Form Table according to xG

[beatific blade]

Goals win you games not expected goals
I was waiting for a bus expected at 16.30 but it didnt turn up

Ridiculous analogy

Love it when people come out with smart arse comments to try to undermine what is undeniably THE most profound advanced analytical data out there.

 

[Pedro Verde]

Ridiculous analogy

Love it when people come out with smart arse comments to try to undermine what is undeniably THE most profound advanced analytical data out there.

Actual goals win you games
Ive never seen an Xg scored so can you prove to me that these win you games
It may be advanced analytical data but it still dont win you games unless you play FIFA on your XBox

 

[Antarctic Monkey]

Based on Didzy’s xG he is currently at 0.03 goals per chance

 

[Cerberus Blade]

If each individual shot has a 0.1 chance of going in then you'd expect a goal on average every 10 shots.

That's not misleading, it's basic probability exactly the same way as saying that rolling a six on a fair dice happens about every six rolls. Now, probabilities don't work very neatly in practice so you'd see things like someone scoring three 0.1's in a row where another striker might have missed 20 in a row, but over a large enough sample size it'll be about right.

The problem is with your assumptions. Shots from 25yds go in far less frequently than 0.1. And secondly, can you imagine a team allowing someone to have twenty shots from the edge of the box in a game?

The maths is essentially proven. You can demonstrate it with coin flips, dice rolls, or any other repeatable chance event. The only problem that XG has is the samples used to determine the value given to any given chance because, unfortunately, that has to come down to a set of selection criteria that will either be subjective, missing, or very hard to factor in. Minor things like distance from goal and the angle to the goal are easy to do, but you could come up with all sorts of other criteria that will have an effect like weather conditions, pitch quality, pressure from defenders, lighting, and so on. All those things will have some impact on the "true" probability of a shot going in but may be impractical to take account of due to either measurement or sample size issues.

So it's not a question of "can we model shots on goal mathematically?". It's a question of how well we can model it. And XG seems to be a pretty good predictor of results.

I think you explained that very well. I agree that XG does appear to be a pretty good predictor of results. It's not by any means perfect though and you've explained why that's the case as well. The thing that interests me more about XG is that it gives an insight into the "performance" of the team, not just the results. And I think, over the course of several matches, it's the performance of the team that is the best indicator of how future results may come.

 

[Ainsley Harriott]

XG is the biggest load of fucking shit statistic I've ever known in football. I'm sure whem we beat Brighton 1-0 our XG was 0.83 or summat.

Absolute testicles

 

[Scooter_mcgavin]

Proof that xG has little bearing is the table in the OP. Nobody has scored the same amount of goals as their xG suggests.

Also xG doesn't take into account the actual players skill involved in the chance.

 

[PokerBlade]

Actual goals win you games
Ive never seen an Xg scored so can you prove to me that these win you games
It may be advanced analytical data but it still dont win you games unless you play FIFA on your XBox

The problem is that it's absurdly reductionist to view the game this way. There's a lot more to the tactics of football than a manager or coach saying "Only goals matter, so go out and score six of them". Scoring is difficult. How you go about achieving the aim (and preventing it at the other end) is how you win games. And XG is a good reflection of how consistently you will achieve the aim of getting goals.

Expectation is a statistical model so, no, you don't see XG winning a game in a simplistic sense. I don't see expected value flipping a coin but I'm still confident that if I flip a coin enough times it will land heads about 50%.

I think you explained that very well. I agree that XG does appear to be a pretty good predictor of results. It's not by any means perfect though and you've explained why that's the case as well. The thing that interests me more about XG is that it gives an insight into the "performance" of the team, not just the results. And I think, over the course of several matches, it's the performance of the team that is the best indicator of how future results may come.

Couldn't agree more with the last part. Again, it's something that playing poker at a decent level really drills into you (and believe me, it tests your sanity at times). I've been in situations where I've got all the money in and my opponent is dead to one card out of the forty four remaining. And they hit it. It's painful but it's silly to look at the situation and think you've done anything wrong. And it's silly for people to say things like "probability doesn't win you money, cards do" because 98% of the time I'm going to win all the money making that same play. What we want is a way to look at how we played today and be able to have confidence that we'll win games in the future because that allows us to work on tactics and strategy. XG is a better tool for that than the actual result.

 

[GRUMPY BLADE]

I think even when you understand the concept we're just very bad at visualising what that actually looks like. After the early ipods they made the shuffle feature intentionally less random because people complained that it did things like playing the same track twice in a row, playing multiple tracks off a single album, stuff that they thought shouldn't happen, but stuff that happens a surprising amount with a random selection.

Another one that gets people is take these two numbers:
3981462705
9231957774

One of them I got from a random generator on google, the other is one I put together intentionally. When they do this test most people will say the first number is the random one. But it isn't. The first one is much less likely to occur by chance because it contains every digit from 0-9. And when people try to make up a random number they usually do what I did with that one. The second one contains two 9's and three 7's in a row and so people think it's less random when in actuality repeated digits are pretty common.

A strategy in poker is using randomisers. It comes up in theoretically optimal play, but also for deception, that you might want to bluff with a certain holding say 70% of the time (so an opponent might think you're bluffing a lot more than you actually are). And it's basically impossible for a human player to do this well without assistance. Online players at high levels will have a RNG open to help them because otherwise you really have no idea if you're doing it well or not. Even when we understand the concepts we still can't apply them well.

But what impact does VAR have?

Is it accounted for?

 

[essexblade73]

Xg is an excuse for spotty number crunchers to seem relevant in the football sphere! Piss off back to rugby.

 

[BushBlade]

XG is the biggest load of fucking shit statistic I've ever known in football. I'm sure whem we beat Brighton 1-0 our XG was 0.83 or summat.

Absolute testicles

I don’t remember us having that many clear cut chances in the Brighton game and their defenders should have dealt with McB, I’d be livid if our defenders played like that. So, we ‘did a job’ on them. That’s what the Xg is telling us, nothing more than that. I’ve found that in most cases, when I leave a game saying ‘we were a bit unlucky today’ or ‘we deserved that’ it’s usually backed up by the Xg stats. They don’t really tell you much more than that, IMO, which is why it can appear to be testicular, as people sometimes try to use it to show things it can’t.

 

[PokerBlade]

But what impact does VAR have?

Is it accounted for?

I'm not sure what you mean, but I suppose some of the goals in XG samples might have been disallowed under VAR. If the sample size is big enough then that effect will be negligible, but as I said earlier in the thread the value of stats always comes down to how good your sampling is.

 

[guancheblade]

If you're interested there's some discussion of why that couldn't work yet in this Tifo football podcast: https://open.spotify.com/episode/0pd4pmVMExgfAAZB1sTWEz?si=UjQcbFrnQdaJSXgu-EpE9w

TLDR basically is that there isn't enough data to build a model that understands the difference between a McGoldrick shot and a Sharp shot. To have a full understanding we'd need to have lots of data on McGoldrick taking those shots hundreds if not thousands of times, with the same for every striker in the league

They only need to wait till the end of the season, he'll have racked up enough sample data by then!

 

[beatific blade]

For the non-believers among us: James Tippett has published a book (with fairly large font and simple language).

It’s called The Expected Goals Philosophy.

Read it. Learn something. Leave your cynicism behind. Liberate yourself from ignorance.

Follow him on Twitter @xGPhilosophy

 

[jt64]

If McGoldrick's personal xg is available for tbhe season it would speak volumes to guancheblade .

It's available on the site the OP listed, last time I checked he was on over 7 xG and in the top ten in the league for xG/90

 

[ThatJa]

Someone like Billy Sharp ?

Curious, if Sharp played every game (90mins) from now until the end of the season, how many do you think he'd score?

 

[Berlin Blade]

It’s embarrassing that so many of you don’t understand XG to be honest. Even MOTD uses it. You have to remember, personality and player ability is completely taken out of the situation. It’s based on if the average player was taking a shot from a certain situation. Based on the ‘long shots’ scenario people are discussing above, look at Michael brown in the 02/03 season. His XG that season was probably about 6 or 7. That means that the average player would have scored 6 or 7 based on the chances he had that season but he scored over 20. That’s why you know he is having an unbelievable season and is an outlier. That’s why he went to spurs. If you start including things like ‘this guy is good at long shots so up the XG’ it completely defeats the object. That’s how you see if someone is above the average of the standard you’re playing at. I mean, how can you not understand this? The only problem with XG is the data pool. The bigger the pool gets, the more refined and accurate it will get.

 

[Berlin Blade]

Put it this way, mcgoldrick’s XG suggests that there is a 0.08% chance an average player would have scored 0 goals this season based on the chances he’s had. Is that because he has been incredibly unlucky or because he is massively underperforming infront of goal? That isn’t clear, but you can make your own assumptions based on the data. If billy sharp had taken the same amount of chances, the XG would be exactly the same but he probably would have got off the mark by now.

 

[bartman]

Curious, if Sharp played every game (90mins) from now until the end of the season, how many do you think he'd score?

No idea but it would likely be more than McGoldrick( who he is ahead of already), Robinson ( who he is equal to in league games despite only starting once) and probably on par( or just behind) with the other two.

 

[DronnieBlade]

Or, when you watch the poker on TV they give mathematical percentages for the likelihood of winning a hand. They’re purely statistical and don’t account for how the players will play their cards but they’re still useful. I wouldn’t necessarily know if an ace/king is better than a pair of sixes but the percentage tells me immediately.

I do .

Fancy a game ?.

UTB

 

[PokerBlade]

It’s embarrassing that so many of you don’t understand XG to be honest. Even MOTD uses it. You have to remember, personality and player ability is completely taken out of the situation. It’s based on if the average player was taking a shot from a certain situation. Based on the ‘long shots’ scenario people are discussing above, look at Michael brown in the 02/03 season. His XG that season was probably about 6 or 7. That means that the average player would have scored 6 or 7 based on the chances he had that season but he scored over 20. That’s why you know he is having an unbelievable season and is an outlier. That’s why he went to spurs. If you start including things like ‘this guy is good at long shots so up the XG’ it completely defeats the object. That’s how you see if someone is above the average of the standard you’re playing at. I mean, how can you not understand this? The only problem with XG is the data pool. The bigger the pool gets, the more refined and accurate it will get.

Michael Brown is a really good example when you then consider he only scored about 12 goals in the rest of his career after that season. Statistical variance is a crazy thing and a single season isn't enough for variables like this to play out for an individual player. It's quite likely that Brown got really lucky for a season when it came to goals and was never going to sustain it. Granted he had plenty of other qualities that made him a decent player, but nobody beats the odds for long.

 

[Sabella's Socks]

For the non-believers

Liberate yourself from ignorance.

Follow him

But keep your scepticism close at hand, always

 

[beatific blade]

Put it this way, mcgoldrick’s XG suggests that there is a 0.08% chance an average player would have scored 0 goals this season based on the chances he’s had. Is that because he has been incredibly unlucky or because he is massively underperforming infront of goal? That isn’t clear, but you can make your own assumptions based on the data. If billy sharp had taken the same amount of chances, the XG would be exactly the same but he probably would have got off the mark by now.

We must consider that part of the reason why McGoldrick has a very healthy xG output is because he gets into goalscoring positions often and helps create the chances that come to him.

It is too literal to suggest that Billy Sharp would take those chances. He might not get the same quality or quantity of chances in the first place.

That being said, Billy does also have a very good xG-per-90mins output - albeit with a smaller sample size than McGoldrick.