|
On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track).
Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall.
I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO.
|
Varane and Konate affected by a virus (after Rabiot and Upamecano, most likely Covid let's be honest) :
www.lequipe.fr
Horrendous for France these are our best CBs. Please let the finals not be decided by viruses...
|
Netherlands were less likely to win against Argentina than England and Morocco to win against France. In terms of performance and chances created.
That said, Argentina and France were still deserving winners in all their matches.
|
On December 16 2022 23:28 MyLovelyLurker wrote:Show nested quote +On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO.
An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team.
If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game.
This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK.
|
France12886 Posts
On December 17 2022 00:51 Stratos_speAr wrote:Show nested quote +On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK.
I think you give far too much credits to football statistics, they seem barely useful for predictions from what I see. It is funny to read but I hope you don’t take what you say without a grain of salt.
In the hypothesis that we gather far more useful statistics about the squads / players / crowd / conditions etc, you could make decent models, but quite frankly it is not worth the time put into it.
About the viruses, well if we win with our completely demolished squad, it just shows how much of a farce the World Cup is in terms of competitiveness.
|
On December 17 2022 01:03 Poopi wrote:Show nested quote +On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. I think you give far too much credits to football statistics, they seem barely useful for predictions from what I see. It is funny to read but I hope you don’t take what you say without a grain of salt. In the hypothesis that we gather far more useful statistics about the squads / players / crowd / conditions etc, you could make decent models, but quite frankly it is not worth the time put into it. About the viruses, well if we win with our completely demolished squad, it just shows how much of a farce the World Cup is in terms of competitiveness.
As I've mentioned before, statistics aren't perfectly predictive; this is an error that a lot of people make. If someone gives Argentina a 52% chance to win, that means that out of 100 theoretical simulations Argentina would only win 4 more times. That is effectively a coin flip. It doesn't mean that Argentina will definitely win or that France is really worse than Argentina at all. In fact, they could still be considered the better team but there may be other factors contributing to why Argentina may be very slightly favored (e.g. having an extra day of rest).
Statistics give you an empirical picture of how a team has performed so far, and the best predictor for future performance is past performance. There are a slew of statistics that have been shown to have meaningful predictive value, but, again, no statistic is perfect.
Footballing statistics are in their infancy, but they've already proven to be very useful. Downplaying footballing statistics is just the latest sport to try to hold onto their Boomer ideals where "vibes" and "eye tests" and "grit" are things that people use to judge teams.
I think it's also weird that your arguments for France's quality dismisses conceding penalties when talking about their defensive performance. PK's are gained via offensive quality and conceded via defensive failures. Conceding a PK is no different than conceding an extremely high-quality scoring chance; in fact, that is literally exactly what a PK is. A solid 75+% of PK's are scored.
|
On December 17 2022 00:51 Stratos_speAr wrote:Show nested quote +On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK.
Yes and your point and mine are the same: variance dominates over a single tournament. Only one world actually happens. Sure, if you ran 1,000,000 world cups in parallel universes, you'd expect ergodicity, and the LLN to apply. But we only run one world cup.
|
On December 17 2022 01:23 Stratos_speAr wrote:Show nested quote +On December 17 2022 01:03 Poopi wrote:On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. I think you give far too much credits to football statistics, they seem barely useful for predictions from what I see. It is funny to read but I hope you don’t take what you say without a grain of salt. In the hypothesis that we gather far more useful statistics about the squads / players / crowd / conditions etc, you could make decent models, but quite frankly it is not worth the time put into it. About the viruses, well if we win with our completely demolished squad, it just shows how much of a farce the World Cup is in terms of competitiveness. Footballing statistics are in their infancy, but they've already proven to be very useful. Downplaying footballing statistics is just the latest sport to try to hold onto their Boomer ideals where "vibes" and "eye tests" and "grit" are things that people use to judge teams.
If you think that 'football statistics are in their infancy', you're not talking to the right people. Bookmakers in the UK regularly ban accounts doing xgboost and random-forest based statistical arbitrage. I work in AI research and know that people are modelling pitch movement with deep graph neural networks and diffusion models. What you're missing is as soon as everyone starts doing that, the edge fades. And don't expect your TV pundit to ever opine on those topics anyway.
You're coarse-graining 'statistics' as if it were a unified discipline, when there are tons of distinct regimes in stats and probability, and tons of known failure modes of summary statistics, expectations falling to predict skewness (one of the shortcomings of xG) being a famous one. Football is closer to conformal prediction, large deviations and extreme value theory - all extremely hard than the standard 'just concentrate these random variables and take the average' regime. The fundamental limit that there might always be just 2 or 3 shots on goal per side and match will always remain.
|
On December 17 2022 01:29 MyLovelyLurker wrote:Show nested quote +On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. Yes and your point and mine are the same: variance dominates over a single tournament. Only one world actually happens. Sure, if you ran 1,000,000 world cups in parallel universes, you'd expect ergodicity, and the LLN to apply. But we only run one world cup.
Yes, and this is part of the excitement of the World Cup, or any knockout competition for that matter. It's why it's everywhere in American sports and it's also why a lot of European football leagues are seen as really boring; they're so long that the best teams will almost always come out on top due to the larger sample size.
That said, stats still aren't useless for international competitions and they give us good insight into how a team is playing. Simply put, the stats show us that France's defense has been a weak point and that this is a valid weakness to look at when trying to predict how the final is going to go. It doesn't determine that anything is going to or going to not happen; it's still perfectly possible that France scores pretty early, sits deep, and Argentina fails to score, but it's also naive to try to interpret France's performance as some kind of 4D chess calculation and that this is why they are so good and why they will be able to win. If France had a better defense and midfield out there on Sunday that led to conceding fewer chances and holding the ball more they would certainly have a better chance to win.
If you think that 'football statistics are in their infancy', you're not talking to the right people. Bookmakers in the UK regularly ban accounts doing xgboost and random-forest based statistical arbitrage. I work in AI research and know that people are modelling pitch movement with deep graph neural networks and diffusion models. What you're missing is as soon as everyone starts doing that, the edge fades. And don't expect your TV pundit to ever opine on those topics anyway.
You're coarse-graining 'statistics' as if it were a unified discipline, when there are tons of distinct regimes in stats and probability, and tons of known failure modes of summary statistics, expectations falling to predict skewness (one of the shortcomings of xG) being a famous one. Football is closer to conformal prediction, large deviations and extreme value theory - all extremely hard than the standard 'just concentrate these random variables and take the average' regime. The fundamental limit that there might always be just 2 or 3 shots on goal per side and match will always remain.
I don't really see how anything you said is contradictory or relevant. I've said maybe four or five times now that [commonly used] statistics aren't perfect but that they also aren't useless as a predictive tool (what multiple people here have tried to claim). Are you trying to dispute that?
As for variability in statistics? OK. I've never said anything to the contrary. As for infancy? This is a relative term. Sure, there are always people deep in the weeds that are using statistics in ways that the overwhelming majority don't understand, but this doesn't change the fact that statistical analysis as it is applied to football is very new and rough-around-the-edges compared to other team sports. American football has been doing it since our parents were children. Baseball went total Moneyball a full generation ago. Basketball switfly followed suit. I don't follow cricket at all but from what I do know it seems like the cultural use of stats is more akin to baseball than anything else. Hockey is the most recent holdout but it still started several years before football, where xG was pretty much unheard of and unknown to the broader base of fans and analysts until maybe three our four years ago.
|
France12886 Posts
On December 17 2022 01:38 MyLovelyLurker wrote:Show nested quote +On December 17 2022 01:23 Stratos_speAr wrote:On December 17 2022 01:03 Poopi wrote:On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. I think you give far too much credits to football statistics, they seem barely useful for predictions from what I see. It is funny to read but I hope you don’t take what you say without a grain of salt. In the hypothesis that we gather far more useful statistics about the squads / players / crowd / conditions etc, you could make decent models, but quite frankly it is not worth the time put into it. About the viruses, well if we win with our completely demolished squad, it just shows how much of a farce the World Cup is in terms of competitiveness. Footballing statistics are in their infancy, but they've already proven to be very useful. Downplaying footballing statistics is just the latest sport to try to hold onto their Boomer ideals where "vibes" and "eye tests" and "grit" are things that people use to judge teams. If you think that 'football statistics are in their infancy', you're not talking to the right people. Bookmakers in the UK regularly ban accounts doing xgboost and random-forest based statistical arbitrage. I work in AI research and know that people are modelling pitch movement with deep graph neural networks and diffusion models. What you're missing is as soon as everyone starts doing that, the edge fades. And don't expect your TV pundit to ever opine on those topics anyway. You're coarse-graining 'statistics' as if it were a unified discipline, when there are tons of distinct regimes in stats and probability, and tons of known failure modes of summary statistics, expectations falling to predict skewness (one of the shortcomings of xG) being a famous one. Football is closer to conformal prediction, large deviations and extreme value theory - all extremely hard than the standard 'just concentrate these random variables and take the average' regime. The fundamental limit that there might always be just 2 or 3 shots on goal per side and match will always remain.
That is why the external factors (the REAL health status of players, the REAL group cohesion inside the locker room, the mindgames between players from the same club when they are facing each other in national team, etc.) that aren't easily known from the public are probably the best to get a glimpse of the "real" odds, but that's insider info that we don't get the luxury to have.
|
On December 17 2022 01:46 Poopi wrote:Show nested quote +On December 17 2022 01:38 MyLovelyLurker wrote:On December 17 2022 01:23 Stratos_speAr wrote:On December 17 2022 01:03 Poopi wrote:On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. I think you give far too much credits to football statistics, they seem barely useful for predictions from what I see. It is funny to read but I hope you don’t take what you say without a grain of salt. In the hypothesis that we gather far more useful statistics about the squads / players / crowd / conditions etc, you could make decent models, but quite frankly it is not worth the time put into it. About the viruses, well if we win with our completely demolished squad, it just shows how much of a farce the World Cup is in terms of competitiveness. Footballing statistics are in their infancy, but they've already proven to be very useful. Downplaying footballing statistics is just the latest sport to try to hold onto their Boomer ideals where "vibes" and "eye tests" and "grit" are things that people use to judge teams. If you think that 'football statistics are in their infancy', you're not talking to the right people. Bookmakers in the UK regularly ban accounts doing xgboost and random-forest based statistical arbitrage. I work in AI research and know that people are modelling pitch movement with deep graph neural networks and diffusion models. What you're missing is as soon as everyone starts doing that, the edge fades. And don't expect your TV pundit to ever opine on those topics anyway. You're coarse-graining 'statistics' as if it were a unified discipline, when there are tons of distinct regimes in stats and probability, and tons of known failure modes of summary statistics, expectations falling to predict skewness (one of the shortcomings of xG) being a famous one. Football is closer to conformal prediction, large deviations and extreme value theory - all extremely hard than the standard 'just concentrate these random variables and take the average' regime. The fundamental limit that there might always be just 2 or 3 shots on goal per side and match will always remain. That is why the external factors (the REAL health status of players, the REAL group cohesion inside the locker room, the mindgames between players from the same club when they are facing each other in national team, etc.) that aren't easily known from the public are probably the best to get a glimpse of the "real" odds, but that's insider info that we don't get the luxury to have.
Exactly, in every prediction market, insider information is huge. What we're seeing from France now is very worrying with this contagious virus and a close-to-B-team already. In fact there is a very real scenario where Benzema plays on Sunday simply because it'd be the only reasonable thing if Giroud or MBappe fell sick.
|
|
|
On December 17 2022 01:43 Stratos_speAr wrote:Show nested quote +On December 17 2022 01:29 MyLovelyLurker wrote:On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. Yes and your point and mine are the same: variance dominates over a single tournament. Only one world actually happens. Sure, if you ran 1,000,000 world cups in parallel universes, you'd expect ergodicity, and the LLN to apply. But we only run one world cup. Yes, and this is part of the excitement of the World Cup, or any knockout competition for that matter. It's why it's everywhere in American sports and it's also why a lot of European football leagues are seen as really boring; they're so long that the best teams will almost always come out on top due to the larger sample size. That said, stats still aren't useless for international competitions and they give us good insight into how a team is playing. Simply put, the stats show us that France's defense has been a weak point and that this is a valid weakness to look at when trying to predict how the final is going to go. It doesn't determine that anything is going to or going to not happen; it's still perfectly possible that France scores pretty early, sits deep, and Argentina fails to score, but it's also naive to try to interpret France's performance as some kind of 4D chess calculation and that this is why they are so good and why they will be able to win. If France had a better defense and midfield out there on Sunday that led to conceding fewer chances and holding the ball more they would certainly have a better chance to win.
... it's a trend, which is why you look at the xG-outcome correlation historically (see one of my previous posts) and realize it's clocked sub 50% throughout many previous world cups. Which means putting your money behind it to predict anything would be hazardous to say the least.
There's a well known quote in machine learning and stats that 'all models are wrong and some are useful'. We can replace 'models' with 'stats' or 'estimators'  There'd be a ton of money in finding a football betting martingale, both with bookies and obviously with club strategists. If you find it, share it with us But xG et al aren't it.
Here is Walid Regragui - Morocco coach and as you know world cup semifinalist after three months on the job - roasting possession and xG stats for 3 minutes. "We're here to win... [...] Guardiola at some point was driving me crazy as well (with possession) Of course when you have DeBruyne, possession is good. What matters to me is to win.'
twitter.com
|
France12886 Posts
On December 17 2022 01:43 Stratos_speAr wrote:
I don't really see how anything you said is contradictory or relevant. I've said maybe four or five times now that [commonly used] statistics aren't perfect but that they also aren't useless as a predictive tool (what multiple people here have tried to claim). Are you trying to dispute that?
Well, I can still say that they are mostly useless from my experience: aligulac is not a good performer in liquibet, the people who had the best results in "my petit prono" / fantasy leagues apps during WC2018 were mostly girls with no football / statistics knowledge, same with Formula 1 fantasy league: I was around top 1/2 in the TLnet league despite barely any F1 knowledge etc and not thinking too much about which teams to use in the fantasy leagues. I performed decently in liquibets when I was actively following starcraft 2, but this season I am performing even better despite taking approximately 0.5s for each vote, and not following current results. (mostly my personal opinion)
Statistics in different forms (using old school stats, AI or whatever) can be useful for betting, especially if you can make A LOT of bets over time, but as far as I know the "house" doesn't really allow you to win too much so they might prevent you from betting and stuff like that. I never bet so I am not sure if they actually do it, but that is why it's imo a waste of time if the betting websites do such things (and they are businesses so it's probably best for them to do it).
So to put it simple: statistics are useful (albeit not magical / perfect, as you perfectly said) but IMHO (from what I have heard), it is not really worth your time since winning strategies end up getting spotted by the house / betting websites, and they apparently prevent you from winning too much.
If they did not interfere, it could be cool, but as the other French poster said, if everyone starts using the state of the art in terms of predictive capabilities, nobody gets an edge anymore (except those who have insider info and can calibrate their models even better, since they have more and/or better data to work with).
|
On December 17 2022 02:00 MyLovelyLurker wrote:Show nested quote +On December 17 2022 01:43 Stratos_speAr wrote:On December 17 2022 01:29 MyLovelyLurker wrote:On December 17 2022 00:51 Stratos_speAr wrote:On December 16 2022 23:28 MyLovelyLurker wrote:On December 16 2022 23:19 RKC wrote:
I'm actually a huge fan of possession football, which explains my support for Spain. But football is a low-scoring game (unlike basketball, American football, or tennis) and therefore different in a sense that possession doesn't strongly correlate with goal-scoring chances and goals. In fact, this WC, a lot of the teams making deep runs have less possession than the opponent especially after taking the lead (France v England & Morocco, Morocco v Spain & Portugal, Croatia v Brazil). Of course the majority of mismatches between top teams and underdogs will have the top team dominating and winning the game - but there's also an unusual amount of matches of underdogs winning with significant less possession (Japan v Germany & Spain).
I do still believe that for an evenly match game (even in terms of both teams having enough time to train and prep in advance like club football, not necessarily even in terms of quality), higher possession will more strongly correlate with higher chances, goals, and wins. But when the game is between two top teams closely even in quality, possession becomes less of a factor as some top teams deliberately cede possession and hit on the break and still become very successful with such tactic (eg Mourinho's Inter, RM in CL last season, etc). Football at the highest level involves a lot of intangibles (or data that we have yet to properly track). Exactly this. France vs Morocco was a masterclass of this: happy to lure Moroccans in attack to create spaces in their otherwise impenetrable defensive wall. I think we could track quantitatively why this strategy of ceding the ball makes sense - with no disrepect to the brilliant Moroccans, because you have confidence your quality of defense can easily hold their attackers. xG is only a first-order measure and should probably be weighted by attacker-defense differential priors, IMHO. An important difference though is that while Morocco did make a Cinderella run, they actually aren't a particularly good team. If you increased the sample size significantly (e.g. run a hypothetical "World Cup league"), they would almost certainly fall off a cliff in terms of performance. It's easy to sit back and defend against a team that doesn't have notable offensive quality; Morocco has only scored .5 goals/game whereas Argentina has scored 2 goals/game. This is seen best in comparing the Morocco game vs. the England game; Morocco did very little with their possession because they didn't have the offensive quality (they were still under 1.0 xG until 2nd half stoppage time) whereas England put up a nearly 2.5 xG, had a ton of great chances if you just stick to the "eye test", and would've pushed France into Extra Time if Kane hadn't whiffed on a PK. Yes and your point and mine are the same: variance dominates over a single tournament. Only one world actually happens. Sure, if you ran 1,000,000 world cups in parallel universes, you'd expect ergodicity, and the LLN to apply. But we only run one world cup. Yes, and this is part of the excitement of the World Cup, or any knockout competition for that matter. It's why it's everywhere in American sports and it's also why a lot of European football leagues are seen as really boring; they're so long that the best teams will almost always come out on top due to the larger sample size. That said, stats still aren't useless for international competitions and they give us good insight into how a team is playing. Simply put, the stats show us that France's defense has been a weak point and that this is a valid weakness to look at when trying to predict how the final is going to go. It doesn't determine that anything is going to or going to not happen; it's still perfectly possible that France scores pretty early, sits deep, and Argentina fails to score, but it's also naive to try to interpret France's performance as some kind of 4D chess calculation and that this is why they are so good and why they will be able to win. If France had a better defense and midfield out there on Sunday that led to conceding fewer chances and holding the ball more they would certainly have a better chance to win. ... it's a trend, which is why you look at the xG-outcome correlation historically (see one of my previous posts) and realize it's clocked sub 50% throughout many previous world cups. Which means putting your money behind it to predict anything would be hazardous to say the least. There's a well known quote in machine learning and stats that 'all models are wrong and some are useful'. We can replace 'models' with 'stats' or 'estimators' There'd be a ton of money in finding a football betting martingale, both with bookies and obviously with club strategists. If you find it, share it with us But xG et al aren't it. Here is Walid Regragui - Morocco coach and as you know world cup semifinalist after three months on the job - roasting possession and xG stats for 3 minutes. "We're here to win... [...] Guardiola at some point was driving me crazy as well (with possession) Of course when you have DeBruyne, possession is good. What matters to me is to win.' twitter.com
...I really don't get where this conversation is going at this point.
We've agreed multiple times on the point that the World Cup simply has too small of a sample size to be reliably predicted by statistics. That still doesn't make them useless, and I never said that they were the end-all of predictions, just that they're better than "vibes" and "eye tests" and other nonsense that homers are trying to use to craft a new narrative for France's performance. I've also never advocated for definitive use of anything for betting markets.
I was a big fan of Morocco and all but maybe if Regragui's team had more offensive quality they would've been able to drum up some semblance of offensive threat against France and may have advanced; something that would've been reflected in the statistics.
Honestly it's really bizarre how this conversation has turned and it reminds me a lot of fundamentalist Christians in America when discussing basically anything that is empirically-based. They often try to argue against "science" and using data to support public policy as if it was an ideology or a monolith. Instead, they want to use vibes and feelings and their belief in something that is absolutely impossible to empirically support or to be epistemically shared between two people as the basis for these decisions.
But science isn't an ideology and statistics are the same. They're both just measuring things in the world and adjusting our knowledge, actions, and expectations to this. It's not like a smokey back-room with a cabal of cigar-smoking elites chose a particular statistic (or two) and decided that these need to work to predict things. Stats are just an after-the-fact numerical analysis of something that happened. They start with a bunch of nerds who are weirdly passionate about something crunching a bunch of numbers, and the numbers that generally find consistent use are the ones that tend to show some level of actual value in both explaining past performance and predicting future performance. However, if stats were really, really good at predicting sports, they would be boring! The entire point of watching sports is the excitement, the unknowns, and the surprises that you can get when watching them.
Well, I can still say that they are mostly useless from my experience: aligulac is not a good performer in liquibet, the people who had the best results in "my petit prono" / fantasy leagues apps during WC2018 were mostly girls with no football / statistics knowledge, same with Formula 1 fantasy league: I was around top 1/2 in the TLnet league despite barely any F1 knowledge etc and not thinking too much about which teams to use in the fantasy leagues. I performed decently in liquibets when I was actively following starcraft 2, but this season I am performing even better despite taking approximately 0.5s for each vote, and not following current results. (mostly my personal opinion)
Statistics in different forms (using old school stats, AI or whatever) can be useful for betting, especially if you can make A LOT of bets over time, but as far as I know the "house" doesn't really allow you to win too much so they might prevent you from betting and stuff like that. I never bet so I am not sure if they actually do it, but that is why it's imo a waste of time if the betting websites do such things (and they are businesses so it's probably best for them to do it).
So to put it simple: statistics are useful (albeit not magical / perfect, as you perfectly said) but IMHO (from what I have heard), it is not really worth your time since winning strategies end up getting spotted by the house / betting websites, and they apparently prevent you from winning too much.
If they did not interfere, it could be cool, but as the other French poster said, if everyone starts using the state of the art in terms of predictive capabilities, nobody gets an edge anymore (except those who have insider info and can calibrate their models even better, since they have more and/or better data to work with).
I mean I can get behind what you're saying in a general sense, but I also don't really know where the betting markets came from. I don't gamble and I was never explicitly talking about how people should bet money or how betting markets should be run.
Pundits and fans use stats as little more than a hobby and a passion when it relates to talking about sports, and that is all I was referring to. We went way off the beaten path here, but my original point was just that I don't think that narratives about France being "dominant" throughout the tournament or "clear favorites" going into the final are accurate, based on past performances as elucidated by statistical analysis.
|
France12886 Posts
France has more weapons at its disposal than Argentina, so if the referees aren’t too biased towards them, they should be the clear underdogs, which makes France the clear favorites.
When you consider how easily the Argentina squad seem to lose their temper, that helps understand why France is clearly favored if they maintain a decent squad (still unsure about the veracity of the viruses thing, maybe they can get better in two days?)
|
My point is that scientifically, you can upper bound the usefulness of predictive stats in many instances. The canonical example in Bayesian stats is the Cramer-Rao bound : you already know that no matter what one does, their prediction is never gonna be better than X% accuracy, where X can be, theoretically, computed.
Similarly it's very easy to state that football is a low signal-to-noise environment for match- and tournament-level predictions. And thus without giving a hard bound-like result, you can still say scientifically these will have poor predictive value, due to non-stationarity, adversarial behaviour (see Scaloni's changes in formation), low sample size, etc., as previously stated. (Basically there is a statistical literature around power of estimators and negative results, which I encourage you to read. If you're unaware of it, then it might be why my point seems a bit remote.)
Hence, you can decline to take pundit stats seriously without being a 'boomer', as you previously mentioned
If you want to have fun with stats for nerding around, then by all means - don't we all
|
On December 17 2022 02:35 Poopi wrote:
France has more weapons at its disposal than Argentina, so if the referees aren’t too biased towards them, they should be the clear underdogs, which makes France the clear favorites.
When you consider how easily the Argentina squad seem to lose their temper, that helps understand why France is clearly favored if they maintain a decent squad (still unsure about the veracity of the viruses thing, maybe they can get better in two days?)
The virus thing is bizarre. On the one hand it looks contagious, on the other, we saw Deschamps take (supposedly sick) Coman in his arms after France-Morocco ended, displaying little signs of worry. All I'm hoping for is a good finals win or lose, would hate it if we witnessed mistakes and poor play due to C-teamers.
|
Honestly I think that Poopi's points are pretty much just homer rationalization at this point and I just don't really agree on any of them so I'm going to move on:
Two days left until the final. I just found this article on ESPN.
https://www.espn.com/sports/insider/soccer/insider/story/_/id/35259622/lionel-messi-world-cup-proves-best-male-athlete-ever
The author lays out a number of stats and claims that they prove that Messi dramatically outperforms Ronaldo (and anyone else) as the best footballer of all time.
He then goes on to a wild ride of claiming that Messi is the best male athlete of all time in any sport.
Thoughts?
Important information:
-His definition of "athlete" is "person who plays a sport professionally" and his general metric is "how dominant were they in their sport".
-He pretty much summarily disregards any players from older generations because the competition wasn't as good back then.
-He refused to take American football, basketball, or hockey into account because they're not popular enough.
-He claims that the only comparable athlete could be Usain Bolt, but doesn't actually lay out any argument as to why Messi would be considered the better athlete.
|
France12886 Posts
On December 17 2022 02:48 Stratos_speAr wrote:Honestly I think that Poopi's points are pretty much just homer rationalization at this point and I just don't really agree on any of them so I'm going to move on: Two days left until the final. I just found this article on ESPN. https://www.espn.com/sports/insider/soccer/insider/story/_/id/35259622/lionel-messi-world-cup-proves-best-male-athlete-everThe author lays out a number of stats and claims that they prove that Messi dramatically outperforms Ronaldo (and anyone else) as the best footballer of all time. He then goes on to a wild ride of claiming that Messi is the best male athlete of all time in any sport.Thoughts? Important information: -His definition of "athlete" is "person who plays a sport professionally". -He pretty much summarily disregards any players from older generations because the competition wasn't as good back then. -He refused to take American football, basketball, or hockey into account because they're not popular enough. -He claims that the only comparable athlete could be Usain Bolt, but doesn't actually lay out any argument as to why Messi would be considered the better athlete.
Michael Jordan easily comes to mind. The fact that he is (afaik) the first athlete to become a billionaire despite starting with « low salaries » (compared to current NBA) is pretty telling.
|
|
|
|
|
|
EPT
