Members

Changes to Opta’s expected goals (xG) stats explained

As some Premium Members of Fantasy Football Scout have picked up on already, Opta’s expected goals (xG) model has had a revamp.

A decade on from when it first appeared, the divisive metric, championed by our Career Hall of Fame No 1 Fábio Borges, has become firmly established in the Fantasy Football conversation. 

It’s the only stat I use (along with xA) when making FPL decisions. Of course, it is not perfect but if the model is reliable – and you know how to interpret it – I believe it is, by far, the most useful stat for FPL.

interview with Fábio Borges

In essence, all stats in football aim to tell the story of the match.

While the only one that really matters is goals scored, knowing the number of shots a team had can tell you a little more about what happened in the game – was it a dominant performance or a smash and grab?

Expected goals adds colour to that picture by looking at the type of shots taken.

If one team was restricted to lots of efforts from distance but gave up a glorious chance on the counter, then expected goals will help paint a more nuanced depiction.

Context is everything

With the release of xG 2.0, expected goals has become better at evaluating the quality of a single shot.

It now considers over 35 contextual factors. The most significant improvements are the inclusion of goalkeeper position, defensive pressure and shot clarity. 

And, in a move that will please its detractors, Opta has removed ‘big chances’, previously used as a proxy to estimate those details.

The definition of a big chance begins “a situation where a player should reasonably be expected to score” and left it open to criticisms of subjectivity:

… it’s hard for me to believe that there would be zero outcome bias, that a shot being scored wouldn’t make the preceding chance look bigger.

– Michael Caley

Those concerns are no longer an issue.

A more accurate picture

On Fantasy Football Scout, the expected goal values for this season have all been updated to the latest model. Let’s take a look at a few examples to see how they compare:

Firmino v Watford
(see video link to Liverpool YouTube channel)

The chance above has the highest probability of resulting in a goal of any shot this season. Firmino is central, inside the six-yard box and the goalkeeper is stranded.

According to the updated model, the Brazilian would score 96 per cent of the time. A more intuitively satisfying result than the 81 per cent (0.81 xG) it received under the old model.

But why not 100 per cent? Could he really miss from there?

Well, remember expected goals doesn’t account for who takes the shot, so while Firmino may not have missed others have fluffed similar opportunities. Just ask Nathan Redmond, Timo Werner or Cristiano Ronaldo.

Likewise, goalkeeper position and the absence of defensive pressure boosted the likelihood of a goal from around 47 per cent to over 87 per cent in these strikes from James Maddison and Patson Daka:

Maddison v Brentford and Daka v Newcastle
(see video links to Leicester City YouTube channel: BRE, NEW)

Equally, some opportunities are now deemed harder thanks to the new contextual factors.

Richarlison v Wolves
(see video link to Wolves YouTube channel)

Richarlison was previously given a four in five chance of converting this attempt against Wolves but xG 2.0 puts it closer to one in five thanks to José Sá smothering the shot.

Unusually revealing

Pedro Neto v Leicester City
(see video link to Wolves YouTube channel)

The most dramatic change in xG value, however, is found in a long-range effort from Pedro Neto, which had a 19-fold increase from 0.035 to 0.656.

There will always be differences of opinion when it comes to single-shot xG – and the outcome could be affecting our perception of its quality – but if that chance was played out a thousand times, would you expect 656 goals?

Admittedly it’s a little unkind to Opta to pick out this one unusual example from the 6,899 non-penalty shots so far this term – it’s clearly an outlier; Kasper Schmeichel is 30 yards off his goal line – but doing so is revealing.

When factoring in goalkeeper position, Opta measure the distance to goal, distance to the shot, the distance from striker’s line of sight and the ‘keeper’s angle to the shot.

As the improvements in xG 2.0 demonstrate, the strength of any expected goals model depends on the variables it is taking into account.

That the new variables have upped the likelihood of Neto scoring from less than one in 25 to two in three shows the massive difference contextual factors can make. But could adding more data points further enhance the descriptive power of the model?

Aiming for perfection

Sadio Mané v Leeds United
(see video link to Liverpool YouTube channel)

This miss by Sadio Mané is rated the fourth worst this season, with an 83 per cent chance of a goal being scored.

If you look closely, however, you’ll notice that Jota’s pass takes a deflection off Meslier so at the last second Mané has to adjust his stride to reach the ball.

This information is not currently considered by xG 2.0, and, in my opinion, makes it a much harder opportunity than it first appears.

Despite taking into account details like which foot the player used to shoot, what type of chance it was, and the type of assist that created the opening, recording last-second deflections or the ball going slightly behind the striker may improve expected goal values even more.

It might not be perfect, as Fábio said, but if you know how to interpret it, it’s the most useful stat for FPL.

Best performers

The table below shows the top ten players for non-penalty expected goals to Gameweek 28. 

Aside from Son Heung-min and Mohamed Salah, the top ten have scored fewer goals than expected.

But this isn’t due to rebounds inflating xG totals, a concern voiced by Bowstring The Carp among others.

Adjusted xG

Bowstring rightly points out that it isn’t possible to score more than one goal from a single phase of play

For example, against Leicester in Gameweek 20, Salah’s penalty was saved, he then headed the follow up against the bar and volleyed the subsequent rebound wide. All in the space of three seconds.

Taken individually, those three shots had a cumulative xG value of 1.84 made up of 0.79 for the penalty, 0.54 for the header and 0.51 for the volley.

Salah’s xG, however, is adjusted to account for the same pattern of play.

So instead of 1.84 being added to his season total, only 0.95 is counted. This gives a wholly more realistic expectation of the number of goals he should have scored.

Cumulative xG

There are situations when you might prefer a cumulative xG total to Opta’s adjusted tally. Measuring finishing ability for instance.

Why is it better to look at cumulative xG for finishing ability? Because each chance should be considered individually.

Continuing with our example, when we allow for the fact Salah could only score one goal from that sequence, the two rebound efforts have an adjusted xG of about 0.08 each. As opposed to 0.54 and 0.51.

Ask yourself – if Salah was presented with that second rebound a million times, would you expect him to miss on nine out of ten occasions? Or is scoring one in two more realistic? Probably the latter.

The table below looks at which players get the best quality shots, excluding penalties.

PlayerShotsCumulative xGxG per Shot
Raheem Sterling429.170.22
Diogo Jota6413.680.21
Sadio Mané7614.880.20
Phil Foden468.820.19
Joshua King478.220.17
Conor Gallagher416.650.16
Son Heung-Min558.900.16
Mohamed Salah10015.870.16
Cristiano Ronaldo7712.080.16
Jarrod Bowen6610.260.16

Raheem Sterling tops the list with an average of 0.22 xG across his 42 shots. Meaning, we’d expect him to score at a better rate than one in five.

In fifth place we find Joshua King, who averages 0.17 xG per shot. But as any frustrated owner will tell you, this doesn’t mean much given his finishing ability this season.

Interpreting player performance

PlayerShotsCumulative xGxG per ShotxG per GoalGoals
João Cancelo593.340.063.341
Bryan Mbeumo466.000.133.002
Gabriel Jesus415.500.132.752
Adam Armstrong444.750.112.372
Joshua King478.220.172.064
Cristiano Ronaldo 7712.080.161.737
Chris Wood456.600.151.654
Leandro Trossard574.540.081.513
Harvey Barnes444.240.101.413
Daniel James465.390.121.354

Indeed, King needs, on average, two xG to score one goal ranking him among the poorest finishers in the league. 

João Cancelo is deemed the worst, however, scoring just once from an xG of 3.34. Although it’s worth noting that despite taking 59 shots up to Gameweek 28, he’s averaged just 0.06 xG per shot. So his odds of scoring any individual attempt is a lowly six per cent.

In fact, you could make a convincing argument for King being the worst given his superior xG per shot.

PlayerShotsCumulative xGxG Per ShotxG per GoalGoals
Kevin De Bruyne534.010.080.459
Emmanuel Dennis526.780.130.759
Mason Mount465.290.110.767
Youri Tielemans443.150.070.794
Son Heung-min 558.900.160.8111
Conor Gallagher416.650.160.838
Bruno Fernandes647.650.120.859
James Maddison516.010.120.867
Demarai Gray424.370.100.875
Riyad Mahrez425.270.130.886

Turning to the best finishers, Kevin De Bruyne has scored a goal every 0.45 xG this term. If Raheem Sterling could finish at the same rate he would have 20 goals instead of 10.

But the comparison is unfair – and this is where it’s important to interpret those results.

De Bruyne’s xG per shot is a mere 0.08, and looking at his shot heatmap we can understand why:

Just under half his goal attempts, 26 of 53, have come from outside the penalty area. Sterling, in contrast, has taken just eight of his 42 non-penalty shots from outside the box.

Despite the added contextual factors in xG 2.0, distance to goal remains a crucial component of expected goals.

So given his general shot location, De Bruyne is not going to have the same volume of xG as Sterling, despite taking more shots than him.

Indeed, looking at the table of best finishers, there are several players who are good at striking the ball from distance.

This makes sense because scoring with a shot from outside the box will boost your goals to xG ratio. For instance, Youri Tielemans’ belter against Brentford had an xG value of just 0.02.

Equally, missing a long-range effort isn’t going to add much to your xG total.

To me, the two players who stand out from the table of best finishers are Son Heung-min and Conor Gallagher, who combine a high xG per shot of 0.16 with a low xG per goal.

Predicting the future

It’s tempting to look at these tables and think that those who are under or over-performing will regress to the mean.

However, we must beware of Gambler’s Fallacy: the players have already ‘banked’ their over or under-performance. To draw an analogy, if we flipped a coin and (improbably) got ten heads in a row, it doesn’t mean the next ten flips will all be tails.

Nonetheless, it’s worth asking the question: are there players who consistently exceed or fail to reach their expected goals tally?

It’s an idea of Fabio’s that FPL Virgin picked up on in his examination of career xG earlier in the season.

Site improvements 

We recently revamped several tabs in our Members’ Player Stats and added to the metrics available in Members’ custom tables. Following Opta’s revamp of expected goals we will be making further improvements.

‘Cumulative xG’ will be added to allow members to make a comparison with Opta’s adjusted xG numbers already on site.

New stats ‘xG per Shot’ and ‘xG per Goal’ will make use of cumulative xG.

There are also plans to incorporate various career metrics. Whether that’s by creating individual stats, such as ‘Career xG’, or by adding an ‘All Seasons’ filter option. Testing will dictate which route is the most effective.

TopMarx Fan of Fantasy Football and Monty Python. "Archimedes out to Socrates, Socrates back to Archimedes, Archimedes out to Heraclitus, he beats Hegel. Heraclitus a little flick, here he comes on the far post, Socrates is there, Socrates heads it in! Socrates has scored! The Greeks are going mad, the Greeks are going mad! Socrates scores, got a beautiful cross from Archimedes. The Germans are disputing it. Hegel is arguing that the reality is merely an a priori adjunct of non-naturalistic ethics, Kant via the categorical imperative is holding that ontologically it exists only in the imagination, and Marx is claiming it was offside. Follow them on Twitter

122 Comments Post a Comment
  1. Baines on Toast...
    • Fantasy Football Scout Member
    • 13 Years
    1 year, 12 months ago

    Fascinating article, thank you. I do worry that the keeper position (particularly the Neto example) can slightly over-inflate how teams/players appear to be performing, but I suppose that's the reason we need cumulative stats as well.

    1. McSauce
      • 14 Years
      1 year, 12 months ago

      this is my team name!

    2. TopMarx
      • Fantasy Football Scout Member
      • Has Moderation Rights
      • 11 Years
      1 year, 12 months ago

      Yeah, I have emailed Opta to ask them what's going on in the Neto example, I'm trying to think of similar type chances but it's very much an outlier. And Opta have trained their xG model on something like 2.5million shots, so maybe chances where the goal is open and the goalkeeper is miles away are scored 66 per cent of the time? I'm interested to hear what they say. I've assumed that it's goalkeeper position leading to an inflated xG value but it might not just be caused by goalkeeper position, and it might not be that inflated! I'll reply here when I get an answer.

  2. Weak Become Heros
    • Fantasy Football Scout Member
    • 5 Years
    1 year, 12 months ago

    I own Rudiger and Cancelo. Who is the next transfer in:

    A. James
    B. Laporte

    1. Hotdogs for Tea
      • 8 Years
      1 year, 12 months ago

      Chalobah worth a punt as a second way in to Chelsea defence? May well play in PL games while others are rested

    2. Botman and Robben
      • 7 Years
      1 year, 12 months ago

      C. Robbo

      1. bso
        • Fantasy Football Scout Member
        • 9 Years
        1 year, 12 months ago

        Trying to do this myself but short of cash. May have to be James instead.
        Who are your transfer targets LR? I may chance it with Havertz.

        1. NABIL - FPL otai
          • 11 Years
          1 year, 12 months ago

          LRL

          LR is LateRiser or La Roja.

        2. Botman and Robben
          • 7 Years
          1 year, 12 months ago

          Was planning Bowen -> Kulu but waiting for pressers first. May just roll FT.

          1. bso
            • Fantasy Football Scout Member
            • 9 Years
            1 year, 12 months ago

            Good move.
            Worried now that Havertz will not start regularly with other games around it. Such a dilemma. Want to get rid of Coutinho. Not keen on Mount. Who would you go for with my team as is?
            Thanks mate.

  3. McSauce
    • 14 Years
    1 year, 12 months ago

    TAA --->

    A) Robbo
    B) James
    C) Cancelo

    1. NABIL - FPL otai
      • 11 Years
      1 year, 12 months ago

      C

      I want all 3. That's why my WC is active

  4. NABIL - FPL otai
    • 11 Years
    1 year, 12 months ago

    Unpopular opinion: Mahrez and Foden will outscore Havertz in GW31-33.

    1. bso
      • Fantasy Football Scout Member
      • 9 Years
      1 year, 12 months ago

      Maybe, if you have a high quality crystal ball and know who Pep will start.

      1. NABIL - FPL otai
        • 11 Years
        1 year, 12 months ago

        It doesn't matter. I've seen it.

        Pep wants rhythm, both will start.

        Kai will play more in the cup games from now on.

    2. Camzy
      • Fantasy Football Scout Member
      • 13 Years
      1 year, 12 months ago

      That seems very likely.

      Havertz I think plays 31 and 33. Rested in 32 around the Real Madrid game.

      1. NABIL - FPL otai
        • 11 Years
        1 year, 12 months ago

        Now we're talking.

        Remember GW33 is 3 days after the FA Cup SF game. I think Kai plays in the cup game.

        1. NATSTER
          • 13 Years
          1 year, 12 months ago

          City have UCL and FA Cup and also rotation too.

  5. AC/DC AFC
    • Fantasy Football Scout Member
    • 8 Years
    1 year, 12 months ago

    Nice article and it's clearly and logically explained.

    The paragraph below lost me a bit on cumulative xG.

    I get field position, distance and quality of strike on the ball as well as defensive positions are all factors in goal scoring conversion rates.

    And sheer volume of shots count.

    But can someone explain cumulative xG in simple terms please?

    1. AC/DC AFC
      • Fantasy Football Scout Member
      • 8 Years
      1 year, 12 months ago

      This is the part and table below it i didn't get right away.

      How it relates to the stats.

      Ask yourself – if Salah was presented with that second rebound a million times, would you expect him to miss on nine out of ten occasions? Or is scoring one in two more realistic? Probably the latter.

      1. TopMarx
        • Fantasy Football Scout Member
        • Has Moderation Rights
        • 11 Years
        1 year, 12 months ago

        Basically it's adding up the individual xG shot values and not adjusting them for the fact that you can only score once from a single passage of play.

        For most players there will be no difference between the two totals.

        But in that Salah example where his pen was saved and then he had two rebound shots, his cumulative xG was 1.81 and his adjusted xG (or just his xG) was 0.95.

        So when adjusting for the fact that you can only score one goal from that sequence of shots, the xG value of is second and third shots are around 0.08 xG each as opposed to 0.5 xG when take individually.

        Is that any help?

        1. AC/DC AFC
          • Fantasy Football Scout Member
          • 8 Years
          1 year, 12 months ago

          Yes, cheers.

          Of course fitting the data to the maximum outcome of 1 goal in a sequence is important and a big improvement if xG 1.0 glossed over that fact.

          1. TopMarx
            • Fantasy Football Scout Member
            • Has Moderation Rights
            • 11 Years
            1 year, 12 months ago

            xG 1.0 hadn't but we had, so I wanted to make it clear that's how it works.

  6. Rainer
    • 7 Years
    1 year, 12 months ago

    Brilliantly put together piece, TopMarx!

    The Neto xG seems too drastic a jump but the eye test alongside xG might help in these (rare) situations.

    1. TopMarx
      • Fantasy Football Scout Member
      • Has Moderation Rights
      • 11 Years
      1 year, 12 months ago

      Thank you, and yes exactly, my point with the Mané example too; the eye test is still important - is there something that's obvious to us when watching that isn't being picked up by the data collectors?

      I think the data collection process is fascinating, and that does evolve every season - Opta adjust and add to the detail of the events they collect. These xG models can only be as accurate as the data you input.

      I sometimes think about what data I would collect if I was starting from scratch, especially with FPL in mind. Although I don't want to become a modern-day Charles Reep!

  7. OptimusBlack
    • 10 Years
    1 year, 12 months ago

    Best Striker to get if u have Both Kane & Laca (Jiménez)
    Jiménez >

  8. MHG
    • 6 Years
    1 year, 12 months ago

    Looking to try and squeeze Foden into my WC:

    A) Rudiger & Kulu

    or

    B) Chalobah & Foden?

    Thanks

    1. NABIL - FPL otai
      • 11 Years
      1 year, 12 months ago

      Neither. Kai to Foden

      1. MHG
        • 6 Years
        1 year, 12 months ago

        Thanks for checking my team. It's a option for sure!

        1. NABIL - FPL otai
          • 11 Years
          1 year, 12 months ago

          I didn't. I only assume you must have Kai in your WC draft 😉

          1. MHG
            • 6 Years
            1 year, 12 months ago

            🙂

  9. Twigs
    • 5 Years
    1 year, 12 months ago

    Jimenez to Weghorst?

    Or

    Dubravka to Pope?

    Rest of team:
    Dubravka/Foster
    James/TAA/Dier/Coady/White
    Son/Salah/Saka/Barnes/Kulu
    Jimenez/Wood/Broja