valuable a players secondary assists are.

It’s not a perfect method given a lot can happen between a primary and secondary assist – particularly if you include dribbling which isn’t recorded in the data – but it helps show which teams are contributing to their sides most valuable chances. Later on I refer to a players xG contribution which is just the xG of the players shots, plus the xG of the shots resulting from their chances created as both the primary and secondary player.

I struggled with what to call a secondary assist that doesn’t result in a goal (as assist implies goal) so settled upon secondary chance created. It’s not a good term as it implies the chance is secondary, rather than the chance having a secondary type or player, but it seemed the best of a bad bunch.

Also, Germany has more chances with a secondary type registered than other leagues and Spain has by far the least, which is a big reason why the player rankings have huge amounts of Bundesliga players and hardly any La Liga. Shots with secondary types registered per league are: Bundesliga – 2046, Premier League – 1849, Ligue 1 – 1438, Serie A – 1336 and La Liga – 789. This is really interesting on its own and quite surprising the country known for possession football has the lowest number of shots following a secondary type.

Rather than produce a table of all the combinations of secondary and primary actions I just produced a similar table as the one in part two only accounting for the xG per shot depending secondary assist. The values tend to be much higher than when looking at the primary assist but given the smaller sample size (around 20% of shots have a credited 2nd assist, compared to around 80% credited as having primary) plus the fact the probability a team scores increases as they make more passes this may not be a worrying thing.

Here’s the table for xG per shot by 2nd assist type:

Set Pieces, Second Balls and Liverpool

One of the biggest differences between the above table and the one seen in part two is the rise of corners and high crosses as a means of scoring. When the primary assist is a corner or high cross the odds of scoring are fairly low – despite them being fairly common ways to create chances – however when it’s the secondary assist that’s a high cross or corner the chances of scoring dramatically increase.

This shouldn’t come as too big a surprise. The ball is in a dangerous area and the resulting chance is likely to be following a rebound or knockdown rather than a headed chance directly from the cross – given, on average, headed chances are harder to score and rebounds are easier you can see why it’s an effective way of scoring. If a corner isn’t cleared properly or leads to a rebound there’s a huge number of players around to poke it home.

From an attacking point of view this becomes interesting – the average xG of a chance directly following a high cross is 0.143 and that of a corner is 0.098, when a chance is taken with the secondary action being a high cross or corner however it has a huge effect on xG values. High crosses almost double to 0.273 while corners more than triple to 0.312.

Could teams make more effective use of corners by aiming for a knockdown, flick-on or short corner routine rather than a direct header?

If you have a target man is it better to play for them knocking it down to players in the box rather than just looking to score headed goals?

On the defensive side the most interesting example of set pieces as secondary assists comes from Liverpool.

Lots has been made of Liverpool’s defending on set pieces since the arrival of Jurgen Klopp, with statistics pointing to how they’ve conceded x amount of goals regularly cited on TV whenever the club concede a set piece. However, similar to the bit about Gylfi Sigurdsson in part two, this doesn’t tell the full story.

Starting with chances created where the primary assist was a set piece, Liverpool conceded around 0.093 xG per game just from set pieces, only the 11th most in the Premier League.

However the xG per shot for the chances they gave up via set pieces is the 3rd highest in the league – interestingly tied with the two Manchester clubs – while only Swansea and Hull gave away higher value chances via set pieces last season. 10.9% of Liverpool’s xG against came from chances where the primary assist was a set piece, only Manchester United, Everton and  Crystal Palace had a higher percentage.

Just as big a problem for Liverpool though are the chances they concede where the secondary assist is a set piece. Klopp’s side give away 0.211 chances per game where the secondary assist is a set piece, the 6th most in the Premier League, with the average xG value of these shots being 0.358 – the 5th highest in the league.

Where it gets interesting though is Liverpool conceded 8.8% of their total xG from chances that originated from a set piece as the 2nd assist, the second highest in Europe behind only Genoa. The graph below shows the % of xG against originating from set pieces:

Click to Enlarge

You can see that while Liverpool can improve in both areas the percentage following a secondary set piece is more troublesome than the primary as their 10.9% isn’t much higher than the European average of 9.5%.

It still means however almost 20% of Liverpool xG against last season originated from set pieces, the average across Europe was 12.8%, while the average for chances where the secondary assist was a set piece was 3.3%. Liverpool are almost double the average overall and almost triple the average for secondary assists being set pieces.

If Liverpool gave away just the average percentage of xG from set pieces last year they could have had just under 10% less expected goals against last season. Liverpool’s style of play usually means they don’t concede many shots but the shots they do concede tend to be high quality (this was touched upon in part one), this is a by-product of how they play.

If a team is able to break the press they find themselves up against a somewhat unprotected defence. Given how this doesn’t happen all that frequently it’s not a huge concern, but set pieces are something Liverpool should be dealing with a lot better.

Liverpool had the 3rd highest percentage of expected goals against come from set pieces (primary and secondary) in Europe’s top five leagues last season yet nothing seems to have changed over the summer. While it’s a tiny sample size from the first 4 games of 2017-18 22.2% of Liverpool’s xG against has been via set pieces.

If Liverpool wish to progress and turn into title challengers this is an area where they should be seeking improvement. I’m no expert on set pieces but I’d imagine Liverpool could get down to at least the average percentage given some time and organisation over the summer.

It seems strange they don’t look to have changed anything given it’s where almost 20% of their xG against has come from.

With how competitive the top of the table can be, if you were offered a chance to reduce your xG against by around 10% without signing anyone or compromising your style of play surely you should be taking it?

Young Players, Bayern Munich and Trying to Find Midfielders

After looking over the second assist data two things seemed to stand out – second assists could be a good way to find talented young players and Bayern Munich seem to love second assists.

3 of the top 5 players for secondary chances created per 90 are from Bayern Munich (Kingsley Coman, Arjen Robben and Douglas Costa) while they also have 2 (Phillip Lahm and Joshua Kimmich) of the next 5, meaning they have half the top 10.

There also seems to be a good representation of young players in the mix too, Kingsley Coman, Joshua Kimmich, Nadiem Amiri, Alex Iwobi and Julian Brandt all find themselves in the top 20 with Schalke’s Max Meyer in 21st too.

It means 5 of the top 20 are 22 or younger, while it may not be crazy it seems more than you’d find for a lot of other metrics.

Could secondary assists be a good way to highlight talented young players? It highlights players who are getting involved in build-up and occupying dangerous areas but not necessarily taking the shots or playing the final passes just yet. This could be the case for a lot of young players who are just breaking into teams and are not yet the primary goalscorers or creators.

Just plotting some graphs of different metrics starts to highlight some interesting players. For instance there’s some interesting players in this graph showing secondary chances created and the average xG of the resulting shot (>= 900 mins & >=10 secondary chances created):


Click to Enlarge

Some interesting names from the above:

Philipp Max – 23 – Augsburg: The young Augsburg left-back has around the same number of secondary chances created as Arsenal new boy Sead Kolasinac (who you can just about make out below the ‘Bayern Munchen’ part of Robert Lewandowski) but on average the chances tend to be slightly higher value.

It’s worth pointing out Max did spend time at both left-back and left-wing last season, as he also manages a higher xA p90 than Kolasinac with his 0.263 almost doubling the Bosnian’s 0.128.

This isn’t to say Arsenal should have been in for Max instead, I’m only making the comparison as they happened to occupy a similar space on the above graph, It’s just pointing out Max could be a decent pick up for a club. 23-years-old and capable of playing left-back and left-wing while having a tendency to be involved with good quality chances are all useful attributes.

Taking the xG value of the shots he took, assisted or second assisted leads to an xG contribution of 0.396 p90, not bad for a young player who’s Augsburg side only managed 35 goals last season.

Robert Bauer – 22 – Werder Bremen: Bauer took up a lot of different positions for Werder Bremen last season, playing at least one game across every defensive and midfield position, however his primary two positions (in terms of appearances) were right-back and left-back.

His numbers aren’t as impressive as Max’s – xG contribution of 0.260 p90 – and tend to fall under the category of someone who doesn’t create much but when they create it’s high quality, however a versatile 22-year-old who’s actions can lead to high quality chances could be a useful acquisition.

Another interesting graph we could plot in a hope of finding some deeper lying midfield players is the same as above but accounting for passes only (>= 900 mins & >= 10 secondary chances created via passing).

Click to Enlarge

Fabregas leads the way by quite some distance here, so to make it slightly easier to look at we’ll exclude him, giving us this:

Click to Enlarge

Some interesting names to pop up here (looking mainly for midfielders) are:

Leandro Paredes – 23 – Zenit: I previously talked about Paredes in my 5 transfers you may have missed piece as he slipped under the radar going from Roma to Zenit. His passing and defensive stats are great for a deep lying playmaker and the above gives further proof of this. While he may not create a huge quantity his secondary chances created do have a higher average value than any other player.

His xG contribution of 0.534 is hugely impressive for a deep lying player (although he did take some set pieces too) and it’s surprising a bigger club or a club in a bigger league didn’t look to pay Roma the £20m for him.

Danilo Cataldi – 23 – Lazio: Cataldi spent last season on loan at Genoa and occupies a similar space to Xabi Alonso on the graph. It’s worth pointing out though that Cataldi only played 930 minutes at Genoa and his xG contribution p90 (0.181) isn’t outstanding. If he gets more game time at Benevento though he should be a player worth watching.

Nadiem Amiri – 20 – Hoffenheim: If we define the top right corner as 1 secondary chance created p90 and an average of 0.5 xG for the resulting shot then Nadiem Amiri is the 13th closest to the corner out of the above batch of players, which is impressive given his young age. His xG contribution of 0.576 p90 is hugely impressive and Hoffenheim will be glad they renewed his contract in June as if he continues this into the 2017-18 season it wouldn’t be surprising to see bigger clubs in for him next summer.

Joshua Kimmich – 22 – Bayern Munuch: While originally being a midfielder Joshua Kimmich seems to have become a permanent right-back now, regularly appearing there for both club and country. Despite the position switch his passing influence seems to be just as good as most midfielders as he’s the 7th closest to the top right corner and the closest for those 23 or under.

His goal contribution of 0.511 is impressive for both a player so young and a defensive player. The right-back role may be a hard one to fill at Bayern after Phillip Lahm’s retirement but Kimmich seemed to have fit straight in.

Other interesting players are Alex Iwobi (9th closest to the top right corner) who has great numbers for secondary chances created and passes into the box (these aren’t from StrataBet but were discussed when talking about him as a potential replacement for Mesut Ozil here) which shows through his xG contribution of 0.547 p90.

Riechedly Bazoer only played 974 minutes for a struggling Wolfsburg last season but at only 20-years-old was the 25th closest to the top right corner below the trio of Kevin Kampl, Kevin De Bruyne and Xabo Alonso for distance.

Another young Bundesliga player to impress was 21-year-old Max Meyer from Schalke who was the 18th closest to the top right corner and had an impressive xG contribution of 0.482. Given Schalke have had lots of talented midfielders over the years, including the likes of Mesut Ozil, Julian Draxler and Leroy Sane, Meyer is a player who could be worth keeping a close eye on.

This kind of stuff is just scratching the surface of looking at different ways to filter and use the secondary assist data, hopefully throughout the current season I can continue digging through the data and find some more interesting points.

Dangerous Moments V2.0

It turns out that my interpretation of Dangerous Moments in part two wasn’t quite right. Rather than a qualifier that precedes a shot they’re usually treated as their own event, so the example I gave with Danny Welbeck and Hector Bellerin on the final day of the season isn’t necessarily a common occurrence.

A lot of the time a player may not be there to sweep up the ball and score like Bellerin was but instead the Dangerous Moment would be the ‘chance’. I thought it’d be worth pointing out the wrong interpretation in the previous part and looking at the players who contribute the most to dangerous moments – whether they’d be the player, primary player or secondary player involved.

So this’ll just be a quick run down and I’ll try to look deeper into dangerous moments as the season progresses.

The top 10 ten players who have been most involved in dangerous moments p90 (>=900 mins) are: John Guidetti (Celta Vigo), Cristiano Ronaldo (Real Madrid), Kingsley Coman (Bayern Munich), Dries Mertens (Napoli), Ezequiel Ponce (Granada), Lucas Vasquez (Real Madrid), Ante Rebic (Frankfurt), Lazar Markovic (Hull), Cheick Diabate (Metz) and Lionel Messi (Barcelona).

Out of these players Ezequiel Ponce is probably the most interesting case. Playing for a doomed Granada side last season he not only had 0.480 Dangerous Moments per game he also had a pretty impressive xG p90 of 0.397. It’ll be interesting to follow how he does now he’s out on loan at Marcelo Bielsa’s Lille side and whether he can turn some of these Dangerous Moments into shots and preferably goals.

One idea using Dangerous Moments may be to work out hypothetical xG values for Dangerous Moments, so what the xG would be if the player was able to get a shot off, in an attempt to see which teams or players are suffering most due to not being able to turn these moments into shots.


Taking a step back and looking at the secondary assist can bring up a lot of interesting things, most of which I’ve hardly touched on here. To go into proper detail I feel as though it’ll be better to focus on a specific team or player, rather than just giving run downs in the way that I am.

However for this piece I was mostly focused on just pointing out a couple of interesting things and making sure it doesn’t run as long as part two (which I was just about successful at). Expect to see more content looking how teams and/or players are creating chances some point in the future though.

This may be the final part of ‘Exploring Expected Goals with StrataBet Data’ too – I’m thinking about doing something involving their box entry data but haven’t got anything yet and it also doesn’t really fit the ‘expected goals’ part of the title.

I mostly did these three pieces to delve into and understand the StrataBet data while just pointing out anything interesting I found along the way. I feel now I have a better grip on the data and I can start using it to look into specific things rather than just giving brief rundowns of the different qualifiers.

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.


Leave a Reply

Your email address will not be published. Required fields are marked *