Explaining Expected Goals

I often get asked “what do these numbers even mean?” when talking about Expected Goals and how it applies to United, individual players, and football in general. It’s a fair question and one I’m happy to explain (hopefully in an interesting way!)

Opta, the god of all football statistics, define Expected Goals as:

 “Expected goals (xG) measures the quality of a shot based on several variables such as assist type, shot angle and distance from goal, whether it was a headed shot and whether it was defined as a big chance.”

In essence Expected Goals is a metric which assesses every chance, essentially answering the question of whether a player should have scored from a certain opportunity. It does this by pulling together data on thousands of shots from all over the pitch, and looking at the percentage of times that similar shots resulted in a goal. That percentage is the Expected Goals number for that particular chance, usually expressed as a decimal eg 0.40.

Ever watched Match of the Day and heard the phrase “9 time out of 10 he scores from there” – well, does he? Let’s take a look at an example:

Here’s our very own Leon Clarke heading straight through on goal against Middlesbrough a few games ago.

clarke chance.png

For every Blades fan out there, this is a simple must score, heading 1 on 1 with the keeper with no defenders. However, is it actually a “must score”? Here’s how my xG model sees this opportunity:

  • No defensive pressure – Big tick

  • Shot type – Left foot shot

  • Shot zone – Zone 2 (more on this later, but in essence quite central and quite close)

  • Assist type – controlled pass – big tick

This shot at goal that Leon “must score” is actually only given an xG value of 0.43. Only 43% of the time on shots I studied with the same details as this one – same location, same shot type, same amount of defensive pressure, same kind of assist - actually ended in a goal. Less than half of the time, this would be scored. Shocking right!

Less than half the time? Are you serious?

Instantly I can feel the outcry, “that’s just nonsense, he has to score”. Well let’s look at some further examples. Generally, in most xG models, a penalty is assigned an xG of 0.77. This is because only 77% of penalties (not in shootouts) are actually converted. I imagine that a lot of fans would expect their penalty taker – their best penalty taker, remember, as these are not pens from shootouts – to score 9 times out of 10, maybe 19 times out of 20. On average, it’s a fair bit less than that.

Add to that the fact that only 1 in approximately every 9 shots in the Premier League actually result in a goal and now you’re starting to see that not many shots actually end up in the back of the net. Which makes perfect logical sense: think about how many shots as a team has in a game compared to how many goals they actually score in a game. Not exactly equal.

Let’s a view another example, from our game against Swansea.

mcgoldrick chance.png

Here David McGoldrick has recovered the ball and is driving towards goal. He has defenders in front of him making his shot slightly harder, but still, at the edge of the 18 yard box and on his strongest foot I think we are all expecting to see him hit the target and many would be thinking he should score from this transition. Here’s what the model assigns this chance:

  • xG value – 0.154

Only 15% of shots from this situation (all variables included) actually result in a goal.  

Hopefully the message is becoming clear at this point. When you actually look at a large sample size of shots from various locations, lots don’t actually result in goals from positions where you maybe think they should. I for one found this both fascinating and frustrating when I first learnt about xG. For example, my eyes and my heart still tell me the Leon Clarke chance above has to be a goal, and quite frankly I almost still do. However, the fact stands that in actual reality that shot should not – and does not - always result in a goal.

Key Point to consider – not many chances get a rating of 1 xG – it's almost impossible unless the ball is on the goal line, on the floor with zero defenders. So a rating of 0.43 can actually be quite large in the context of other chances within the game.

How do Expected Goals apply to analysing a game?

Big point here – Expected Goals should be based as a long-term metric to build a large sample size, and its usage is not best-placed in analysing single games. I’m reluctant to do so, however that said it can help paint a picture of the game and show whether what you felt when watching is how the game “should” have panned out. In a single-game situation xG summarises all of a team’s chances by adding up all of their chance values to give an overall figure.

So when we see something like this:

Expected Goals:

SUFC 1.45

Swansea 1.11

What can that tell me about the game itself? It tells us that in this particular game the Blades were unlucky to lose. I tend to have an easy rule-of-thumb that if the xG difference between each is side is 0.3 or lower then it’s fair to say the game should have been a draw.

An xG of 1.45 doesn’t specifically say we had loads of chances. One concept with xG is that two high-quality chances of 0.8xG (or "the Billy Sharp range" as I like to call it) is the same as eight chances of 0.2 xG, both would total 1.6 for the game xG. That’s where the analysis of the game comes in: in this particular game Swansea actually had the higher quality chances in terms of xG: they created clearer chances than the Blades in the game. United had more efforts, but from lower-quality situations.

One other thing to take into account when looking at individual games is that the “game state” can affect the overall team xG. For example, United went 3-0 down early on to Middlesbrough, who then sat off for most of the rest of the game. United created plenty of chances and the xG of the team ended up being quite close by full-time – or at least, a lot closer than you’d expect for a 3-0 defeat. That in itself does not suggest that United were “unlucky” to lose 3-0. 

Expected Goals – A player’s example

At its core, Expected Goals was designed as a tool to find players who were taking up great goal scoring positions. Whether they finished and converted those chances is kind of beside the point at the initial stage, you just want to highlight the people who should be scoring lots. Let’s take a look at some of last season’s Expected Goals leaders in the Championship:

Lee Gregory 17/18

  • xG – 18.71

  • Goals scored – 10

From the quality of chances that Gregory had last season, and the goal scoring positions he got himself into, Gregory would have been expected to score around 18-19 goals. The fact he actually only scored 10 is where the real question lies. Did he only score 10 through bad luck, and actually this season Gregory will hit the form he should actually hit and score lots of goals? Or, did he score less because his finishing skills are actually quite poor, despite him getting into good goal scoring?

Leon Clarke 17/18

  • xG – 14.63

  • Goals Scored - 19

Very brief one here: Leon exceeded his Expected Goals. That suggests he’s a player with high finishing skill or quite simply he had a purple patch where everything he hit ended up in the net.

With players who over perform in a season we are always interested in how their next season plays out: if he over performs again (a la Harry Kane numbers) then Clarke is just an exceptional finisher. However, my guess based on his history is that this was simply a purple patch for Leon and he will regress this season. The most positive aspect of his over performance was the fact his Expected Goals total was still the 7th highest in the league, so even if he does regress he would still have scored a bucketful of goals due to him taking up good goal scoring positions and our play in setting him up.

xg to date.png

So who’s looking good this season already for Expected Goals? The results on this table were taken prior to this weekend’s games (18-08-2018) therefore were taken before the SUFC vs Norwich game.

Some names on this won’t surprise anyone and nor should they. Interesting to see Clarke and Gregory in the top 15 again, showing they are again taking up good goal scoring positions, even though it’s not happened for Clarke yet in terms of hitting the net. The most interesting man in this list is our very own Billy Sharp who has the uncanny knack of hardly shooting at all in a game but finding himself in the best goal scoring position possible for the 1 or 2 efforts that do come his way.

Still reading? Good!

In another article I may well go into my methodology and my model but quite frankly I wouldn’t want to bore you Blades! I hope this has given you a brief overview of what xG is, how it is worked out and what it can help to explain. Throughout the season I will be writing articles and showing examples of Expected Goals vs actual performance and I often find the results very interesting. I hope you guys do too!

If you have any questions about this article or anything else related to performance analysis on the Blades, leave a comment below or get me on Twitter @Blades_Analytic.

This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform. (www.stratagem.co), (https://app.stratabet.com), (https://stratatips.co)

Jay Socik3 Comments