|
Post by tachyon on Jan 4, 2020 10:23:41 GMT
Some insightful comments. Bottom line is that an xG approach predicts future goals better than other current metrics, including previous goals. It has more signal and less noise. Outcome based evaluations are full of randomness and prone to human cognitive biases, which leads to faulty conclusions. Chris Sutton’s “he must score there” (for every opportunity….ever) & Owen Hargreaves’ “he has to be more clinical” The only current pundit who gets the probabilistic nature of goal scoring chances is Michael Owen. Prime example 1. Clinical finishing is a myth (sorry Tyrese). The biggest scoring over performers over the first half of a season based on the quality and quantity of their chances perform dead average in the second half. Same for the “couldn’t hit a barn door” brigade, as a group they regress towards the number of goals you’d expect based on the chances they are given. The skill is getting into good scoring positions, once there the identity of the chance taker is largely irrelevant. The skill differential at converting chances is very small at the professional level. (tin hat on). 2) xG didn’t defend or champion Jones, it simply pointed out that Stoke had scored fewer and conceded more goals than the quality of the chances suggested they should and natural regression was on the cards. Whether he was generating the level of process appropriate to the amount of investment was for a different type of analysis. Having taken 1 point from 21, he then took 7 points from the next 21. We’ve since taken 16 points from 36 with a new manager and an interregnum during an easier run of games. So, we’re gradually getting the rewards our process merits and we’re even slightly improving the process (although don’t get carried away with how much effect a manager can have other than by changing the playing squad) 3) Many clubs routinely use xG. For a fine example of how it is used at Liverpool check out this. linkTim’s describing Liverpool’s non shot xG model that doesn’t just quantify chances it quantifies every on field action. We also have a non shot model for all the major European leagues, if I’m feeling brave enough I’ll post up the figures for all our players under Jones & MO’N PROCESS. Keep doing beneficial things for the team. OUTCOME. Which one-off reality the probabilistic universe throws your way.
|
|
|
Post by scfc75 on Jan 4, 2020 10:32:50 GMT
I think it’s fair to say that sample size is crucial. You’ll see big swings, plus and minus, on an individual game basis, but the longer you use xG as your prediction tool the more likely you are to be successful in predicting outcomes provided you remain disciplined and don’t let emotion come into play. From what I’ve seen, xG is consistently more accurate in predicting final league positions than either bookmakers or fans (ie if you pooled all fans predictions and came up with a mean league position over, say, ten seasons, that method would be less accurate overall in successfully predicting Stoke’s final league position each season than xG would be).
|
|
|
Post by scfc75 on Jan 4, 2020 10:35:29 GMT
Some insightful comments. Bottom line is that an xG approach predicts future goals better than other current metrics, including previous goals. It has more signal and less noise. Outcome based evaluations are full of randomness and prone to human cognitive biases, which leads to faulty conclusions. Chris Sutton’s “he must score there” (for every opportunity….ever) & Owen Hargreaves’ “he has to be more clinical” The only current pundit who gets the probabilistic nature of goal scoring chances is Michael Owen. Prime example 1. Clinical finishing is a myth (sorry Tyrese). The biggest scoring over performers over the first half of a season based on the quality and quantity of their chances perform dead average in the second half. Same for the “couldn’t hit a barn door” brigade, as a group they regress towards the number of goals you’d expect based on the chances they are given. The skill is getting into good scoring positions, once there the identity of the chance taker is largely irrelevant. The skill differential at converting chances is very small at the professional level. (tin hat on). 2) xG didn’t defend or champion Jones, it simply pointed out that Stoke had scored fewer and conceded more goals than the quality of the chances suggested they should and natural regression was on the cards. Whether he was generating the level of process appropriate to the amount of investment was for a different type of analysis. Having taken 1 point from 21, he then took 7 points from the next 21. We’ve since taken 16 points from 36 with a new manager and an interregnum during an easier run of games. So, we’re gradually getting the rewards our process merits and we’re even slightly improving the process (although don’t get carried away with how much effect a manager can have other than by changing the playing squad) 3) Many clubs routinely use xG. For a fine example of how it is used at Liverpool check out this. linkTim’s describing Liverpool’s non shot xG model that doesn’t just quantify chances it quantifies every on field action. We also have a non shot model for all the major European leagues, if I’m feeling brave enough I’ll post up the figures for all our players under Jones & MO’N PROCESS. Keep doing beneficial things for the team. OUTCOME. Which one-off reality the probabilistic universe throws your way. As a stat nerd, I really appreciate your posts. Great insights 👍
|
|
|
Post by tachyon on Jan 4, 2020 10:43:11 GMT
I think it’s fair to say that sample size is crucial. You’ll see big swings, plus and minus, on an individual game basis, but the longer you use xG as your prediction tool the more likely you are to be successful in predicting outcomes provided you remain disciplined and don’t let emotion come into play. From what I’ve seen, xG is consistently more accurate in predicting final league positions than either bookmakers or fans (ie if you pooled all fans predictions and came up with a mean league position over, say, ten seasons, that method would be less accurate overall in successfully predicting Stoke’s final league position each season than xG would be). Great point. You can look terrible or brilliant in small sample sizes. (TC warning alert) xG expands sample size compared to actual goals because there are more chances created than goals scored. Non shot xG that quantifies every tackle, pass, interception, off the ball run, post shot keeper save, long throw :-) expands the sample size 50 fold.
|
|
|
Post by sheikhmomo on Jan 4, 2020 10:57:45 GMT
To say Benham is purely an xG advocate massively underestimates the man.
I know one of the most successful traders at one of the biggest bookmakers in the world who now believes pure xG to be complete hokum after being almost a missionary for it for years.
|
|
|
Post by estrangedsonoffaye on Jan 4, 2020 12:51:13 GMT
Bookmakers and gamblers have been using Xg for years, in fact it was largely pioneered by the gambling industry using the principles of expected value (i.e the average return if you placed the same bet 1000s of times.) The owner of Brentford, Matthew Benham is a professional gambler who set up his own xG driven stats based company SmartOdds who sell their xG information that has much more sensitive and accurate algorithms than those produced by Opta. They then sell this data to high rolling gamblers who place the bets on Asian markets to avoid the issue you mentioned about his bets not being taken. They use the same algorithms to inform their recruitment and performance. They have compiled a global “xG fairness” table which has various weightings and basically assesses how well a team is playing and places them on a global league table. If a team is performing well, they take an indepth look at who the major factor is. They’ve done this for the litany of players they’ve signed for peanuts and then sold for 10x profit. This has also allowed them to build systematically towards promotion despite having the second smallest budget in the league for several years. The formula used to calculate odds in expected value (so that the punter is ultimately always likely to lose money) is literally the exact formula used to calculate xG. Benham himself has said he will only sack a manager if their xG performance and thus expected points performance is low. It’s working a treat, many wanted Frank gone early in the year and now they’re firing on all cylinders. I still think it’s highly misunderstood because of the stilted way its been introduced to the common football fan, if it was total nonsense, football would have ditched it years ago. It’s a measure of performance, not result. If the only stat that “mattered” was goals scored then we’d barely talk about what happened in the 90 minutes, all xG is, is another method of saying “we should have scored those sitters eh?”. A very good example is Dortmund in 14/15, halfway through the season they were bottom, but their xG data suggested they should be 5th. They rallied to 7th by seasons end but their xG barely changed, they simply met their expected performance. Thank you for that information, very interesting stats, particularly the Dortmund example. Are you saying Stoke should not have sacked Jones and kept him and the results would have improved? Sounds a bit dodgy to me. The person I knew who couldn't get his bets on anywhere was before the days of internet gambling. He would be in his element now! No I wouldn’t say that, Jones was frazzled, making really weird decisions and I think never settled long enough on a system to provide a large enough sample to bear out the xG data. We played so many different systems that it created way too much variance around the whole thing for my money. Whereas the Dortmund example is a manager who had been successful, had established a way of playing and had the trust of the fans and board so there was patience there which bore out what the stats were suggesting. I think Jones would have picked up some more wins but it was the right decision to sack him IMO for various factors. O’Neill has performed very well against the bottom 5 and the xG shows that, but it also shows we’ve been flat out dreadful against non-bottom 5 sides, regressing even from what Jones was managing against them. I’m not advocating xG as a deus ex machina stat or anything, I just think it can be useful in articulating how games have actually played out and whether you may have something good developing, I also think its really useful for identifying over-achievement or under-achievement early on, like Charlton for example, we played much better and generated much better chances but they took advantage of mistakes and scored 3. They did this a lot early, but the data suggested an eventual regression which has now definitely come to pass. The inverse for Brentford prior to their great run. I think it’s a useful tool to look at, in conjunction with others. I’d never advocate using one stat definitively to run things by. I just think it gets a pretty bad press because it’s a bit different.
|
|
|
Post by estrangedsonoffaye on Jan 4, 2020 13:03:02 GMT
To say Benham is purely an xG advocate massively underestimates the man. I know one of the most successful traders at one of the biggest bookmakers in the world who now believes pure xG to be complete hokum after being almost a missionary for it for years. Oh the man’s a visionary, Brentford’s club structure is superb and has clarity on nearly every level. He’s constantly developing his approach based on what he sees and that’s commendable. But the fact they’ve generated close to 100 million profit from player transfers using xG at the forefront of their recruitment suggests there is something in it, or at least the way SmartOdds do it which as I say is much more sophisticated than the way Opta do it. I’ve no doubt he’ll be using everything he can additionally to help, but their entire set up is more cerebral than most clubs from top to bottom, independent of the stats they use.
|
|
|
Post by shipshape on Jan 4, 2020 22:22:23 GMT
|
|
|
Post by Little Gary Patel on Jan 4, 2020 22:28:33 GMT
To say Benham is purely an xG advocate massively underestimates the man. I know one of the most successful traders at one of the biggest bookmakers in the world who now believes pure xG to be complete hokum after being almost a missionary for it for years. Oh the man’s a visionary, Brentford’s club structure is superb and has clarity on nearly every level. He’s constantly developing his approach based on what he sees and that’s commendable. But the fact they’ve generated close to 100 million profit from player transfers using xG at the forefront of their recruitment suggests there is something in it, or at least the way SmartOdds do it which as I say is much more sophisticated than the way Opta do it. I’ve no doubt he’ll be using everything he can additionally to help, but their entire set up is more cerebral than most clubs from top to bottom, independent of the stats they use. He's a protege of bloom. He worked at star lizard before going alone and after they allegedly fell out. And blooms security around his operation is crazy, since they fell out apparently 🤨 I'd love to know what actually went on beyond the rumours that I've been told.
|
|
|
Post by tachyon on Jan 5, 2020 9:22:34 GMT
If you read the link to Liverpool’s TW describing their EPV model, you’ll know that their group of mainly Welsh physicists and chess champions have an xG model that puts a goal probability value on every possession anywhere on the pitch. We call our EPV model a non-shot xG model, but it’s the same thing. Having the ball on the edge of your own box has a low goal probability value, but it’s much higher if you’ve moved the possession to the edge of your opponent’s box. Instead of a couple of actual goals or a dozen or so attempts, you’ve now got typically a couple of thousand quantified events for each game. Just to pick a random example, we have 1835 events that led to a change in goal probability for the Wigan Fulham game in September, comprising passes, ball carries, interceptions, tackles, throw ins etc. This information is most usefully used to see where, how and by who a side is progressing the ball. Who’s holding the ball up well, who’s giving it away, where a side can be pressed and where you should try to disrupt their most profitable attacking areas of the field. You can generate tons of tactical/player evaluation stuff from this data in the form of heat maps etc, but as a general overview you can simply tally each players positive and negative contributions in a game to get a rating, expressed on a 0-10 scale. We currently do this for all the main European leagues and competitions, including the Championship. We just take a basic feed for the FA Cup because sides largely play understrength teams. Sorry shipshape :-) Here’s Stoke in 2019/20 in the Championship under NJ and under MO’N. Therefore, small sample and strength of schedule caveats. Attachment DeletedA couple of quick takeawys. It doesn't include the caretaker game. It omits very low minute cameos. MO'N has had slightly easier games. Keeper's are evaluated on distribution & cross claiming. but mainly on shot stopping (or not), depending on the post shot characteristics. Pinged into the top corner with power from close range is obviously more difficult than a mishit shot from distance that's straight at you. On that basis Jack's been worse than AF (fairly obvious to the eye test). Although the good news is that keeper under/over performance is almost as random a trait as "clinical finishing". JM's done very well under MO'N, probably position dependent. Ryan's return is very welcome. JM, SC, SV, TS & JA have been our most productive, if somewhat inconsistent performers. TC and SV have been more impressive under MO'N.
|
|
|
Post by scfc75 on Jan 5, 2020 9:35:33 GMT
Here’s a question tachyon .... Do any xG models take into account the height that the ball arrives, the pace, and its position relative to the player who is going to have the goal attempt? For example... A) Ball comes in from the byline, to the edge of the 6 yard box and is at the perfect position and height for the forward to slot home. B) Ball comes in from the exact same position on the byline to the edge of the 6 yard box but this time, the striker is slightly beyond the ball and it’s fizzed in very hard at waist height. Is there a model that factors all these things in? The ball is delivered from and to the same position on the pitch in both examples but clearly to the eye, A is more likely to be scored than B.
|
|
|
Post by sheikhmomo on Jan 5, 2020 10:19:51 GMT
Oh the man’s a visionary, Brentford’s club structure is superb and has clarity on nearly every level. He’s constantly developing his approach based on what he sees and that’s commendable. But the fact they’ve generated close to 100 million profit from player transfers using xG at the forefront of their recruitment suggests there is something in it, or at least the way SmartOdds do it which as I say is much more sophisticated than the way Opta do it. I’ve no doubt he’ll be using everything he can additionally to help, but their entire set up is more cerebral than most clubs from top to bottom, independent of the stats they use. He's a protege of bloom. He worked at star lizard before going alone and after they allegedly fell out. And blooms security around his operation is crazy, since they fell out apparently 🤨 I'd love to know what actually went on beyond the rumours that I've been told. Think it all goes back to the Premier Bet days. Benham was working for himself as well as the firm and Bloom didn't like it, it got messy about who owned modelling etc but it was essentially a pissing contest between two Alpha nerds
|
|
|
Post by tachyon on Jan 5, 2020 10:28:06 GMT
Hi mate, most are x,y co-ordinates rather than x,y,z. Tracking data will help with that omission, but it's expensive and it can take longer to process than it takes to actually play the game.
You can infer the added difficulty from some of the qualifiers that come with the data. For example, from a cross, from a corner and passes have over a dozen qualifiers, such as lofted, chipped, from keeper kick etc.
There's a couple of data suppliers, Opta probably the most well known, and Statsbomb, founded by an ex Brentford guy and former Pinnacle quant. The latter are giving the former a run for their money by including richer data feeds.
Some models give a subjective xG based on just the opinion of the collectors, others use an algorithm.
We get the xG out on the Infogol app's shot map in real time for all major European leagues and the Championship, so we use the latter. So we're tied to the instantaneous data feed coming from Opta's Leeds centre and they go through a post game checking process that can lead to alterations being made.
xG sort of serves two masters. It has to be descriptive, but also predictive. The greater you reflect the former, the more llikely you are to overfit what happend in the past and less likely you generalise into the future.
We try for the fewest inputs into the algorithm, without compromising the accuracy on out of sample test data.
|
|
|
Post by Deleted on Jan 17, 2020 10:52:17 GMT
|
|
|
Post by tachyon on Jan 21, 2020 8:46:46 GMT
xG update. MON's getting an uptick at both ends of the pitch. The rolling average for xG created has taken a gentle upswing and there's been a downward trend in the amount of xG we are conceding. Improving the defensive process was the major factor in turning NI around. For the almost the first time since we dropped into the Championship, we've got some positive separation between our attacking and defensive process. Most importantly for results, the statistically noisy outcomes have started to better track the process (but that always happens eventually). Attachment Deleted The orange line above the blue line in the first defensive plot is where JB was letting in the feeblest of chances (that's stopped happening). And the orange line below the blue line in the attacking plot is where we were bouncing high quality chances off the post (that's also stopped being an issue).
|
|
|
Post by markby on Jan 21, 2020 9:31:58 GMT
Isn't that its weakness though? It basically assumes that Gregory should finish a goal like Mitrovic should because they're in the same section! Conversion rates are both volatile and the skill gap between players is narrow. Lewandowski is over performing his xG by 11% this season. Last season he under performed it by around 20%. Ronaldo under performed his xG last season and Mitrovic also did last season. If you take the biggest over performers, "clinical finishers" if you like, as a group they always regress to being neither under or over performers in future trials. Unless your name is Messi and it is certainly counter intuitive, but finishing IS a skill, of course, but the talent differential between players in the same leagues is invariably too small to measure. The observed difference is almost always random variation that eventually regresses to the expected xG rate. Interesting analysis about "natural" or "clinical" finishers etc and the skills needed for the role.
I always remember John Aldridge talking about this many years ago. Someone remarked that he always seemed to be in the right spot to finish off a move. He replied that it's actually very difficult to predict exactly where a teammate is going to play a chance (unless the opposition has been completely carved open and outnumbered). For example, when a winger gets to the bye-line, he has the choice of playing it high or low, near post or far, ahead of the defence or cut back. And even if you can guess where the ball is intended to go, it's always possible that your teammate will miss-hit it, or a defender or gk will intercept or block it. Moreover, if it does come in where expected and you get to it, you've often still got a defender breathing down your neck.
Therefore Aldo's solution was simply to run into space i.e. where the defender isn't. If it comes to you then fine, you've got time and space to pick a finish. And if it doesn't reach you, then so be it - like buses, there'll likely be another one along shortly (at least if you played for a team like 80's Liverpool).
|
|
|
Post by elystokie on Jan 21, 2020 9:35:05 GMT
xG update. MON's getting an uptick at both ends of the pitch. The rolling average for xG created has taken a gentle upswing and there's been a downward trend in the amount of xG we are conceding. Improving the defensive process was the major factor in turning NI around. For the almost the first time since we dropped into the Championship, we've got some positive separation between our attacking and defensive process. Most importantly for results, the statistically noisy outcomes have started to better track the process (but that always happens eventually). View Attachment The orange line above the blue line in the first defensive plot is where JB was letting in the feeblest of chances (that's stopped happening). And the orange line below the blue line in the attacking plot is where we were bouncing high quality chances off the post (that's also stopped being an issue). Well mate, I stuck a tenner on us last night at just over 3/1, having been influenced by your stats, so I owe you a beer and thank you very much
|
|
|
Post by swampmongrel on Jan 21, 2020 9:45:03 GMT
xG update. MON's getting an uptick at both ends of the pitch. The rolling average for xG created has taken a gentle upswing and there's been a downward trend in the amount of xG we are conceding. Improving the defensive process was the major factor in turning NI around. For the almost the first time since we dropped into the Championship, we've got some positive separation between our attacking and defensive process. Most importantly for results, the statistically noisy outcomes have started to better track the process (but that always happens eventually). View Attachment The orange line above the blue line in the first defensive plot is where JB was letting in the feeblest of chances (that's stopped happening). And the orange line below the blue line in the attacking plot is where we were bouncing high quality chances off the post (that's also stopped being an issue). “For the almost the first time since we dropped into the Championship, we've got some positive separation between our attacking and defensive process.” Sorry to be thick, but could you expand on this? I’m not sure what it means.
|
|
|
Post by markby on Jan 21, 2020 9:52:19 GMT
So the clubs they have most wrong prove it's a good system?! I work in gambling and firms are becoming more and more dubious about it. The basic premise that all players are equal is just wrong. The basic premise isn't that all players are equal. They differ in their ability to get on to the end of chances and teams differ in their ability to create those chances. It is the conversion rates of those chances that is a narrowly banded skill and under or over performance at the sharp end of a scoring opportunity invariably regresses to the population mean (except Messi). It is used in the betting industry, for instance Matthew Benham's Smartodds uses xG extensively. Re. your bold, I remember once reading that when England manager, Walter Winterbottom once tested his England squad for speed of reaction. Although it varied somewhat, one player stood out as being so far ahead of the rest as to be almost off the scale entirely: Jimmy Greaves.
And if you look at Greavseys' goalscoring record, it's actually v.close to Messi and Ronaldo when it comes to goals per game over a long(ish) period in a top level league. This must surely have owed a great deal to his ability to react to a loose chance quicker than everyone else. And I think I'm right in saying that although he would never have won any prizes in a 100 yards race, over the first 10 yards, he was as quick as anyone.
Re the second part, Greaves never played in all-conquering teams the way Messi or Ronaldo do, nor do they have brutal defenders clogging the legs off them with impunity like JG had to suffer in his day. I'd guess the only player from those days who could "out-stat" Greaves was Gerd Muller, but he had the advantage of playing for exceptional Bayern and W.Germany teams.
P.S. Re Benham, I think I read somewhere that when it comes to signing a striker, he doesn't give a damn how many the player scores himself. Rather, the measure he uses is how many goals the team scores when the striker is on the field, versus how many they score when he isn't.
|
|
|
Post by markby on Jan 21, 2020 10:13:56 GMT
We tend to run 10 game rolling averages spread over multiple seasons to look at longer term trends. Single game xG is more a descriptive than predictive metric. For overall team ratings we take the last 40 games, weighted towards the most recent xG outcomes. Getting back to Benham at Brentford, am I correct in saying that he doesn't believe the accepted wisdom that "the best team always wins the league"?
That is, with football games being so random, you can have two or three broadly matched teams, but one of them prevails by a small margin, which could be traced back to a couple of individual games where the winner "got lucky" and/or the runner-up was unlucky.
Instead, he prefers to assess teams over as many as 100 games (i.e. two to three seasons), rather than the 40 you suggest, to determine which is the "best".
This would explain eg how Leicester won the Prem i.e. their stats will have indicated that they should have been higher the season prior to their title, when they narrowly escaped relegation, whereas the season following their title was a better marker of their worth (good, but not outstanding) in a reversion to the mean.
Which if correct, would just mean that they won the title during a "hot spell" where they outperformed their actual merit.
|
|
|
Post by markby on Jan 21, 2020 10:20:47 GMT
I've met Benham many times. He's a genius. I know he relies in his own models much more than any external ones. You say he's a genius, which is true, I'm sure, but would you say he's modest as well?
I believe he likes to recruit the very best Oxbridge graduates he can, such that he gets "worried" if he isn't the least intelligent guy in the room lol.
And as for using his own models, that would tie in for his aversion to recruiting staff from elsewhere in the industry, since he wants people with an open mind, rather than those who merely reproduce what they've learned in their previous job, which may not be up-to-date or reliable.
|
|
|
Post by tachyon on Jan 21, 2020 10:26:23 GMT
xG update. MON's getting an uptick at both ends of the pitch. The rolling average for xG created has taken a gentle upswing and there's been a downward trend in the amount of xG we are conceding. Improving the defensive process was the major factor in turning NI around. For the almost the first time since we dropped into the Championship, we've got some positive separation between our attacking and defensive process. Most importantly for results, the statistically noisy outcomes have started to better track the process (but that always happens eventually). View Attachment The orange line above the blue line in the first defensive plot is where JB was letting in the feeblest of chances (that's stopped happening). And the orange line below the blue line in the attacking plot is where we were bouncing high quality chances off the post (that's also stopped being an issue). Well mate, I stuck a tenner on us last night at just over 3/1, having been influenced by your stats, so I owe you a beer and thank you very much Cool. A win's enough reward :-) Attachment DeletedWe were fairly bullish on Stoke's chances compared to the general price. We made us around a 2/1 shot, partly down to our decent underlying numbers and partly down to WBA being someway short of being a legitimate table topping team.
|
|
|
Post by estrangedsonoffaye on Jan 21, 2020 10:26:31 GMT
I've met Benham many times. He's a genius. I know he relies in his own models much more than any external ones. You say he's a genius, which is true, I'm sure, but would you say he's modest as well? I believe he likes to recruit the very best Oxbridge graduates he can, such that he gets "worried" if he isn't the least intelligent guy in the room lol. And as for using his own models, that would tie in for his aversion to recruiting staff from elsewhere in the industry, since he wants people with an open mind, rather than those who merely reproduce what they've learned in their previous job, which may not be up-to-date or reliable.
I've never met him but his track record of constantly looking to update his methods, team and operations is truly admirable. His model is infinitely more sophisticated than Opta's where most clubs purchase their data on. At the moment, I believe they're working in a "danger" metric into their calculations which take into account moves where a shot never occurs, but you'd say it should have been a chance. Like the ball going through the 6 yard box but the striker just not getting his toe to it. He's a true visionary and in truth, I genuinely hope Brentford make it to the PL, I think they'd be as big a breath of fresh air as Sheffield United have.
|
|
|
Post by estrangedsonoffaye on Jan 21, 2020 10:36:07 GMT
We tend to run 10 game rolling averages spread over multiple seasons to look at longer term trends. Single game xG is more a descriptive than predictive metric. For overall team ratings we take the last 40 games, weighted towards the most recent xG outcomes. Getting back to Benham at Brentford, am I correct in saying that he doesn't believe the accepted wisdom that "the best team always wins the league"? That is, with football games being so random, you can have two or three broadly matched teams, but one of them prevails by a small margin, which could be traced back to a couple of individual games where the winner "got lucky" and/or the runner-up was unlucky. Instead, he prefers to assess teams over as many as 100 games (i.e. two to three seasons), rather than the 40 you suggest, to determine which is the "best". This would explain eg how Leicester won the Prem i.e. their stats will have indicated that they should have been higher the season prior to their title, when they narrowly escaped relegation, whereas the season following their title was a better marker of their worth (good, but not outstanding) in a reversion to the mean. Which if correct, would just mean that they won the title during a "hot spell" where they outperformed their actual merit.
Well even if you look at statistics, 46 is a pretty small sample size for any kind of pattern to emerge, never mind in a game as random as football. His entire logic behind hiring and firing managers is based on their statistical performance rather than their on pitch one. Which sounds like madness, but the ultimately correct decision to sack Mark Warburton all those years ago is testament to the fact he's made it work. Leicester in 15/16 was an incredibly random season, they were pretty much the only side to never pick up an injury. The teams around them collapsed into nothing and they came back from 2-0 down 4-5 times. You'd be hard pressed to find one of those factors for a team in a season, never mind all 3 in 1!. This basically comes down to his "global fairness" table which ranks clubs across leagues with weightings and uses an expected points model (based off individual game xG) to determine how good a side actually is. It's how they have plucked a lot of players from lower leagues and sold them for nigh on 100 million in profit. This is my ultimate frustration with xG and how a lot of people view it, football is a random game, xG attempts to articulate some of that randomness and in doing so provides you with a guide of whether your system is working or not. People seem to think it's advocates are trying to get it to replace what happens on the pitch, it's never been the intention.
|
|
|
Post by markby on Jan 21, 2020 10:40:47 GMT
How's your stats pHD going out of interest? Every time tachyon comes on here and explains you refuse to listen and stick to your idiotic guns about it being "nonsense". You complain like fuck about our recruitment, yet admire Brentfords. Are they just plucking it out of thin air? Think they might have some predictors about future player performance or players who are undervalued? Nah, they're nonsense. According to Benham, whenever the stats contradict the evidence of his own eyes, he goes with the stats.
Every. Single. Time.
|
|
|
Post by tachyon on Jan 21, 2020 10:46:07 GMT
You say he's a genius, which is true, I'm sure, but would you say he's modest as well? I believe he likes to recruit the very best Oxbridge graduates he can, such that he gets "worried" if he isn't the least intelligent guy in the room lol. And as for using his own models, that would tie in for his aversion to recruiting staff from elsewhere in the industry, since he wants people with an open mind, rather than those who merely reproduce what they've learned in their previous job, which may not be up-to-date or reliable.
I've never met him but his track record of constantly looking to update his methods, team and operations is truly admirable. His model is infinitely more sophisticated than Opta's where most clubs purchase their data on. At the moment, I believe they're working in a "danger" metric into their calculations which take into account moves where a shot never occurs, but you'd say it should have been a chance. Like the ball going through the 6 yard box but the striker just not getting his toe to it. He's a true visionary and in truth, I genuinely hope Brentford make it to the PL, I think they'd be as big a breath of fresh air as Sheffield United have. A few clubs have a "danger" metric. It goes under a multitude of different names. One of Liverpool's quants showcased their EPV model here linkThe post from PrestwichPotter above is an aggregated version of our xG model that doesn't require a shot to credit the player moving the ball into a more dangerous area. link
|
|
|
Post by mrcoke on Jan 21, 2020 10:47:31 GMT
The basic premise isn't that all players are equal. They differ in their ability to get on to the end of chances and teams differ in their ability to create those chances. It is the conversion rates of those chances that is a narrowly banded skill and under or over performance at the sharp end of a scoring opportunity invariably regresses to the population mean (except Messi). It is used in the betting industry, for instance Matthew Benham's Smartodds uses xG extensively. Re. your bold, I remember once reading that when England manager, Walter Winterbottom once tested his England squad for speed of reaction. Although it varied somewhat, one player stood out as being so far ahead of the rest as to be almost off the scale entirely: Jimmy Greaves.
And if you look at Greavseys' goalscoring record, it's actually v.close to Messi and Ronaldo when it comes to goals per game over a long(ish) period in a top level league. This must surely have owed a great deal to his ability to react to a loose chance quicker than everyone else. And I think I'm right in saying that although he would never have won any prizes in a 100 yards race, over the first 10 yards, he was as quick as anyone.
Re the second part, Greaves never played in all-conquering teams the way Messi or Ronaldo do, nor do they have brutal defenders clogging the legs off them with impunity like JG had to suffer in his day. I'd guess the only player from those days who could "out-stat" Greaves was Gerd Muller, but he had the advantage of playing for exceptional Bayern and W.Germany teams.
P.S. Re Benham, I think I read somewhere that when it comes to signing a striker, he doesn't give a damn how many the player scores himself. Rather, the measure he uses is how many goals the team scores when the striker is on the field, versus how many they score when he isn't.
IMO Jimmy Greaves was the best striker ever, not just the best England striker. Unlike the so called greats who regularly play rubbish opposition, Jimmy always played in the toughest league in the world. He played against players who were allowed to tackle and on mud baths, not South American pitches. He topped the league scorers season after season, unlike anyone else before or since. Even though he was ill in the 65-66 season and Ramsay dropped him (or to be more correct he couldn't get back into a winning team after a severe cut injury), the following season he was the top scorer again in the top flight league. He has the fastest score rate of any top England scorers. He would be the highest ever scorer had Ramsay decided he was not going to use him for the 1970 world cup and he stop playing for England at a relatively young age of 27 in 1967. He scored more hat tricks than anybody. To top all that, he achieved his record scoring feats despite never taking a penalty! www.englandfootballonline.com/TeamGoals/Goals_10-49.htmlEvery Saturday night the BBC trot out that Alan Shearer is the highest goal scorer in the Premier League, and indoctrinate a younger generation. But that is only since the top England league was rebranded. Holder of most of the real records is in fact Jimmy Greaves. He scored 357 goals in 516 league matches between 1957 and 1972 for Chelsea, Tottenham Hotspur and West Ham United, which is actually 74 more goals than Shearer in 43 fewer games. I suggest that isn't just better, but a better class by far.
|
|
|
Post by tachyon on Jan 21, 2020 10:53:01 GMT
Getting back to Benham at Brentford, am I correct in saying that he doesn't believe the accepted wisdom that "the best team always wins the league"? That is, with football games being so random, you can have two or three broadly matched teams, but one of them prevails by a small margin, which could be traced back to a couple of individual games where the winner "got lucky" and/or the runner-up was unlucky. Instead, he prefers to assess teams over as many as 100 games (i.e. two to three seasons), rather than the 40 you suggest, to determine which is the "best". This would explain eg how Leicester won the Prem i.e. their stats will have indicated that they should have been higher the season prior to their title, when they narrowly escaped relegation, whereas the season following their title was a better marker of their worth (good, but not outstanding) in a reversion to the mean. Which if correct, would just mean that they won the title during a "hot spell" where they outperformed their actual merit.
Well even if you look at statistics, 46 is a pretty small sample size for any kind of pattern to emerge, never mind in a game as random as football. His entire logic behind hiring and firing managers is based on their statistical performance rather than their on pitch one. Which sounds like madness, but the ultimately correct decision to sack Mark Warburton all those years ago is testament to the fact he's made it work. Leicester in 15/16 was an incredibly random season, they were pretty much the only side to never pick up an injury. The teams around them collapsed into nothing and they came back from 2-0 down 4-5 times. You'd be hard pressed to find one of those factors for a team in a season, never mind all 3 in 1!. This basically comes down to his "global fairness" table which ranks clubs across leagues with weightings and uses an expected points model (based off individual game xG) to determine how good a side actually is. It's how they have plucked a lot of players from lower leagues and sold them for nigh on 100 million in profit. This is my ultimate frustration with xG and how a lot of people view it, football is a random game, xG attempts to articulate some of that randomness and in doing so provides you with a guide of whether your system is working or not. People seem to think it's advocates are trying to get it to replace what happens on the pitch, it's never been the intention. take your point. xG was developed precisely because football has randomness and is low scoring. 40 games with around 20+ shot based xG data points per game trades off between more recent evidence and more data points than simpy goals. With a non shot based xG model it is 40 games with over 1,000 data points per game.
|
|
|
Post by estrangedsonoffaye on Jan 21, 2020 10:57:32 GMT
Re. your bold, I remember once reading that when England manager, Walter Winterbottom once tested his England squad for speed of reaction. Although it varied somewhat, one player stood out as being so far ahead of the rest as to be almost off the scale entirely: Jimmy Greaves. And if you look at Greavseys' goalscoring record, it's actually v.close to Messi and Ronaldo when it comes to goals per game over a long(ish) period in a top level league. This must surely have owed a great deal to his ability to react to a loose chance quicker than everyone else. And I think I'm right in saying that although he would never have won any prizes in a 100 yards race, over the first 10 yards, he was as quick as anyone. Re the second part, Greaves never played in all-conquering teams the way Messi or Ronaldo do, nor do they have brutal defenders clogging the legs off them with impunity like JG had to suffer in his day. I'd guess the only player from those days who could "out-stat" Greaves was Gerd Muller, but he had the advantage of playing for exceptional Bayern and W.Germany teams. P.S. Re Benham, I think I read somewhere that when it comes to signing a striker, he doesn't give a damn how many the player scores himself. Rather, the measure he uses is how many goals the team scores when the striker is on the field, versus how many they score when he isn't.
IMO Jimmy Greaves was the best striker ever, not just the best England striker. Unlike the so called greats who regularly play rubbish opposition, Jimmy always played in the toughest league in the world. He played against players who were allowed to tackle and on mud baths, not South American pitches. He topped the league scorers season after season, unlike anyone else before or since. Even though he was ill in the 65-66 season and Ramsay dropped him (or to be more correct he couldn't get back into a winning team after a severe cut injury), the following season he was the top scorer again in the top flight league. He has the fastest score rate of any top England scorers. He would be the highest ever scorer had Ramsay decided he was not going to use him for the 1970 world cup and he stop playing for England at a relatively young age of 27 in 1967. He scored more hat tricks than anybody. To top all that, he achieved his record scoring feats despite never taking a penalty! www.englandfootballonline.com/TeamGoals/Goals_10-49.htmlEvery Saturday night the BBC trot out that Alan Shearer is the highest goal scorer in the Premier League, and indoctrinate a younger generation. But that is only since the top England league was rebranded. Holder of most of the real records is in fact Jimmy Greaves. He scored 357 goals in 516 league matches between 1957 and 1972 for Chelsea, Tottenham Hotspur and West Ham United, which is actually 74 more goals than Shearer in 43 fewer games. I suggest that isn't just better, but a better class by far. Not saying Greaves wasn't class, he palpably was. But his career did co-incide with a much higher goals per game. For all but the last few years of his top flight career, the goals per game in the top flight was always above 3, in some seasons as high as 3.5. Compared to Shearer who played in the top flight when the average goals per game was around 2.5-2.6 and didn't really fluctuate. It's a bit like the massive upsurge following the alteration of the offside rule in 1925. fivethirtyeight.com/features/in-126-years-english-football-has-seen-13475-nil-nil-draws/
|
|
|
Post by markby on Jan 21, 2020 11:10:12 GMT
xG didn’t defend or champion Jones, it simply pointed out that Stoke had scored fewer and conceded more goals than the quality of the chances suggested they should and natural regression was on the cards. Whether he was generating the level of process appropriate to the amount of investment was for a different type of analysis. Having taken 1 point from 21, he then took 7 points from the next 21. We’ve since taken 16 points from 36 with a new manager and an interregnum during an easier run of games. So, we’re gradually getting the rewards our process merits and we’re even slightly improving the process (although don’t get carried away with how much effect a manager can have other than by changing the playing squad) What you say about Jones/xG/results etc is all very true, but only insofar as it goes.
That is, while actual results weren't reflective of performance under Jones, that (unrewarded) performance was reflective of the players' efforts.
I'm open to correction, but surely the problem with Jones was that the players' efforts were in danger of receding as they lost confidence in him (or "losing the dressing room" as it's commonly known)?
Therefore Stoke were v.unlikely to have continued reproducing those xG stats the further the season went on, resulting in no reversion to the (old) mean, but instead a reduction in the level of the mean.
The point being that the players have a belief in O'Neill, which means they're redoubling their efforts, which means that with an xG which reflects their ability, they're starting to get results that would have been completely beyond Jones had he stayed.
I've met MO'N a couple of times (albeit briefly), but also people who've worked closely with him and everyone agrees he's very impressive. I saw nearly all his games in charge of NI from he took over. During the opening two years (first qualifying campaign), the results were terrible, yet without having access to the stats, fans generally agreed that the performances were much better than you'd think from the results. Of course, Michael will have appreciated that more keenly than anyone, yet he still had the humility to drive all the way from Edinburgh to Southampton in the close season to speak to his captain, the vastly experienced Steven Davis. Basically he asked him if it was worth carrying on. Davis reassured him that the players were still with him ("retained the dressing room") and he should stick with it.
And the rest, as they say, is history.
|
|