Total xG Explained: The Most Misunderstood Statistic in Football | Part 1
Vložit
- čas přidán 19. 12. 2023
- Total xG is perhaps the most misinterpreted statistic in modern-day sports. In this video, I delve into the mathematical principles behind the expected goals model. Stats are great at trying to uncover patterns in the game but these tools are only as good as our ability to understand and implement them. In the intro I lay out 3 things to cover but editing took longer than I expected so this is the first of a three-part series uncovering the tricky details behind xG. Why is it so widespread if it’s not accurate? Find out in Part 2.
In the intro, I mention 170 games but wrote 169. This is because the Bournemouth v Luton game has been suspended. Thoughts go out to Tom Lockyer, his family, and everyone involved.
Thank you for all the support!
why you have 100subs?
Thanks a lot. Didnt understand the equations but got the gist of it. Well presented.
The biggest problem is teams react to the real time scoreboard, and ultimately don’t care about ‘winning xG’
If a team scores a long shot early on, one that maybe only has a 0.05 chance of scoring, the scoreboard is 1-0. They don’t care about 0.05-0. The rest of the game they are likely to play defensively, leading to less chances created AND reducing the real world quality of their opponents chances.
Had the long shot not gone in they would have played differently, creating more themselves and allowing more space for their opponents.
Other gripe with xG - it treats all players as equal. Salah taking a shot on his left foot from the edge of the box is far more likely to score than a league average player in the same scenario. Best example of this is penalties, which are always ~0.78, yet there are some pen specialists who average well over 90%.
I feel like all those aren't disadvantages but actually help filter the standout players from the baseline average.
@@glokta1 you’re right, and I believe that’s what xG was initially meant for, but for how it’s colloquially used, ‘which team deserved to win based on quality of chances’, then it doesn’t do a good enough job for me
Another day another banger 🔥
excellent analysis.subbed
I say this for ages, but all these laptop managers discussing football with me, not every played a ball, are actually maths fetishists.
yup, most people either come from a playing background or a stats background. very few people who have a large voice have a solid understanding of both.
I read on reddit at least 20+ reasons why its a crap statistic. Comparing xG/xA is one step away from comparing whether or not if batman or superman would win in a fight.
Very well presented video Aqua Melon and an interesting take.
I notice you said at the end that you're still learning how to make videos. You've got your visuals down well.
My constructive criticism would be to focus on your audio a bit more. It sounds like you're trying to speak quietly so you don't disturb anyone.
Your pacing in how you speak is good but the level of your voice is inconsistent which is partly due to your recording and partly due to the audio editing. 2:09 for example 'the moneyballification of all modern sports began', is difficult to hear on a first listen. Whereas 6:58, the audio is much better and can hear your voice way more clearly.
Can't give you specific advice on how to improve it as it depends on your recording setup and your editing software but look into it. Remember audio is 50% of your video and people are much more forgiving of poor visuals if the audio is good. Whereas if the audio is bad but the video is good, it's less paletable.
Some general tips would be to listen back to your video at the end without headphones and also listen at 2x speed to make sure everything is easily understandable. And also be careful with adding background music, make sure the levels prioritise your voice.
Left a like and subbed for more, just trying to help :)
I’ll keep that in mind, thanks!! 🙏
I feel like this video itself misunderstands the point of xG. It's not there to predict the final scoreline, we already have a statistic that predicts that perfectly, which is the final scoreline.
It's simply there to give a more accurate indication of the general performance of the teams as the final scoreline rarely (or even indeed almost never - which is where the 1/6 chance comes from) represents what teams 'deserve' given the chances they get.
You do have a point that it doesn't perfectly indicate the winning chances of each team, but even in this extreme example of 20 super low-scoring opportunities vs 2 high scoring ones, the result is only a couple of percentage points. And that whole calculation assumes the xG model to be accurate on every individual shot (as it is fairly sensitive to errors on the individual chance estimations), which is not really the case. XG is more a measure that accurately averages out larger amounts of chances.
So it's true that it's not 100% accurate but it's an intuitive, concise and better indicator of the quality of a teams chances than the final scoreline, which in my opinion is a good reason to include it in the statistics alongside the scoreline. It's not there as a replacement to the final scoreline nor to proclaim an alternative winner, it's there to give a general indication of the amount and quality of both teams' chances in a better way than either the scoreline, possession, total shots (on/off target) or any other statistic could possibly tell you.
Of course it's only a statistic calculated by inherently flawed models so we shouldn't blindly put all our faith in it, and I imagine seeing other people do this is probably where your backlash to it comes from.
yup, you’re right. my main gripe isn’t about xG itself, it’s about adding total xG when there’s a small sample size (as in a single match). I gave a simple example with clean numbers but in a normal game the individual xG values vary greatly. In a case like that, the variance plays a bigger roles to show how the score line can vary from the expected total goals in a game. Happy to get into that a bit more but it’s a little messy with the math- lmk!
Amazing video and style man! Loved it totally on such a unique topic and I request you to not change your video styles and narration voice even when you go big! (which I know you will). Keep making videos regularly, will be looking forward to the next one
thanks so much man!! lots of room to improve 😤🙏
This video is amazing
I agree that the xG obfuscates the quality of chances, but is xG trying to give you an accurate scoreline? Or is xG a base statistic to see which team theoretically played better?
I think it is fairly intuitive that higher the xG, the more goals a team would score so why does it matter that xG cannot predict the scoreline precisely? What does your stats show on xG correctly predicting the outcome of the game instead of the scoreline?
soccer/football is a game of small margins and the sample size for game changing results is so small that it’s hard to give predictive stats on a game-by-game basis a lot of confidence. xG should be used when looking at a lot of shots. you’re correct in that higher xG is in line with intuition for what is a better chance. we should be comparing individual chances for better analysis. when we start to add xG up without a large sample size, we introduce more error such that it makes the stat less helpful.
Hey, good video! For me personally, xG helps me to get a cryptic idea of how a game that i haven’t watched went. A few years ago you didn‘t really have access to these kind of statistics- you only had possession, shots on target etc. There were situations in which I couldn‘t explain to myself how a team lost that had 15 shots on target, scoring zero goals while the other team won, having scored twice from four shots on target. After watching the highlights I understood because the losing weren‘t able to break the opponents defense- just shooting from outside the box while the winning team stayed patient and worked the ball into the box. xG isn‘t accurate in predicting scorelines but I think it gives a pretty good overview combined with the number of shots on target. Also, there is a reason why professional scouts and Managers like Thomas Tuchel value the information xG gives. Nevertheless you made good points! I hope you gain some subscribers, you definitely deserve more.
yupp xG is a really good tool if used properly (I’m sure tuchel does). I started this series to see how valid xG is in terms of statistical significance because, like you, I found myself looking at the total xG for games I hadn’t watched. stats are definitely going in the right direction but we need to continue improving metrics to be more representative and intuitive. thanks for the feedback and food for thought!!
Very interesting video. Is there somewhere that records these cumulative probabilies? Would be interesting to see for other matches.
not that I know of- addressing this in my next couple vids!
Great 👍🏻
Didn't understand how the win probability was calculated though
I got the win probabilty by adding up the cases where Tottenham scores more than Man City (6:37). The percentage values in those cells are found by multiplying the probability of each team scoring 'n' number of goals. For example, we get 18% for a 1-0 win to Tottenham by multiplying the chance of Tottenham scoring 1 goal (5:37) by the chance of Man City scoring 0 goals (5:57). Hope that helps! I'll try to pace my videos better in the future. :)
@@realdoritosman Clear!
I was watching a game and the striker had an open goal, which was apparently only .10 xG, Like what?
I feel like you can't really discuss this without mentioning variance