Expected Goals


#1

So there was a tweet by Saddlers Stats after the Wimbledon game that showed the following:

This is the perfect example of why expected goals is complete nonsense. We never really looked like conceding outside of 10-15 minutes in the second half. They took potshot after potshot first half. Anybody can do that. We looked like scoring every time we broke forward. I can’t remember Liam Roberts having to make a serious save.


#2

It’s definitely nonsense. Means absolutely nothing but pundits seem to love it.


#3

It means nothing useful on its own, but with explanation, yes it is.


#4

Nah, pure nonsense.


#5

Lies, damned Lies and then statistics! No doubt keeps some otherwise unemployable graduate in a job.


#6

Page 6 of this week’s Racing and Football Outlook looks in depth at the stats for shots to goals ratio. It remains almost mystically constant at 1 in 10 scored, but it’s much better to get the ball in the area before shooting , even if some moves break down in the attempt.

Remember a few years ago we had a rather tedious troll saying, " Walsall need to shoot more ?"Turned out we were low scorers but having plenty of shots off target. So the stats actually suggested that “Walsall need to shoot less.”


#7

During the Villa-Brentford commentary they mentioned that Brentford and Dean Smith in particular put a lot of time into ‘Expected Goals’ and it’s findings as part of their statistical analysis which they’ve adopted in recent years thanks to their chairman’s company Smartodds.

They base signings around the approach and haven’t done too shabby recently.


#8

He’s also an excellent coach who preachers a very proven style of play.

If he does use expected goals for certain signings then it must surely be one of many factors because every time I’ve seen it used it’s been guff.


#9

It’s not nonsense though. The data and explanation all stack up. Wimbledon had lots of shots (19) with a low probability of scoring (say 10%) 19 *10% =1.9 close to the 2.3 the tweet says.

Can’t remember all Walsall’s attacks in detail but for three goals we scored:

  • Morris’s goal was a ‘worldy’ as Merson would say, i.e would only happen say 1 in every 5 times (0.2)

  • The 2nd goal was an complete fluke that would happen 1 in 100 times (0.01)

  • The 3rd goal was also a deflection again, say that would happen 1 in 4 times (0.25)

Add these up you get close to 0.5. Add in Walsall’s other attacks I can easily see you get to the 0.8 quoted in the tweet.

Or to think of it another way if walsall hadn’t scored their 1st and 2nd goals and 2 of Wimbledon’s long range shots had deflected in (not that unlikely seeing they had 19 of them), the score would have been 2-1. Not too far off the XG score of 2.3 v 0.8.

It’s far from a perfect metric, but I think XG is a useful statistic. The key stat is 1-3 though :grinning:


#10

and the only one that counts for anything.


#11

You’re forgetting, and so is the data, the three or four times we broke, particularly down the left in the first half and cut back with their defenders at sixes and sevens with the merest touch needed for (another) goal. I’d rather rely on that than hopelessly belting it miles away. The cookie goal is harsh to say you shouldn’t “expect” him to score. Ok he nearly made a mess of it but it’s found him yards out. I’d expect him to score that a lot more than I’d expect one of theirs from 20 30 yards.

If you watched that game and “expected” Wimbledon to score more than we did then you are delusional, that’s why this information is nonsense.


#12

You’re forgetting, and so is the data, the three or four times we broke, particularly down the left in the first half and cut back with their defenders at sixes and sevens with the merest touch needed for (another) goal.

Well I did say “add in Walsall’s other attacks and you get to 0.8”. You may be right though, I haven’t got time to watch the whole match to comment, although I just rewatched the highlights and other than a Cook shot from distance in the first half they didn’t show these chances you describe which sort of implies they weren’t brilliant chances.

Clearly if the data is excluding Walsall’s chances then obviously the output will be nonsense, That’s true of any statistic. It doesn’t make the concept of Expected Goals nonsense though.

I’d rather rely on that than hopelessly belting it miles away

So would I!

That’s kind of the whole point of XG isn’t it though? It puts a higher weighting on well crafted chances compared to long range blasts. However, if you have enough long range shots one will go in eventually. So, say 10 long range shots are equivalent to 1 well crafted chance in terms of the goals you would expect to occur from doing this.

If you watched that game and “expected” Wimbledon to score more than we did then you are delusional, that’s why this information is nonsense.

That’s because you get more information from watching the game. E.g the Wimbledon own goal was probably caused by the pressure Walsall put on Wimbledon around this time. xG doesn’t capture that, but watching the game does.

As I said in my original post, XG is far from perfect, but to claim it is nonsense is just showing a lack of understanding.


#13

Today we “only” had 6 shots to Rochdale’s 13. However, we had 5 on target to their 4. That’s good shootin’ and a much, much better conversion rate than we used to manage under DS, where we out-possessed and out-shot other teams on a regular basis without always getting the right result (although, let’s face it, we did OK under DS, to say the least).
Conclusion? There’s more than one way of skinning a cat. This season, we seem, for the moment anyway, to have that clinical touch (and a bit of good fortune, to boot). Long may it continue :grinning:


#14

I know you did. And it adds to more than 0.8 which is why I thought you’d forgotten.

Why would they show our wingers continuously skinning their full back and pulling the ball back or firing it across the face with no shots?

It’s not a lack of understanding it’s common sense. If I watch a game and I think I expect one team to score and the other to not, and a stat tells me the opposite that’s nonsense. It happens regularly with that “stat”. And I mean if you think that isn’t a fair assessment find me a report that says we were lucky to win or indeed “expected” them to score more than us.


#15

I know you did. And it adds to more than 0.8 which is why I thought you’d forgotten.

Why does it add to more than 0.8?

Why would they show our wingers continuously skinning their full back and pulling the ball back or firing it across the face with no shots?

Either these moments you describe were goal scoring opportunities for Walsall or they weren’t

If they were I’d expect to see them on the highlights, but they weren’t, which was my point.

If they weren’t goal scoring opportunities then it makes sense that Wasall have low XG because they wouldn’t contribute to XG

It’s not a lack of understanding it’s common sense. If I watch a game and I think I expect one team to score and the other to not, and a stat tells me the opposite that’s nonsense. It happens regularly with that “stat”. And I mean if you think that isn’t a fair assessment find me a report that says we were lucky to win or indeed “expected” them to score more than us

But that’s just subjective. I can’t debate with your view of which team you expect to score next, you’re not giving me any information why you are thinking that.

The xG stat just looks on the chances that actually happened for both sides and comes up with an overall value for both sides.

It could be that you have a different view of the chance of a certain player scoring a goal.

For example if Messi had a shot from 30 yards and Ian Roper had a shot from 30 yards, XG would give the same value (say) 0.1 of a goal.

You and I both know that’s nonsense, of course Messi would be more likely to score and so should have a higher XG. but XG isn’t sophisticated enough to know that.

That’s why I’m saying XG isn’t perfect, but if you understand what it is measuring then it’s reasonably useful.


#16

Of course they were goal scoring opportunities? Would you count them as that? Or are you going to totally discount them because they weren’t included in a two minute clip of a game that had 4 goals? Actual ones not expected ones, three of which were scored by us.

I’ve given you plenty of examples of why I thought that throughout this thread! How many times? That every time we broke forward we looked very dangerous and that every chance they had seemed to be a pot shot from miles away. And, you know actually watching the game. I can’t make it any clearer for you to be honest why I think that, I’m at a loss.

It’s measuring something that is completely useless though. You may as well use shots at goal if it’s only good enough to analyse something that vague. That’s my point and that’s why you think I don’t understand it. Because I’m not actually arguing about the method or the maths behind it at all. It’s neither here nor there to my point. I’m talking about the result of the analysis and it’s usefulness. Which to me is completely nothing when actually watching a game can tell you a completely different story.