Using Expected Shooting Percentage to determine Projected Goals
written by: @TheGersReport
A few days ago, I introduced Expected Shooting Percentage (xSh%) which projects what rate of shots on target should have beaten the keeper based on the kinds of shots a player takes. The value of this stat is that even though players' scoring rates will typically regress to the mean, what I like about xSh% is that it helps establish what each player's mean may be.
For example, both Jamie Maclaren & David Milinkovic scored on 50% of their shots on target, which is a rate that is unsustainable & (if they return to Scottish football) will be very safe bets for regression. But when applying Expected Shooting Percentages you can see that they have very different 'floors of regression.'
Based on the kinds of shots on target that Maclaren took last season, his xSh% was 42%, while Milinkovic's was 32%. Both players would be primed to score at a lower rate next season, but Milinkovic's scoring rate would likely bottom out at a more severe rate than Maclaren's.
While I was playing around with these numbers, I began to wonder...what would happen if I applied the same concept to Shot Accuracy (which is the rate of total shots that are on target)? If an Expected Shot Accuracy was determined, you could then combine Expected Shot Accuracy with xSh% to basically project how many goals a player should have scored last season.
Odsonne Edouard led the league in non-penalty goals per 90 last season with 0.76. Let's breakdown his actual shot statistics before applying the Expected rates. When looking at only his unblocked shots, his Shot Accuracy was 67% & his Shooting Percentage was 50% (remember that means 67% of his unblocked shots were on target & half of those shots on targets beat the keeper for a goal). His goal scoring rate will regress next season. In 2016-17, each of the players in the top 15 for goals per 90 minutes who had a Sh% of 40% or higher saw that rate decline this past season, which in turn led to a decrease in their goal rates.
Now, if you apply his Expected rates we can get a clearer picture of what his regression may look like next season. If he was able to get his shots on target with the same accuracy as the league average, based on the kinds of shots he took, his Expected Shot Accuracy on unblocked shots was 56%. Then, you can factor in his Expected Shooting Percentage (40%) & that is how can come up with Edouard's Projected Goals per 90 which was 0.51. That rate is still very, very good & is basically equal to Moussa Dembele's actual goal rate this season (maybe Matt Rhein was onto something)!!!
In my last post, I tabbed Kyle Lafferty as a player who was poised to see a decline in goals next season & suggested that if teams come in with transfer offers, that Craig Levein would be smart to part ways with the 30-year old striker. This past season his actual goals per 90 (non-penalties) was 0.36. His Projected Goal rate was noticeably lower, at 0.23. For context, Kenny Miller (who had a REALLY bad season) averaged........0.23 goals per 90.
How did that difference in actual & projected rates happen? Below you'll find a visual that splits out Lafferty's Expected & actual rates for Shot Accuracy & Shooting Percentage.
You can see that his inflated goal rate came from the fact that 46% of his shots on target were goals. Couple that with the fact that his xSh% was 24% & it wouldn't be a surprise to see his goal output take a serious hit next season. Additionally, you can see that he wasn't very good at getting his shots on target & if that continues he won't be able to balance out his regression with a high volume of chances to test the keeper.
I want to come back to using these kinds of data visuals as entry point for coaches, but first let's see how the numbers play out for others.
First...here are the goals per 90 leaders from last season (at least 800 minutes played), along with their Projected Goal Rates.
Leigh Griffiths scored nine goals this season, in a year derailed by nagging injuries, but you can see that he is still arguably the best forward in the Scottish Premiership. Only two players scored at a higher rate this season, but nobody really came close to matching his Projected Goals per 90 of 0.66. The next best were Dembele at 0.54, & Edouard at 0.51. The non-Celtic trio of Florian Kamberi, Kris Boyd, & Alfredo Morelos weren't really close with rates of 0.47 (Kamberi) & 0.46 (Boyd & Morelos).
Griffiths has been the league's best goal scorers for years now & there is no real sign that a decline is imminent.
Notice the difference between Lafferty & Griffiths. There was very little margin between Griffiths' actual & Expected shot rates, which makes you think he can easily sustain his goal scoring rates next season. On the other hand, there is an extreme difference in Lafferty's rates, particularly his Sh%, which supports the claim that his goal rates will decline.
Now, let's look at top 15 players based on their Projected Goals per 90 rates:
There's been a lot of talk about whether Alfredo Morelos can repeat his goal scoring rate next season, including my own confident proclamation of:
Rangers supporters (& management) can take some solace in the fact that Morelos' actual goal rate basically mirrored his Projected Goal rate this past season...which may be a surprise to many who bemoaned the chances he missed.
Again, I'm a solid maybe on whether he will keep scoring at such a good rate. The numbers suggest that he should be able to get more shots on target, BUT he is due for a noticeable regression in his Shooting Percentage. Can these two factors balance each other out? I'm not that sure. I wrote last week about how Morelos is too reliant on shots inside the box, but on the edges. His unusually high Shooting Percentage on these kinds of shots drove up his success rate & history suggests it's not sustainable.
The player who really sticks out on the visual above of the players with top 15 Projected Goals per 90 rates is Scott Sinclair. He ranked 7th in Projected Goals per 90 with a rate of 0.44, but in actual goals per 90 he ranked 23rd with 0.31 per 90. His goal scoring output was going to drop last season, it was inevitable given that his Shooting Percentage in 2016-17 was 64% when he averaged 0.73 goals per 90 minutes. But it shouldn't have been that big of a decline if you look at his Projected Goal rates.
Let me be very clear. I do not see Projected Goals as a replacement to Expected Goals (which is slowly becoming mainstream).
However, I see it's benefit as an entry point for coaches to hone in on areas to support the player. Look at Sinclair's shot rates.
There was literally no difference between his actual Shot Accuracy & his Expected rate. But, you can see that the discrepancy lies in the fact that he was not finishing his shots on target at the rate he should have been (when you apply league average rates to the shots he generated, aka xSh%).
It was at this point that I decided to walk down the halls of the Modern Fitba headquarters to the film room where our tactics team (Alex Lawrence) spends his time re-watching random matches from October yelling at the walls about "half-spaces", "mid-blocks" & "vertical compactness."
I asked him to put his "coaching hat" on & to watch some clips of Scott Sinclair. I smuggled over a video of all of Sinclair's shots on target that were saved...to see what he saw.
First, Alex astutely pointed out that there is some survivorship bias here. I nodded, listened & acknowledged.
Alex pointed out, "A couple of things stand out. Firstly, shot location. This is probably worth looking at and comparing to last season."
"The second thing ties in with the first a little. When you watch footage of Sinclair's goals last season you will notice how many of them are either from counter attacks or in frontal 1v1 situations where he can use his quick feet to move the ball outside the defender and get a clear 'view' of the goal. This is vintage Sinclair, yet to my mind I don't remember it being a hugely prevalent part of his game this season."
"Also see how often he manages to make runs in behind the defensive line to go 1v1 with the goalkeeper. This was certainly a noticeable feature of his game last season (he tended to make diagonal runs in behind from the left wing when playing there in the 4-2-3-1 or 3-4-2-1 rather than move to a left '10' position as McGregor does), that wasn't really so strong this year."
"It seemed as though this season he spent most times in advanced areas outside the width of the penalty area, let alone outside the width of the six yard box/goal (where you would typically want a goalscoring wide player to play)."
It was at that precise moment that Dougie Wright was making his rounds in the Modern Fitba offices, chatting up the interns & gossiping about the latest spats on Twitter & he stumbled upon my conversation with Alex.
He calmly asked - "Can you playback the footage? Let me see what I can see (as I put on my blue-tinted glasses)."
"Well......from a location perspective, as well as being wide, the angle is pretty narrow. He's shooting from an angle of 20 degrees; average conversion from there is 12% and Sinclair's on around 14%."
Dougie cleared his throat & added, "Technically, he's maybe guilty of sacrificing power for accuracy in some of these clips. Furthermore, there were occasions where he took it first time when he had time to take a touch and pick his spot."
"As Alex says though, probably a bit of survivorship bias coming into play there."
Then, I decided to blow their minds. Take a look at this...Sinclair had three shots which went dinged off the woodwork (& technically are classified as being "off target"). If these shots that hit the post had actually scored...Sinclair's goals per 90 would have been 0.42. Remember what his Projected goals per 90 was?
I'll save you the scroll...it was 0.44.
Ultimately, I believe Projected Goals & the subsequent breakdown to Expected Shot Accuracy & Expected Shooting Percentage is an example of how data can help clubs. To an extent, it can help in recruitment or squad management.
But, where I foresee it's greatest value is giving performance analysts & coaches entry points to help players find areas of improvement in how they approach shooting situations. I think it's hard to not watch that clip of Scott Sinclair's shots that were saved & not see a player "sacrificing power for accuracy."
I'm sure if we made similar videos for those players under (or over) performing their Shot Accuracy or Shooting Percentage stats that coaches could really hone in some tangible areas to focus on as they work on the player's development.
- Stats used in this post come from my season long tracking of Scottish Premiership data. Next season, Modern Fitba will work off a unified set of stats.
- Only unblocked shots were factored into shot accuracy stats.
- Most of you know Dougie's work (& it's really good). Also, make sure you check out the tactical wing of Modern Fitba, like this Scottish Cup preview that Alex wrote.
- Who were the outliers when comparing Projected Goals to actual goals per 90? Besides Sinclair - Kenny Miller, Conor Sammon, Miles Storey (hi!), Ryan Christie, Billy McKay, Tom Rogic, & Sofien Moussa were the ones who most underperformed their Projected Goal rates. There are arguments that Sammon & Moussa just kind of suck, while Miller is just too old. Storey, Christie, & Rogic may be the safest bets to see an increase in output next season, while Billy McKay is my early prediction for Scottish Championship Player of the Year. Meanwhile - Jamie Maclaren, Odsonne Edouard, Steven Naismith, & Michael O'Halloran had the biggest differences in the other direction. Of course, there is no guarantee that Maclaren or Edouard will be playing in the Scottish Premiership next season, while Naismith is still technically a Norwich City player for another season. Who knows where Michael O'Halloran is? Of the other players we actually think will be in the Premiership next season, expect declines in goal rates from Kyle Lafferty, Ryan Bowman, & Kris Boyd.
- This was written under the influence of ShitKid, the Brian Jonestown Massacre, Parcels, & Extra Classic.