Just like any prediction, XG is best used as a directional, not an absolute metric. No model could use past chance conversion stats of say, Rashford and predicts his performance this season!
The models are not reality is why: They're dependent on the input data which means they're as blind about what is not put into them as any other predictive system. As good as Rashford has been this season, without an out-and-out striker eg Kane, I'd anticipate his numbers to go down again as ManU are worked out in future projection for example.
Idk, early on in the season Rashford was accumulating a lot of xG, he just wasn't converting them. It was reasonable to predict, based off of his previous season performances, that he was about to start scoring prolifically.
Even seeing it as directional is the gamblers fallacy, assuming there is a definitive predictble underlying pattern or trend that is likely to continue. XG is based on retrospective analysis. There is no certainty at all whether the score of the next game will be up or down on the previous xG. Would xG predict Liverpool 7 Man Utd 0? Beside if players or teams are getting feedback on their xG they are likely to start changing their game to hit the number, but maybe not the back of the net (e.g. always shooting when they are directly in front of goal instead of passing to a team mate in a better position.)
I feel like sometimes xG is used differently, or incorrectly. However, if you collect all of Brentford's stats and put them together, you will see they correlate amazingly. Brentford are super close to their xG, xGa, and xPts, and are one of only 3 teams whose expected position matches their actual position. They have the best shot conversion percentage in the league, with one of the highest xG/chance (can't remember, I think behind City, maybe?), and this isn't accident, it's through design. The entire system is designed to increase the quality of chances we create and decrease the quality of chances we concede. It's why it seems like teams score a crazy amount of screamers against us (think Leicester last season with 3 screamers in 2 games). I would be very interested to see a video breakdown of Brentford's use of xG, xGa, etc., to mold playing style.
Those who have studied Statistics or use Statistics in their profession would not find this surprising. Even the best models need some additional context to go along with them in order for them to be useful. Great video as always.
XG does not include factors like external pressure on the player, his fatigue or the frenzy that a player had just gone through before shooting. For example: some years ago my club VfB Stuttgart had a high pressing coach, Alexander Zorniger, who - if I remember correctly - said that most goals were scored within 10 or 15 seconds and that this was the reason why his team should try to score as fast as possible when having regained the ball. The XG of Stuttgart was such that one would have expected a much better league position. Eventually the coach was sacked because of the poor results (and some anger issues he had). Now, some argued that the club should have sticked with him longer because he had been unlucky - by pointing to the XG. Yet in one article the author suggested that the XG was in fact misleading: Because the players were switching from high-press to attack and were drilled to immediately look for a chance to score, they generally were not calm enough to convert the chance. Or let us take another example: Let us suppose a team is playing counterattacking football and that a player often has to run 50 yards at full speed before taking his chance. The XG might be favorable but the physical exhaustion of the player might then have a negative impact on his actual scoring. Now, as long as this external factors only happen sometimes, they probably are not relevant but if - as the two examples suggest - they are a consequence or a part of the tactics, then the XG will be higher than one should in reality expect.
The primary problem that I have for xG being a representation of a team’s over-performance or underperformance is that it doesn’t consider the thought process of players choosing to take a shot or forego taking a shot. So it completely ignores how likely players will look to do other actions other than shooting. For example, let’s say a player receives the ball in a position where, if he chooses to shoot, he would have an xG of 0.1. Would he shoot? Well that depends on the player and the situation he finds himself in. If he finds a player, where, if he can make the pass, his teammate would have an xG of 0.25, then maybe he should be looking to pass rather than shoot, if the pass is not too difficult. Whether he actually makes the pass or shoots depends on the player, of course, but also depends on the tactics of the team. Some teams look to create more tap-in goals with high xG, while others may be more direct. Whether a team overperforms or underperforms their xG depends significantly on the tactics and coaching being employed by the team.
There's also the issue that xG is tracked over the whole season but points are shared on a game by game basis. So I can have an xG of +10 on match week 10, that'll push me way up the xG table but it still only accounts for three points. One place we shouldn't be expecting xG to level out in individual goals. When a player has higher or lower xG it probably means he's a better or worse finisher than the average. It's very unlikely that he's entirely average. Indeed, only 1%-ish of people should literally be exactly in the mean.
That's what "expected points" is designed to deal with, with regards to your first point. Sample sizes in football are very small (statistically speaking) but when you look at a player's xG Vs goals over a 4-5 year period, you often find they are very close. So while players have good and bad form, very few consistently overperform their xG.
Lower or higher xG can usually be explained by players form. Like Son or Rashford can go through a period when almost every shot is a goal from even low xG then go on a long run of shooting without scoring at all which overtime balances it back out. It’s only really poor finishers and the very best that consistently over perform or under perform their xG.
I understand xG at a basic level, but expected assists has always been a step too far in terms of dependencies for me. Maybe Tifo could make a video explaining this in Layman's terms?
Transition matrix of how often a pass from that location transitioned into a goal. It all depends on what variables you feed to your model. Do you take into account positional data as well? Or just event data location. Let’s you don’t use a regression function or more difficult model, you can just count the number of passes from that location and check how often they became a goal. Now you can start increasing the difficulty of this model by introducing extra variables.
I'm surprised that you did not mention Union Berlin. They are the epitome of xG not telling the whole story this season. They should be fighting for relegation, and yet, Champions League is looking very likely next season.
This is the problem with stats-based football discussions. The majority of the people using them either misuse them or don't understand the stats contextually.
Also worth noting that teams can over or underperform xG depending on who's taking the shots: a team whose forwards are poor at shooting will very likely underperform their xG and vice versa. Many teams that heavily overperform their xG do so by focusing many chances on a single high quality finisher, and that is a sustainable strategy.
@@andrasszabo1570 No different to any other player getting sold or injured. A team could have a striker worth +0.1 goals a game above their xG, but also they could have another player worth +0.1 goals a game, all of it described by xGD. Losing the former costs the team +0.1 goals a game without changing xGD, while losing the latter costs the exact same amount, the xGD just also drops. Clearly if a team sells their top striker and expects to keep outperforming xG they're making an error, but selling their best creative player and expecting xGD to stay the same would be identically foolish. Whether a player's value is captured by xGD or not doesn't change anything about how good the team is.
@@1882Stu Try and figure out how to discount is appropriately. Maybe on average one division underperforms xG by 20%, then just take xG and multiply it by 0.8. Requires work to figure out which discount to go with though. At the end of the day xG is just a model, and you can adjust it for your own purposes if you so desire.
Nice video. However being interested in statistics (albeit not professional) there are som problems. Of course XG in a league table is a dangerous metric - it has the same problem that goal scoring difference has - you might win a game 20-0 and because of that have a positive foal difference, but that doesn't mean you get more than 3 points. The biggest fundamental problem i have with Xg is how you accumulate several chances in quick succession. If you have have 4 shots on goal in quick succession from the 6-yard box, then you might get a combined number of xG of above 1 even though it is only one chance to score and should never be above 1(the team cannot score more than one goal on the opportunity as a whole). If you do it the other way and only look at one of the shots, then you underestimate the xG for the combined score - 4 shots does give a lot higher likelihood to score than one shots does. I haven't seen any statistics that accounts for this.
Stats is a tool. It is only good as the person using it. There are things that can be measured and things that can' t. Xg allow a more measurable look at those shots. It can' t ever be completely objective, but it can help. One thing to note, it can also likely not the metric, the top clubs are using. Each stat team in those club would protect their trade secrets they spend hours finding and used to win games. Xg is simply what accessible to media and pundits.
There are tables that also use expected points using the expected goals of each match to determine the result. Like you said just because Arsenal are outperforming their GD doesn't mean they would be outperforming their points in fact they could be performing better
Most of the xG models account for this situation by considering the total probability of scoring within that cluster of shots. e.g. a 0.6 xG shot quickly followed by a 0.4xG shot would be: prob of first shot going in + prob of first shot not going in * prob of second shot going in 0.6 + (1-0.6) * 0.4 = 0.76xG
Current issues with xG that can have a big effect in a game by game basis. 1: A player gets into a good position but doesn't shoot (i.e. tries to dribble keeper, whiffs the ball) and gets 0xG. 2: With VAR sometimes clear offsides are not stopped. If its not a goal game isn't always stopped. That is added to the xG even tho it would not have counted.
Glad you guys did this video. For years, I’ve been saying that xG is not a raw data that can be heavily relied on. It requires context. I feel like it’s been misused the last couple of years. Now I’m hearing people use xG to help make bets?!?! Insane.
Preach! The vast majority of the time when I see someone use xG to make an argument online the stat is woefully misused and the argument circumspect at best (and downright moronic at worst). The good ole eye test is usually better than vomiting up a single stat with no context.
It's the same with any stats You could use the league table to help place bets, but with that you could have a team who is bottom after 10 games, but within those have played the 8 strongest teams in the league. Likelihood is they will turn things around as they now have an easier run of games ahead of them.
@@JoshAston23 No it’s not the same for any stat. You clearly didn’t pay attention to the video. xG is a stat of context but more importantly it’s about chances created. What you are saying is more in line with average goals per game. That’s NOT the same as xG. By you using the League table (GF/GA) now you in the realm of Goals scored per game. That’s a direct and raw stat. Goals per game/goals against per game is an average of a complete set of games played. It’s not taking into account chances created. Again, xG is more a stat weighed heavy on the potential chances created than an average goals scored per game.
@@georgebrobbey6882 I understand what xG is, I'm just making the point that all stats require some context alongside them (some more-so than others). Many people just look at the numbers without drilling further into them.
Let's use this extreme scenario: A player takes a shot from a tight angle with a 0.1 xG. The keeper saves it but right onto the line, and the attacking team kick the rebound for a 0.95 xG. However, this rebound hits the post, lands back where it started and the player eventually scores the second 0.95 xG shot. This team had 2.00 xG over the one chance, however this team was never going to score more than 1 goal in this scenario. It makes them look like they should be scoring more than they actually do. Also, I don't believe xG takes into account the player taking the shot, or even their strong foot (but correct me if I'm wrong). If Messi takes a left footed shot from outside the box, the xG is the same as if Maguire took the exact same left footed shot from outside the box. If Messi then scores and Maguire doesn't, it makes it look like Man United are underperforming their xG while PSG are overperforming it, but clearly, your IRL perception of who is more likely to score is very different given that Messi is not only on his strong foot, but a far better scorer of goals than Maguire. The top teams are not outperforming their xG because of luck, they have good strikers who score more often than the average player but hit their expected values. Post-shot xG, or xGoT, is a better indicator of how many goals a team is expected to score in a game because it doesn't rely on the quality of the player taking the shot, only the shot itself, although it still doesn't factor in the GK ability or if a team has multiple rebounds on the same chance.
True but for the messi scenario ch was basically made to improve the quality of chances that's why many teams always look for tap ins and other chances they are likely to score
This video misses the mark a bit on the context that xStats are missing. Leeds aren’t simply better at creating chances when they are down 2 goals, they’re creating a higher volume of xG because they are trying to get back into a match against a side that is likely sitting back & setting up the counter-attack or seeing out a comfortable lead. Arsenal are consistently getting into favorable game states where opponents are chasing the match leading to efficient counter-attack opportunities. Leeds aren’t underperforming, they’re creating a lot of low xG shots at a high volume. Arsenal aren’t over preforming, they’re frequently in game states that lead to high xG shots at a lower volume. The metric being misused to measure their performance doesn’t account for game state.
The weirdest part with expected goals, is that you don't generate any xG if you don't shoot. Let's say there's a perfect low cross, keeper is beaten, and right before the #9 can tap it in the CB tackles it to put it just out of his reach. That's 0 xG. For me that's the biggest issue to address. xG give us information about the quality oh shot a team generates (or allow)which is really valuable. But goals shouldn't be expected only when a player shoots.
Or a 1-on-1 with the keeper, where the keeper smothers the ball is also 0 xG, even tho a it would have been maybe 0.5 xG if the striker had just touch the ball before the smother.
It levels out over time. E.g. a team like City might take fewer shots in certain positions because they prioritise high-quality shots, but over a large sample this will be addressed by the fact that they’ll have a lot more 0.8 xG types of chances.
One small things that's always bugged me with the xG model, is that it doesn't account for specific team or player statistics, for instance I'd imagine a top 5 striker (Lewandowski for example) would have a better chance of scoring any given shot than your average middle of the road 10 goal a season striker does. I think it matters even more from distance, I'm sure Steven Gerrard or Paul Scholes might do better from 25 yards than some players do from 18 yards. Positioning as well, some players really love certain types of chances, like Mbappe loves hitting it across goal when coming in from the left and also being right on the back post to pick up stray shots or deflections, but he'd probably do worse in the air than average, I think that could be an interesting stat to have alongside xG to show the difference certain players make, and showcase exactly how they exceed the norm
@mavis cruet his numbers aren't great, he could be a victim to the stats though because he's often double marked and has to do a lot of work to win them in the first place
You could try and figure out the average xG over/underperformance and simply add that to their xG to form xG-L for Lewandowski. If he overperforms on average by 15%, give him a 1.15 multiplier on whatever xG score he's assigned. Not perfect, but no model can be at least without professional tools.
@@lewisblackwiththenicehair I don't suppose he's that tall either, he just seems so athletic that I would have thought he could. FM give him a 7 for jumping reach and heading, so I guess his numbers aren't indeed great.
@mavis cruet no I believe he's only 5'10/5'11-ish, surprised FM haven't given him a little more jumping reach, it's more so just the quality of his heading that let's him down rather than not being able to reach chances
From Opta - “Expected goals (or xG) measures the quality of a chance by calculating the likelihood that it will be scored from a particular position on the pitch during a particular phase of play. This value is based on several factors from before the shot was taken. xG is measured on a scale between zero and one, where zero represents a chance that is impossible to score and one represents a chance that a player would be expected to score every single time.” So if a player takes a shot from halfway line they’ll generate lower xG vs. the same player taking shot in opposition box would generate xG closer to 1. If a team creates better goalscoring chances than the opposition, they’ll generate higher xG. Unless the team is Brighton who always win on xG regardless of the opposition /s.
I think xG is useful as a team metric, but woefully misused for individuals. I have no idea how big the sample size is, whether i's just top flight teams, or down to div 3, but even in the Premier League, I expect Haaland to score more chances that Woods. The idea that Haaland might "revert to the mean" because he's outperforming "his xG" is ridiculous, because he's not an average striker. And xG will include non strikers too, I imagine. Like with most data, a lot of journalists have no idea how to use it.
Everything reverts to the mean, but one should think a bit about which mean. In this case, Haaland probably has a certain "finishing ability" which is equal to the rate by which he outperforms his xGoals. He will probably revert to the mean of that value. But that doesn't mean he will revert to only scoring xGoals amount of goals.
@@lnhart7157 yeah. It's useful for pegging him, so to speak, against himself, but the whole point of an elite level striker is that he scores more than Neal Maupay.
I think as a team metric, xG can be misused as well, especially with a small sample size. For example, comparing the xGs of two opposing teams from a single match doesn't always depict an accurate story. There is always an element of randomness. If I see xG shown in the match stats on TV, I feel like it can be misleading to viewers who aren't aware that it needs context and that the stat is probably better used with a large sample size of data.
@@adamafriedland It actually gets worse with more data points in this context. xG is for measuring the value of an individual shot. At a team level for a full match, I’ve seen xG say that a team should have won handedly by 5-1, but the reality was a 2-2 where the team accumulating an xG of over 5 was the one lucky to have gotten the draw. Applied to multiple matches it tends to always say top teams are over-preforming and bottom teams are underperforming. It lacks consideration for game state.
@@adamafriedland with immaculate timing, I've just watched sky sports show stats about how Burnley have changed their style. It just showed 5 seasons of possession and long ball numbers. Totally lacking the context that they're no longer playing PL teams. Context is vital!
The big issue with xG is that each outlet has its own model, meaning different outlets will have a different xG for the same game. How do we know which one is more accurate?
I suppose people out there will constantly be testing the accuracy of each model by aggregating the average differences between xG and goals. But even then, you sort of have to make the assumption that there are as many teams over-performing as under-performing, so you can assume that overall teams should be expected to be close on average. I think the more you look into it, the more you realise that no matter what you do, it's always going to spit out a number that needs to be taken with a pinch of salt, but that doesn't stop it being useful as long as you've got a good understanding of the pitfalls
the more accurate ones are the ones that attach a players individual scoring chances to their own personal history. The average of all players scoring a pen is 80%, but some players have a 90-95% success rate, like jorginho. If they are using personal histories then jorginho would get .9-.95 instead of .8
One thing I don't understand/wonder about xG is, is the model continually updated? Let's take the 0.8 for penalties as an example. If over consecutive seasons players average better or worse than 80% for penalties, is a penalty's xG value then adjusted in the following season?
Most models are going to be updated periodically. But it requires a huge amount of data to be statistically significant. So the new data will be added into the mix with the old data and is unlikely to noticeably shift it.
I think it depends on what you're looking at, the xG model either refers to individuals (players, teams, leagues etc.) or the actual situation on the pitch Every penalty has the same rules, the ball is always in the same spot, no defenders but the goalie. This makes a baseline model, that every player that takes a penalty, they will in all scenarios at least have a 0.8 xG, or 80% chance of scoring. Now, whether a player actually converts those pens, that will determine what their individual xG is in comparison to the baseline. If they score each one, then they will be outperforming the baseline penalty xG, and if they're missing a lot of pens, they're xG will sit way lower. So you can refer to the stats to see if a player is a good penalty taker or not, because we're comparing actual chance across all penalties and if a player actually over or under performs by the expectation . The same can work to tell if a goallie is a good pen stopper or not, but it becomes a but difficult as saving a pen, is not the same as a pen being missed. I hope I'm making sense here, I think xG is a really cool concept, but it will never be able to be a true predictor of what's supposed to happen, and that adds to the unpredictability of football and the beauty of it.
The next most logical stat should be avg xG/ shot which will distil the actual quality of each shot and give a better understanding of how good a team's attack actually was.
Rather than just XG per match. I’d be interested in linking xG with possession stats. A way of knowing how productive possession is in creating chances
What about quality of striker. Players like Son usually over performed his xG (maybe not this season) because he was able to score very difficult low xG shots.
I think there were changes to how xG was calculated at the start of this (23/24) season for the EPL. For a few weeks, the xG was basically just the same as the actual goals. But when you watched the games, you could see that some goals were coming from outside chances and others were dead-certs and it wasn't really adding up. Seems to have been fixed around about game 5.
The only time I truly felt I understood xG was for Martinelli's goal against Aston Villa with an xG of 0.99. A guaranteed goal (empty net) barring a slip or being struck by lightning.
Something I’ve always wondered about with Xg is this scenario: a team is being battered and sat deep in the own half all game, in the 90th minute they counter attack and their striker gets a golden opportunity worth say 0.9Xg, but he misses and hits the post, only for it to rebound to his team mate who gets an equally good opportunity worth again 0.9Xg, would the overall Xg for the game reflect this miss or would they count both chances towards the overall Xg
Interesting question because technically it's 2 shots so an xG of 1.8 total but only 1 chance. This next part is me just using guesswork: Because you can't score 2 goals in one chance but you can have 2 shots, the xG can't be greater than 1, even if the sum of the two shots is above 1. Because you can't have a chance that is above 1.0xG, I'd say that you have the initial chance of 0.9xG and there's probably an algorithm that determines the percentage of the second shot that is added on. So the overall chance comes to like 0.9 + (0.9x0.1) = 0.99xG. These numbers are purely fictional ofc.
(If your model is bad, you just take the shot with the highest xG for the whole chance). But really, the chances are conditional, so you calculate the xG of a possession with multiple shots through working out the likilhood of there being no goal, then subtracting that from 1: prob(goal) = 1 - p(no goal) So possession with 4 shots: p(no goal) = (1 - xG1) * (1 - xG2) * (1 - xG3) * (1 - xG4) For your example: (1 - 0.9) * (1 - 0.9) = 0.01 = p(no goal) 1 - 0.01 = 0.99 = p(goal) I hope this is clear
@@Theme1412 yeah that makes sense on a team level, but would a player who misses the first chance still register 0.9xG? So over the course of a match the sum of all players xG could be greater than the teams overall xG for the match?
@@johnbennett7784 Well it all depends on the model, and who's doing the tracking, and why. If you bother to keep those (which you should). Because at the end of the day, they're two similar but different things you're measuring, and for different reasons You use the first to measure a teams performance, and the second to measure a player's. Stuff like League xG tables use the former, whereas tracking a player's positioning and finishing typically uses the latter. However you can use the former to track a player, also, for other reasons. Sorry, I'm running out of time and haven't explained myself right So like, if in your example from earlier, the same player had the two 0.9 shots, the team won't look better because it's a 0.99 chance. But the player will look worse, by scoring 1 goal from 1.8 xG. And yes, mass media and stuff will usually not make it exactly which they're using, but it's usually pretty straightforward to guess which they'd be using
@@Theme1412 thanks for explaining, that makes a lot of sense. It would probably be easier for people who aren’t as familiar with it to understand if there was a clear indicator of which is being used
I think there is much deeper analysis about xg that needs to be looked at. Why is a goal that is scored but ruled out for a marginal offside given as 0 xg? Surely a slight offside has a probability just as a shot which hits the post might go in on other occasions.
Do you have to shoot for it to count as an xg? For example if someone shoots but shot is blocked by defender does that count?, If a good cross comes in but st misses the ball in heading it? Or for example if a good sequence of attacking play happens and the pass into the box towards the striker is over hit, however if it were to be passed correctly he'd be in a 1v1 or open goal? Please answer. Thanks
Yeah its a major limitation of the model - only shots count towards xG. If a striker narrowly misses a tap in because they are just short of the ball, no xG is generated
When it was introduced, It used to considered before shooting but when players would sky the ball in front of an open net but they had to slide/dive to reach it (to name one of many such scenarios), it usually did not give an near accurate representation on how hard the chance was. So analysts started to split the metric into Pre Shot xG(which did not consider anything but the position of the attacker and the ball right before the ball was struck) and Post Shot xG (which considered the ball height, shot power, shot location, etc. But it got confusing to follow 2 metrics. So they kept it as one and used more data to refine the model by adding more layers. Based on your scenarios, defenders blocking in front of the shot is a low xG chance but if the defender sprints from behind to catch the attacker (1v1 with the keeper) and block it is considered a higher xG chance i.e the striker bottled it. as for crossing, xG is calculated if the attacker makes a good enough contact and then misses. it is not based on the quality of the cross as far as i know. what is good enough contact is left to individual xG models and as for striker running through on goal from a pass, until the striker has an ability to reach it and shoot only then it is considered. so weight of the pass is irrelevant at that time. but quality of the cross, cutbacks, through balls, long balls, etc is measured in another metric call xT (Expected Threat) this info may not be entirely accurate but i am sharing what i know
This is one of the limitations with shots-based models, which have become the standard. Worth mentioning that the reason they are standard is not because they are good but because they are easy, you can just look at each individual shot and assign it a value and then add them all together. There are plenty of people trying to make better models and I believe at least one is in use professionally, but will not be available to the casual viewer.
I think it would have been instructive to mention that the xg values are based on the average player. Arsenal over performing their xg but have incredible attacking players so this is expected, in a way. For almost a decade messi would over perform his xg by a considerable margin, but he was so average player. That context is important because the best teams almost always over perform their xg because they have above average players
No one ever talks about how big chances that doesn’t end up in shots don’t count in the XG statistics. Whereas if you just keep pulling the trigger even from very bad positions the number will eventually ramp up.
Controversial opinion: Xg killed the long shot. As it has long been said - perfect is* the enemy of good/done. Seeking the most perfectly executed move all the time can chain some of your team’s most magical/creative players to a zone or task that doesn’t allow them to show individuality in ways that could potentially be useful. I think one of the biggest things Xg misses is the particularities of the INDIVIDUAL whom the chance falls to. For instance, a Messi shot from 19-22 is better money for me than a LOT of players (many on big money, too!) from 12’-15’.
*I don’t think it is, but it CAN BE. For instance, Tiki-Taka is GREAT with Messi and Iniesta and other technically gifted players (mages? wizards?) - but without some consummate finishers you get….Spain in recent times.
There simply isn't enough data to accurately measure how good a player is at shooting. But it doesn't matter too much as few players veer far from their xG over their careers. Messi is one of those exceptions
I think it's inevitable, why risk conceding possession for a 30 yarder when you can keep the ball for another 5 minutes to try and work an unmissable tap in. Even though the reality is it probably never appears.
No single stat always tells the whole truth. It's only by looking at all of them in conjunction with each other do you get a better understanding of what actually happened on a football pitch
I think the main problem is that its not absolutely factual. two people could look at the same chance and determine different probabilities for scoring
xG is a largely misleading metric. There is no way to exactly calculate the quality of a chance. There are too many factors and they cannot be considered
What I've always wondered about xG is does it factor in who the player taking the shot is? For example for a chance on the left side of the box, would Haaland have a higher xG from that shot than Sterling for example?
No, it doesn't matter who takes the shot. The expected number is for the shot, not the shooter. That's why better players usually outperform xG over a long period of time, since it predicts what are the chances an average player would score from that situation.
xG is the probability that an average player would score in that situation What you're asking about is the probability that a single player would score in that situation. That is a similar metric that would require a compilation of the game statistics of a single player throughout their career, instead of the statistics of all players in a league. However this model won't be very refined since there's only so many games and so many shots a player takes during their career. Also, how they play changes a lot during a player's career so the average probability will be somewhat useless due to the high amount of variance in performance throughout the years
I get that xG is a good measure or the quality of a shot. The accumulation of them what confuses me. Adding up the probabilities of each shot as the xG is a score itself does not make much sense. A better reading of the stats would would be, eg; Team A had 5 instances of xG of 0.5, and Team B only had 2. Team B created many xGs that were below 0.5 i.e difficult chances, hence did not score and lost.
This video gets it wrong. If a team conceded fewer goals than their xG goals against, that doesn’t mean their defences overperformed, it just means the teams they played against had poor finishing. Also, aggregate xG for and xG against won’t always be a good predictor of league position, because that’s decided by points. You could lose a game 0-10 and win a game 1-0. Another team could lose a game 2-3 and lose another game 2-3. Your team would be higher in the league but your aggregate goal or xG performance would be worse.
In my football Manager game I saw a good example of the limits of XG we had 6 shots in quick succession from around 4 yards out that just kept getting blocked on the line before it was cleared after the 6th shot, the stats said my expected goals were 4 as a result but clearly this is flawed as you cant score 4 goals in one attack, it feels like Xg should only count the highest Xg shot per attack.
Are these videos targeted at casual fans who dont understand the game properly? As I feel like they have started to become extremely top level and very light in actual meaning or content
Disclaimer: Haven’t watched the video yet but I recently saw someone say that Gavin Bazunu of Southampton is a poor GK because he’s conceded 12 more than his “XG” that person didn’t factor the “XG” of their forwards which plays a big part (don’t score under more pressure) stats do tell a story to an extent but data is honestly only relevant to the person who interprets it.. apologies in advanced if this was mentioned in the vid
@@sportsjefe I hear you loud & clear I’m not debating that at all..but naturally if you don’t score you’re naturally put under more pressure/face more shots meaning there’s a higher chance you’ll concede more…my main point was that stats can’t be relied upon to paint a full picture
xG only provides clarity into the quality (and amount) of chances created, it doesn't take into account something like the quality of defending or transitions or anything else.
The people I hear using advanced metrics the most tend to watch the most football and are often involved in coaching. How many football games do you actually watch in a year?
Where are all those chaps who told their teachers they didn’t need to know the difference between a mean, a mode and a median. Seem like for Fulham the median xg difference might be a better statistic.
Most fans these days are statmen. If you ask them why you think x-team deserved to win against y-team they'll say it's because their xg was better. Stats shouldn't be the whole argument when it coming to discuss about anything in football, it should merely be used to back-up a point you're trying to make. I do think that only people that are part of the football club(staff) should be the only ones with access to stats because most use it wrongly and without any context applied.
XG is a mug's statistic that is leaned on heavily only by people who don't really understand the game. XG tells some information about a game, but in isolation does not tell the story of a football game. XG can flatter certain styles of football and misrepresent others. It doesn't represent all the subtleties of football both physical and mental, strategic etc.
Can xG models factor relative player ability and game state? For example, Haaland or Mbappe have a higher likelihood of scoring the same chance when compared to a League 2 striker. Likewise, a striker may be less likely to score a last minute 1 v 1 to win a game if we compared the same chance but this time the team is 4-0 up and cruising.
The models analyze all shots from players of varying abilities in the included leagues. This results in an xG value that represents an average of all players, from the top stars to the third choice strikers to center backs. Because all shots from a large number of matches are included in the data set, different game states will be represented. By comparing goals scored to xG, you can determine a player's finishing ability, assuming a sufficiently large sample size (Goals>xG = better finishing). Additionally, expected goals on target (xGoT) is an often-overlooked statistic that measures a shot's chance of scoring based on where the ball crosses or would cross the goal line. If a shot's xGoT value exceeds its xG value, it suggests that the shot placement improved the likelihood of scoring. TL;DR: xG is just an average.
Also, as always, data needs context. I'm not saying any of the stats I mentioned should be seen as the end all be all. They are just other ways to judge performance outside of the eye test or traditional stats.
I believe I’ve heard of some xStat models trying to account for the individual player taking the shot, like their historic ability to convert similar shots, but those aren’t very prevalently used . I’ve never heard of a model that takes into account game state, and I believe it’s one of the main reasons xStats always seem to say top teams are over-preforming and bottom teams are under-preforming. For example, Arsenal’s are often in a positive game, playing with a lead, so they can be more efficient with their counter-attack taking high xG shots at a lower volume of shots. Leeds are often in a negative game state forcing them to push forward against an opponent with a multi-goal lead creating a high volume of shots but with much lower xG values.
These days people care more about XG, shots on target, runs, etc more than scoring goals, winning matches and trophies. 😂😂 Liverpool fans can't stop crying about the UCL final.
Just like any prediction, XG is best used as a directional, not an absolute metric. No model could use past chance conversion stats of say, Rashford and predicts his performance this season!
you know ball
The models are not reality is why: They're dependent on the input data which means they're as blind about what is not put into them as any other predictive system. As good as Rashford has been this season, without an out-and-out striker eg Kane, I'd anticipate his numbers to go down again as ManU are worked out in future projection for example.
Idk, early on in the season Rashford was accumulating a lot of xG, he just wasn't converting them. It was reasonable to predict, based off of his previous season performances, that he was about to start scoring prolifically.
Even seeing it as directional is the gamblers fallacy, assuming there is a definitive predictble underlying pattern or trend that is likely to continue. XG is based on retrospective analysis. There is no certainty at all whether the score of the next game will be up or down on the previous xG. Would xG predict Liverpool 7 Man Utd 0? Beside if players or teams are getting feedback on their xG they are likely to start changing their game to hit the number, but maybe not the back of the net (e.g. always shooting when they are directly in front of goal instead of passing to a team mate in a better position.)
@commentarytalk1446 you saw future bruv 💀
I feel like sometimes xG is used differently, or incorrectly. However, if you collect all of Brentford's stats and put them together, you will see they correlate amazingly.
Brentford are super close to their xG, xGa, and xPts, and are one of only 3 teams whose expected position matches their actual position.
They have the best shot conversion percentage in the league, with one of the highest xG/chance (can't remember, I think behind City, maybe?), and this isn't accident, it's through design.
The entire system is designed to increase the quality of chances we create and decrease the quality of chances we concede. It's why it seems like teams score a crazy amount of screamers against us (think Leicester last season with 3 screamers in 2 games).
I would be very interested to see a video breakdown of Brentford's use of xG, xGa, etc., to mold playing style.
As good as that all is, take away Toney and they'll suffer in all those stats, however...
@@commentarytalk1446 agree without toney their play style would fall apart
@@commentarytalk1446 Brentford 3-1 Liverpool would like a word.
Lol Brentford have now fallen off a cliff.
Probably a bit late but where did you find these stats?
Those who have studied Statistics or use Statistics in their profession would not find this surprising. Even the best models need some additional context to go along with them in order for them to be useful. Great video as always.
As a statistician the xG model, its usage and assumptions fascinate me TBH. With many familiar issues to statistical models I use and design at work
XG does not include factors like external pressure on the player, his fatigue or the frenzy that a player had just gone through before shooting. For example: some years ago my club VfB Stuttgart had a high pressing coach, Alexander Zorniger, who - if I remember correctly - said that most goals were scored within 10 or 15 seconds and that this was the reason why his team should try to score as fast as possible when having regained the ball. The XG of Stuttgart was such that one would have expected a much better league position. Eventually the coach was sacked because of the poor results (and some anger issues he had). Now, some argued that the club should have sticked with him longer because he had been unlucky - by pointing to the XG. Yet in one article the author suggested that the XG was in fact misleading: Because the players were switching from high-press to attack and were drilled to immediately look for a chance to score, they generally were not calm enough to convert the chance. Or let us take another example: Let us suppose a team is playing counterattacking football and that a player often has to run 50 yards at full speed before taking his chance. The XG might be favorable but the physical exhaustion of the player might then have a negative impact on his actual scoring. Now, as long as this external factors only happen sometimes, they probably are not relevant but if - as the two examples suggest - they are a consequence or a part of the tactics, then the XG will be higher than one should in reality expect.
The primary problem that I have for xG being a representation of a team’s over-performance or underperformance is that it doesn’t consider the thought process of players choosing to take a shot or forego taking a shot. So it completely ignores how likely players will look to do other actions other than shooting. For example, let’s say a player receives the ball in a position where, if he chooses to shoot, he would have an xG of 0.1. Would he shoot? Well that depends on the player and the situation he finds himself in. If he finds a player, where, if he can make the pass, his teammate would have an xG of 0.25, then maybe he should be looking to pass rather than shoot, if the pass is not too difficult. Whether he actually makes the pass or shoots depends on the player, of course, but also depends on the tactics of the team. Some teams look to create more tap-in goals with high xG, while others may be more direct. Whether a team overperforms or underperforms their xG depends significantly on the tactics and coaching being employed by the team.
The sound mixing and editing is incredibly good on this channel. Sounds amazing with the background music and I don’t usually like background music
Exactly. It really locks you into the video and has a very soothing effect. Fantastic work.
There's also the issue that xG is tracked over the whole season but points are shared on a game by game basis. So I can have an xG of +10 on match week 10, that'll push me way up the xG table but it still only accounts for three points. One place we shouldn't be expecting xG to level out in individual goals. When a player has higher or lower xG it probably means he's a better or worse finisher than the average. It's very unlikely that he's entirely average. Indeed, only 1%-ish of people should literally be exactly in the mean.
That's what "expected points" is designed to deal with, with regards to your first point.
Sample sizes in football are very small (statistically speaking) but when you look at a player's xG Vs goals over a 4-5 year period, you often find they are very close. So while players have good and bad form, very few consistently overperform their xG.
Lower or higher xG can usually be explained by players form. Like Son or Rashford can go through a period when almost every shot is a goal from even low xG then go on a long run of shooting without scoring at all which overtime balances it back out. It’s only really poor finishers and the very best that consistently over perform or under perform their xG.
Thank you I never understood xG. That's a stat I see brought up a lot without knowing what exactly it meant
It’s a very good stat for Football, but like any stat, this video just shows it’s can be skewed and that numbers are not always perfect
Your comment looks fake
Moneyball steps in.
I understand xG at a basic level, but expected assists has always been a step too far in terms of dependencies for me. Maybe Tifo could make a video explaining this in Layman's terms?
Transition matrix of how often a pass from that location transitioned into a goal. It all depends on what variables you feed to your model. Do you take into account positional data as well? Or just event data location. Let’s you don’t use a regression function or more difficult model, you can just count the number of passes from that location and check how often they became a goal. Now you can start increasing the difficulty of this model by introducing extra variables.
You aren’t ready for expected threat 😂
I'm surprised that you did not mention Union Berlin. They are the epitome of xG not telling the whole story this season. They should be fighting for relegation, and yet, Champions League is looking very likely next season.
This is the problem with stats-based football discussions. The majority of the people using them either misuse them or don't understand the stats contextually.
Also worth noting that teams can over or underperform xG depending on who's taking the shots: a team whose forwards are poor at shooting will very likely underperform their xG and vice versa. Many teams that heavily overperform their xG do so by focusing many chances on a single high quality finisher, and that is a sustainable strategy.
Until that single high quality finisher gets injured or sold...
@@andrasszabo1570 No different to any other player getting sold or injured. A team could have a striker worth +0.1 goals a game above their xG, but also they could have another player worth +0.1 goals a game, all of it described by xGD. Losing the former costs the team +0.1 goals a game without changing xGD, while losing the latter costs the exact same amount, the xGD just also drops. Clearly if a team sells their top striker and expects to keep outperforming xG they're making an error, but selling their best creative player and expecting xGD to stay the same would be identically foolish. Whether a player's value is captured by xGD or not doesn't change anything about how good the team is.
@@1882Stu Try and figure out how to discount is appropriately. Maybe on average one division underperforms xG by 20%, then just take xG and multiply it by 0.8. Requires work to figure out which discount to go with though. At the end of the day xG is just a model, and you can adjust it for your own purposes if you so desire.
@@1882Stu No, it's football statistics
@@1882Stu No, it's statistics. How meaningful it is, that's another question. It just depends on how and what you use it for.
Nice video. However being interested in statistics (albeit not professional) there are som problems. Of course XG in a league table is a dangerous metric - it has the same problem that goal scoring difference has - you might win a game 20-0 and because of that have a positive foal difference, but that doesn't mean you get more than 3 points.
The biggest fundamental problem i have with Xg is how you accumulate several chances in quick succession. If you have have 4 shots on goal in quick succession from the 6-yard box, then you might get a combined number of xG of above 1 even though it is only one chance to score and should never be above 1(the team cannot score more than one goal on the opportunity as a whole). If you do it the other way and only look at one of the shots, then you underestimate the xG for the combined score - 4 shots does give a lot higher likelihood to score than one shots does.
I haven't seen any statistics that accounts for this.
I would be surprised if no one does this, the maths is pretty easy if you make a couple of basic assumptions so it would be quite negligent not to.
I agree. But i have not seen it uden regularly - atleast i think
Stats is a tool. It is only good as the person using it. There are things that can be measured and things that can' t. Xg allow a more measurable look at those shots. It can' t ever be completely objective, but it can help. One thing to note, it can also likely not the metric, the top clubs are using. Each stat team in those club would protect their trade secrets they spend hours finding and used to win games. Xg is simply what accessible to media and pundits.
There are tables that also use expected points using the expected goals of each match to determine the result. Like you said just because Arsenal are outperforming their GD doesn't mean they would be outperforming their points in fact they could be performing better
Most of the xG models account for this situation by considering the total probability of scoring within that cluster of shots. e.g. a 0.6 xG shot quickly followed by a 0.4xG shot would be: prob of first shot going in + prob of first shot not going in * prob of second shot going in 0.6 + (1-0.6) * 0.4 = 0.76xG
Current issues with xG that can have a big effect in a game by game basis.
1: A player gets into a good position but doesn't shoot (i.e. tries to dribble keeper, whiffs the ball) and gets 0xG.
2: With VAR sometimes clear offsides are not stopped. If its not a goal game isn't always stopped. That is added to the xG even tho it would not have counted.
Glad you guys did this video. For years, I’ve been saying that xG is not a raw data that can be heavily relied on. It requires context. I feel like it’s been misused the last couple of years. Now I’m hearing people use xG to help make bets?!?! Insane.
Preach! The vast majority of the time when I see someone use xG to make an argument online the stat is woefully misused and the argument circumspect at best (and downright moronic at worst). The good ole eye test is usually better than vomiting up a single stat with no context.
It's the same with any stats
You could use the league table to help place bets, but with that you could have a team who is bottom after 10 games, but within those have played the 8 strongest teams in the league.
Likelihood is they will turn things around as they now have an easier run of games ahead of them.
@@JoshAston23 No it’s not the same for any stat. You clearly didn’t pay attention to the video. xG is a stat of context but more importantly it’s about chances created.
What you are saying is more in line with average goals per game. That’s NOT the same as xG. By you using the League table (GF/GA) now you in the realm of Goals scored per game. That’s a direct and raw stat. Goals per game/goals against per game is an average of a complete set of games played. It’s not taking into account chances created. Again, xG is more a stat weighed heavy on the potential chances created than an average goals scored per game.
@@georgebrobbey6882 I understand what xG is, I'm just making the point that all stats require some context alongside them (some more-so than others). Many people just look at the numbers without drilling further into them.
@@JoshAston23 Yes you’re definitely right on that. I get what you’re saying.
Let's use this extreme scenario:
A player takes a shot from a tight angle with a 0.1 xG. The keeper saves it but right onto the line, and the attacking team kick the rebound for a 0.95 xG. However, this rebound hits the post, lands back where it started and the player eventually scores the second 0.95 xG shot. This team had 2.00 xG over the one chance, however this team was never going to score more than 1 goal in this scenario. It makes them look like they should be scoring more than they actually do.
Also, I don't believe xG takes into account the player taking the shot, or even their strong foot (but correct me if I'm wrong). If Messi takes a left footed shot from outside the box, the xG is the same as if Maguire took the exact same left footed shot from outside the box. If Messi then scores and Maguire doesn't, it makes it look like Man United are underperforming their xG while PSG are overperforming it, but clearly, your IRL perception of who is more likely to score is very different given that Messi is not only on his strong foot, but a far better scorer of goals than Maguire. The top teams are not outperforming their xG because of luck, they have good strikers who score more often than the average player but hit their expected values. Post-shot xG, or xGoT, is a better indicator of how many goals a team is expected to score in a game because it doesn't rely on the quality of the player taking the shot, only the shot itself, although it still doesn't factor in the GK ability or if a team has multiple rebounds on the same chance.
total xG for your scenario is 0.9975
True but for the messi scenario ch was basically made to improve the quality of chances that's why many teams always look for tap ins and other chances they are likely to score
This video misses the mark a bit on the context that xStats are missing. Leeds aren’t simply better at creating chances when they are down 2 goals, they’re creating a higher volume of xG because they are trying to get back into a match against a side that is likely sitting back & setting up the counter-attack or seeing out a comfortable lead. Arsenal are consistently getting into favorable game states where opponents are chasing the match leading to efficient counter-attack opportunities. Leeds aren’t underperforming, they’re creating a lot of low xG shots at a high volume. Arsenal aren’t over preforming, they’re frequently in game states that lead to high xG shots at a lower volume. The metric being misused to measure their performance doesn’t account for game state.
Stats are stats. They don’t tell the whole picture.
A wise man once said "statistic is like a bikini model, it shows a lot of things but not everything".
The rest is all imagination. Still talking about both lol.
I'm fairly certain Morata has missed some shots with an XG of 1.0
The weirdest part with expected goals, is that you don't generate any xG if you don't shoot.
Let's say there's a perfect low cross, keeper is beaten, and right before the #9 can tap it in the CB tackles it to put it just out of his reach. That's 0 xG.
For me that's the biggest issue to address. xG give us information about the quality oh shot a team generates (or allow)which is really valuable. But goals shouldn't be expected only when a player shoots.
Or a 1-on-1 with the keeper, where the keeper smothers the ball is also 0 xG, even tho a it would have been maybe 0.5 xG if the striker had just touch the ball before the smother.
It levels out over time. E.g. a team like City might take fewer shots in certain positions because they prioritise high-quality shots, but over a large sample this will be addressed by the fact that they’ll have a lot more 0.8 xG types of chances.
That is an issue with the commonly used model, but theoretically it would be possible to extend xGoals to all situations, not just shots.
One small things that's always bugged me with the xG model, is that it doesn't account for specific team or player statistics, for instance I'd imagine a top 5 striker (Lewandowski for example) would have a better chance of scoring any given shot than your average middle of the road 10 goal a season striker does. I think it matters even more from distance, I'm sure Steven Gerrard or Paul Scholes might do better from 25 yards than some players do from 18 yards. Positioning as well, some players really love certain types of chances, like Mbappe loves hitting it across goal when coming in from the left and also being right on the back post to pick up stray shots or deflections, but he'd probably do worse in the air than average, I think that could be an interesting stat to have alongside xG to show the difference certain players make, and showcase exactly how they exceed the norm
Is Mpappe not good at headers? He's got such a hefty neck I assumed he would be.
@mavis cruet his numbers aren't great, he could be a victim to the stats though because he's often double marked and has to do a lot of work to win them in the first place
You could try and figure out the average xG over/underperformance and simply add that to their xG to form xG-L for Lewandowski. If he overperforms on average by 15%, give him a 1.15 multiplier on whatever xG score he's assigned. Not perfect, but no model can be at least without professional tools.
@@lewisblackwiththenicehair I don't suppose he's that tall either, he just seems so athletic that I would have thought he could. FM give him a 7 for jumping reach and heading, so I guess his numbers aren't indeed great.
@mavis cruet no I believe he's only 5'10/5'11-ish, surprised FM haven't given him a little more jumping reach, it's more so just the quality of his heading that let's him down rather than not being able to reach chances
From Opta -
“Expected goals (or xG) measures the quality of a chance by calculating the likelihood that it will be scored from a particular position on the pitch during a particular phase of play. This value is based on several factors from before the shot was taken. xG is measured on a scale between zero and one, where zero represents a chance that is impossible to score and one represents a chance that a player would be expected to score every single time.”
So if a player takes a shot from halfway line they’ll generate lower xG vs. the same player taking shot in opposition box would generate xG closer to 1. If a team creates better goalscoring chances than the opposition, they’ll generate higher xG.
Unless the team is Brighton who always win on xG regardless of the opposition /s.
I think xG is useful as a team metric, but woefully misused for individuals. I have no idea how big the sample size is, whether i's just top flight teams, or down to div 3, but even in the Premier League, I expect Haaland to score more chances that Woods. The idea that Haaland might "revert to the mean" because he's outperforming "his xG" is ridiculous, because he's not an average striker. And xG will include non strikers too, I imagine. Like with most data, a lot of journalists have no idea how to use it.
Everything reverts to the mean, but one should think a bit about which mean. In this case, Haaland probably has a certain "finishing ability" which is equal to the rate by which he outperforms his xGoals. He will probably revert to the mean of that value. But that doesn't mean he will revert to only scoring xGoals amount of goals.
@@lnhart7157 yeah. It's useful for pegging him, so to speak, against himself, but the whole point of an elite level striker is that he scores more than Neal Maupay.
I think as a team metric, xG can be misused as well, especially with a small sample size. For example, comparing the xGs of two opposing teams from a single match doesn't always depict an accurate story. There is always an element of randomness. If I see xG shown in the match stats on TV, I feel like it can be misleading to viewers who aren't aware that it needs context and that the stat is probably better used with a large sample size of data.
@@adamafriedland It actually gets worse with more data points in this context. xG is for measuring the value of an individual shot. At a team level for a full match, I’ve seen xG say that a team should have won handedly by 5-1, but the reality was a 2-2 where the team accumulating an xG of over 5 was the one lucky to have gotten the draw. Applied to multiple matches it tends to always say top teams are over-preforming and bottom teams are underperforming. It lacks consideration for game state.
@@adamafriedland with immaculate timing, I've just watched sky sports show stats about how Burnley have changed their style. It just showed 5 seasons of possession and long ball numbers. Totally lacking the context that they're no longer playing PL teams. Context is vital!
How are the xG's calculated before the season? On what stats are those expectations based?
Leeds are like that guy that shouts "hit me" at the start of a fight
The big issue with xG is that each outlet has its own model, meaning different outlets will have a different xG for the same game. How do we know which one is more accurate?
learn statistics/data science 😅
I suppose people out there will constantly be testing the accuracy of each model by aggregating the average differences between xG and goals. But even then, you sort of have to make the assumption that there are as many teams over-performing as under-performing, so you can assume that overall teams should be expected to be close on average. I think the more you look into it, the more you realise that no matter what you do, it's always going to spit out a number that needs to be taken with a pinch of salt, but that doesn't stop it being useful as long as you've got a good understanding of the pitfalls
the more accurate ones are the ones that attach a players individual scoring chances to their own personal history. The average of all players scoring a pen is 80%, but some players have a 90-95% success rate, like jorginho. If they are using personal histories then jorginho would get .9-.95 instead of .8
@@T.E.S.S. that just makes you not want to consume 99% of football media that uses stats wrong
@@patrickkataphasis1967 what would that show you if a team was underperforming in a given match, just that they were unlucky that day?
Great Thank You for This Educational Work! Appreciate it.
Excellent video. One of the best in here in a while
One thing I don't understand/wonder about xG is, is the model continually updated? Let's take the 0.8 for penalties as an example. If over consecutive seasons players average better or worse than 80% for penalties, is a penalty's xG value then adjusted in the following season?
Most models are going to be updated periodically. But it requires a huge amount of data to be statistically significant. So the new data will be added into the mix with the old data and is unlikely to noticeably shift it.
I think it depends on what you're looking at, the xG model either refers to individuals (players, teams, leagues etc.) or the actual situation on the pitch
Every penalty has the same rules, the ball is always in the same spot, no defenders but the goalie. This makes a baseline model, that every player that takes a penalty, they will in all scenarios at least have a 0.8 xG, or 80% chance of scoring.
Now, whether a player actually converts those pens, that will determine what their individual xG is in comparison to the baseline. If they score each one, then they will be outperforming the baseline penalty xG, and if they're missing a lot of pens, they're xG will sit way lower.
So you can refer to the stats to see if a player is a good penalty taker or not, because we're comparing actual chance across all penalties and if a player actually over or under performs by the expectation . The same can work to tell if a goallie is a good pen stopper or not, but it becomes a but difficult as saving a pen, is not the same as a pen being missed.
I hope I'm making sense here, I think xG is a really cool concept, but it will never be able to be a true predictor of what's supposed to happen, and that adds to the unpredictability of football and the beauty of it.
The next most logical stat should be avg xG/ shot which will distil the actual quality of each shot and give a better understanding of how good a team's attack actually was.
everybody claims to be the first view rather than digesting the video
Just smooth brain things
Rest
The xC (expected clout) of commenting "first" is 0
@James-MichaelRobinsonMore than the people who put "first" 😂🤫
6th
Rather than just XG per match. I’d be interested in linking xG with possession stats. A way of knowing how productive possession is in creating chances
What about quality of striker. Players like Son usually over performed his xG (maybe not this season) because he was able to score very difficult low xG shots.
I think there were changes to how xG was calculated at the start of this (23/24) season for the EPL. For a few weeks, the xG was basically just the same as the actual goals. But when you watched the games, you could see that some goals were coming from outside chances and others were dead-certs and it wasn't really adding up. Seems to have been fixed around about game 5.
Thank you for explaining this.
If a team shots of XG .3 and above are off target yet XG .01 to XG .10 on target but not scoring that is worrying.
The only time I truly felt I understood xG was for Martinelli's goal against Aston Villa with an xG of 0.99. A guaranteed goal (empty net) barring a slip or being struck by lightning.
Bro that stat doesn't tell you absolutely anything about the match
XG is part of stats used out of contexts
there's also xGOT which is really important but seems to always be ignored.
Like any metric, it's a starting point. If someone is using it in an absolutist, conversation ending way it's pretty safe to ignore that person.
Its videos like this that I love this channel always going deepee than just face value
Something I’ve always wondered about with Xg is this scenario: a team is being battered and sat deep in the own half all game, in the 90th minute they counter attack and their striker gets a golden opportunity worth say 0.9Xg, but he misses and hits the post, only for it to rebound to his team mate who gets an equally good opportunity worth again 0.9Xg, would the overall Xg for the game reflect this miss or would they count both chances towards the overall Xg
Interesting question because technically it's 2 shots so an xG of 1.8 total but only 1 chance. This next part is me just using guesswork:
Because you can't score 2 goals in one chance but you can have 2 shots, the xG can't be greater than 1, even if the sum of the two shots is above 1. Because you can't have a chance that is above 1.0xG, I'd say that you have the initial chance of 0.9xG and there's probably an algorithm that determines the percentage of the second shot that is added on. So the overall chance comes to like 0.9 + (0.9x0.1) = 0.99xG. These numbers are purely fictional ofc.
(If your model is bad, you just take the shot with the highest xG for the whole chance).
But really, the chances are conditional, so you calculate the xG of a possession with multiple shots through working out the likilhood of there being no goal, then subtracting that from 1:
prob(goal) = 1 - p(no goal)
So possession with 4 shots:
p(no goal) = (1 - xG1) * (1 - xG2) * (1 - xG3) * (1 - xG4)
For your example: (1 - 0.9) * (1 - 0.9) = 0.01 = p(no goal)
1 - 0.01 = 0.99 = p(goal)
I hope this is clear
@@Theme1412 yeah that makes sense on a team level, but would a player who misses the first chance still register 0.9xG? So over the course of a match the sum of all players xG could be greater than the teams overall xG for the match?
@@johnbennett7784 Well it all depends on the model, and who's doing the tracking, and why. If you bother to keep those (which you should). Because at the end of the day, they're two similar but different things you're measuring, and for different reasons
You use the first to measure a teams performance, and the second to measure a player's. Stuff like League xG tables use the former, whereas tracking a player's positioning and finishing typically uses the latter. However you can use the former to track a player, also, for other reasons. Sorry, I'm running out of time and haven't explained myself right
So like, if in your example from earlier, the same player had the two 0.9 shots, the team won't look better because it's a 0.99 chance. But the player will look worse, by scoring 1 goal from 1.8 xG.
And yes, mass media and stuff will usually not make it exactly which they're using, but it's usually pretty straightforward to guess which they'd be using
@@Theme1412 thanks for explaining, that makes a lot of sense. It would probably be easier for people who aren’t as familiar with it to understand if there was a clear indicator of which is being used
I think there is much deeper analysis about xg that needs to be looked at. Why is a goal that is scored but ruled out for a marginal offside given as 0 xg? Surely a slight offside has a probability just as a shot which hits the post might go in on other occasions.
all i know is that when my xG on FM is high i am still v unlikely to score it seems
You need better strikers
@@Cos_Why_Not i also need a joint and some gloppy
@@Cos_Why_Not @speedo270 might also need better quality chances in general.
Do you have to shoot for it to count as an xg? For example if someone shoots but shot is blocked by defender does that count?, If a good cross comes in but st misses the ball in heading it?
Or for example if a good sequence of attacking play happens and the pass into the box towards the striker is over hit, however if it were to be passed correctly he'd be in a 1v1 or open goal?
Please answer.
Thanks
Yes, you have to shoot to count as xG. Yes, blocked shots have an associated xG.
Yeah its a major limitation of the model - only shots count towards xG. If a striker narrowly misses a tap in because they are just short of the ball, no xG is generated
When it was introduced, It used to considered before shooting but when players would sky the ball in front of an open net but they had to slide/dive to reach it (to name one of many such scenarios), it usually did not give an near accurate representation on how hard the chance was. So analysts started to split the metric into Pre Shot xG(which did not consider anything but the position of the attacker and the ball right before the ball was struck) and Post Shot xG (which considered the ball height, shot power, shot location, etc. But it got confusing to follow 2 metrics. So they kept it as one and used more data to refine the model by adding more layers.
Based on your scenarios, defenders blocking in front of the shot is a low xG chance but if the defender sprints from behind to catch the attacker (1v1 with the keeper) and block it is considered a higher xG chance i.e the striker bottled it.
as for crossing, xG is calculated if the attacker makes a good enough contact and then misses. it is not based on the quality of the cross as far as i know. what is good enough contact is left to individual xG models
and as for striker running through on goal from a pass, until the striker has an ability to reach it and shoot only then it is considered. so weight of the pass is irrelevant at that time.
but quality of the cross, cutbacks, through balls, long balls, etc is measured in another metric call xT (Expected Threat)
this info may not be entirely accurate but i am sharing what i know
This is one of the limitations with shots-based models, which have become the standard.
Worth mentioning that the reason they are standard is not because they are good but because they are easy, you can just look at each individual shot and assign it a value and then add them all together.
There are plenty of people trying to make better models and I believe at least one is in use professionally, but will not be available to the casual viewer.
@@OneGoodCrusader thanks bro. Helped alot
To anyone wondering, the xG is like calculus, it never goes to 0, ever the craziest goal has a xG of 0.00000001 0 truly is impossible
Today I stop pretending I know what xG means. 🍻
I think it would have been instructive to mention that the xg values are based on the average player. Arsenal over performing their xg but have incredible attacking players so this is expected, in a way. For almost a decade messi would over perform his xg by a considerable margin, but he was so average player. That context is important because the best teams almost always over perform their xg because they have above average players
No one ever talks about how big chances that doesn’t end up in shots don’t count in the XG statistics. Whereas if you just keep pulling the trigger even from very bad positions the number will eventually ramp up.
Short answer: No
Controversial opinion: Xg killed the long shot.
As it has long been said - perfect is* the enemy of good/done.
Seeking the most perfectly executed move all the time can chain some of your team’s most magical/creative players to a zone or task that doesn’t allow them to show individuality in ways that could potentially be useful. I think one of the biggest things Xg misses is the particularities of the INDIVIDUAL whom the chance falls to. For instance, a Messi shot from 19-22 is better money for me than a LOT of players (many on big money, too!) from 12’-15’.
*I don’t think it is, but it CAN BE.
For instance, Tiki-Taka is GREAT with Messi and Iniesta and other technically gifted players (mages? wizards?) - but without some consummate finishers you get….Spain in recent times.
There simply isn't enough data to accurately measure how good a player is at shooting. But it doesn't matter too much as few players veer far from their xG over their careers. Messi is one of those exceptions
professional players don't care about kids talking about xG on tiktok and twitter. they shoot where they can shoot from, it's definitely not dead lol
I think it's inevitable, why risk conceding possession for a 30 yarder when you can keep the ball for another 5 minutes to try and work an unmissable tap in. Even though the reality is it probably never appears.
Good video, but it needed more explanation, shallow
No single stat always tells the whole truth. It's only by looking at all of them in conjunction with each other do you get a better understanding of what actually happened on a football pitch
Now I would really like to know the xG of "...and Smith must score!"
xG can never measure the quality or the scenario the chance is set in
I think the main problem is that its not absolutely factual. two people could look at the same chance and determine different probabilities for scoring
People gambling their life away on XG 😂
Penalty = 0.8xG = Harry Maguire with Accuracy~
xG is a largely misleading metric. There is no way to exactly calculate the quality of a chance. There are too many factors and they cannot be considered
Superbly presented, thanks.
What I've always wondered about xG is does it factor in who the player taking the shot is? For example for a chance on the left side of the box, would Haaland have a higher xG from that shot than Sterling for example?
No, it doesn't matter who takes the shot. The expected number is for the shot, not the shooter.
That's why better players usually outperform xG over a long period of time, since it predicts what are the chances an average player would score from that situation.
xG is the probability that an average player would score in that situation
What you're asking about is the probability that a single player would score in that situation. That is a similar metric that would require a compilation of the game statistics of a single player throughout their career, instead of the statistics of all players in a league.
However this model won't be very refined since there's only so many games and so many shots a player takes during their career.
Also, how they play changes a lot during a player's career so the average probability will be somewhat useless due to the high amount of variance in performance throughout the years
Our xG is badly being underperformed 😢
I get that xG is a good measure or the quality of a shot. The accumulation of them what confuses me. Adding up the probabilities of each shot as the xG is a score itself does not make much sense. A better reading of the stats would would be, eg; Team A had 5 instances of xG of 0.5, and Team B only had 2. Team B created many xGs that were below 0.5 i.e difficult chances, hence did not score and lost.
This video gets it wrong. If a team conceded fewer goals than their xG goals against, that doesn’t mean their defences overperformed, it just means the teams they played against had poor finishing.
Also, aggregate xG for and xG against won’t always be a good predictor of league position, because that’s decided by points. You could lose a game 0-10 and win a game 1-0. Another team could lose a game 2-3 and lose another game 2-3. Your team would be higher in the league but your aggregate goal or xG performance would be worse.
In my football Manager game I saw a good example of the limits of XG we had 6 shots in quick succession from around 4 yards out that just kept getting blocked on the line before it was cleared after the 6th shot, the stats said my expected goals were 4 as a result but clearly this is flawed as you cant score 4 goals in one attack, it feels like Xg should only count the highest Xg shot per attack.
The graphics remind me of jumpers for goalposts
Are these videos targeted at casual fans who dont understand the game properly? As I feel like they have started to become extremely top level and very light in actual meaning or content
People question xG but not all the metrics derived from xG.
Disclaimer: Haven’t watched the video yet but I recently saw someone say that Gavin Bazunu of Southampton is a poor GK because he’s conceded 12 more than his “XG” that person didn’t factor the “XG” of their forwards which plays a big part (don’t score under more pressure) stats do tell a story to an extent but data is honestly only relevant to the person who interprets it.. apologies in advanced if this was mentioned in the vid
A goalkeeper's xG performance has nothing to do with the forwards on his team, it simply means that he isn't stopping shots that he should.
@@sportsjefe I hear you loud & clear I’m not debating that at all..but naturally if you don’t score you’re naturally put under more pressure/face more shots meaning there’s a higher chance you’ll concede more…my main point was that stats can’t be relied upon to paint a full picture
Didn't XG predict that Liverpool should have finished behind City in 19/20, Liverpool were clearly the best team in the league.
xG only provides clarity into the quality (and amount) of chances created, it doesn't take into account something like the quality of defending or transitions or anything else.
They should show this to the xg merchant that they used to have on the show
They did this on Tifo Irl
1:33 That's bruno Fernandes 😜
Stats help, but w/o context it useless.
Football Twitter is in shambles right now
It doesn't
Do Aston villas rise from Steven G's 17th to pushing for a euro spot
Has Duncan seen this video yet 😊
people should watch full games than simply look at stats like xg
just say I DONT UNDERSTAND THIS VIDEO lmao
Also if you create a chance but the striker doesn't get the shot off... 0xG.
Jon "underlying numbers" "Goblin King" Mackenzie has just been fired.
xG is a stat mostly pushed by people who don't actually watch matches and betting addicts.
The people I hear using advanced metrics the most tend to watch the most football and are often involved in coaching. How many football games do you actually watch in a year?
xG is the equivalent of having the most amount of damage in a game of OW, means borderline nothing.
I've always maintained its a terrible stat
short answer, no
long answer, watch this video
Should it be EG? 🤔🤔🤔
Where are all those chaps who told their teachers they didn’t need to know the difference between a mean, a mode and a median. Seem like for Fulham the median xg difference might be a better statistic.
As soon as I heard 'game state' I knew who wrote this one
XG WE CERTIFIED
All these comment about the Athletic Story got me vexed. Just enjoy the video and ignore the drama … sheesh 🙄
Most fans these days are statmen. If you ask them why you think x-team deserved to win against y-team they'll say it's because their xg was better. Stats shouldn't be the whole argument when it coming to discuss about anything in football, it should merely be used to back-up a point you're trying to make. I do think that only people that are part of the football club(staff) should be the only ones with access to stats because most use it wrongly and without any context applied.
An idiot with stats is more likely to draw the correct conclusion than an idiot with "feelings".
But at the end of the day, they're still an idiot.
XG is a mug's statistic that is leaned on heavily only by people who don't really understand the game. XG tells some information about a game, but in isolation does not tell the story of a football game. XG can flatter certain styles of football and misrepresent others. It doesn't represent all the subtleties of football both physical and mental, strategic etc.
Can xG models factor relative player ability and game state? For example, Haaland or Mbappe have a higher likelihood of scoring the same chance when compared to a League 2 striker. Likewise, a striker may be less likely to score a last minute 1 v 1 to win a game if we compared the same chance but this time the team is 4-0 up and cruising.
The models analyze all shots from players of varying abilities in the included leagues. This results in an xG value that represents an average of all players, from the top stars to the third choice strikers to center backs. Because all shots from a large number of matches are included in the data set, different game states will be represented. By comparing goals scored to xG, you can determine a player's finishing ability, assuming a sufficiently large sample size (Goals>xG = better finishing). Additionally, expected goals on target (xGoT) is an often-overlooked statistic that measures a shot's chance of scoring based on where the ball crosses or would cross the goal line. If a shot's xGoT value exceeds its xG value, it suggests that the shot placement improved the likelihood of scoring.
TL;DR: xG is just an average.
@@adamafriedland appreciate it! Makes sense
Also, as always, data needs context. I'm not saying any of the stats I mentioned should be seen as the end all be all. They are just other ways to judge performance outside of the eye test or traditional stats.
I believe I’ve heard of some xStat models trying to account for the individual player taking the shot, like their historic ability to convert similar shots, but those aren’t very prevalently used . I’ve never heard of a model that takes into account game state, and I believe it’s one of the main reasons xStats always seem to say top teams are over-preforming and bottom teams are under-preforming. For example, Arsenal’s are often in a positive game, playing with a lead, so they can be more efficient with their counter-attack taking high xG shots at a lower volume of shots. Leeds are often in a negative game state forcing them to push forward against an opponent with a multi-goal lead creating a high volume of shots but with much lower xG values.
Xg is a metric used to measure what the average player would score like combining all the haalands and league 2 players to form an average
Great insights
No statistic ever tells the WHOLE truth. The end.
Football goals have lqmy variables such as player stamina wining or loosing what etc... its cannot be measured
Well it can be and has been
@@bighamster2 how
These days people care more about XG, shots on target, runs, etc more than scoring goals, winning matches and trophies. 😂😂
Liverpool fans can't stop crying about the UCL final.
If my team has more XG but loses, they've been screwed over. If they have less XG than their opponent but still win, XG is a fake stat. Simple
Im subscribed to Tifo because of your voice bro.. 😅
eye test is best test