Shuff264 Posted November 27, 2019 Share Posted November 27, 2019 27 minutes ago, tomonred said: That seems about right, Lawrence and Waggy both had decent chances. You wouldn't say they should score them, but they should hit the target. Link to comment Share on other sites More sharing options...
Tamworthram Posted January 1, 2020 Share Posted January 1, 2020 Whilst reluctant to raise this topic again (temporary relief from the Rooney hysteria) but, perhaps further proof that often, these stats are nonsense. According to the BBC, based on XG, Man City should be top (reality is they're 14 points adrift) and Newcastle should be bottom (they are 12 points clear). Also, Southampton should be 5th. Link to comment Share on other sites More sharing options...
duncanjwitham Posted January 1, 2020 Share Posted January 1, 2020 They aren’t nonsense at all though. They tell you something interesting about the game - Man City and Southampton probably create and miss a lot of chances, for example. If they always lined up perfectly with the actual results then it’s a useless stat because it’s telling you nothing you don’t already know. Link to comment Share on other sites More sharing options...
Ellafella Posted January 1, 2020 Share Posted January 1, 2020 On 25/11/2019 at 09:52, FindernRam said: last season I sat beside a guy who spent whole game putting stuff into a laptop, thought he might be a scout but he said at half time he was inputting data for a statistics company. He was like greased lightening on the screen tapping this and that but how he kept up with the game was beyond me and I did see him hit wrong name a couple of times. How he, on the fly, from way up in the stand decided what was a shot or a pass seemed ridiculous to me and people believe this stuff! Yes; inevitably there’s human error involved. But it’s only a small proportion and once you start getting 10000 observations you’ll find the “big data” stuff is predictive. And therefore very useful. Link to comment Share on other sites More sharing options...
FindernRam Posted January 1, 2020 Share Posted January 1, 2020 1 hour ago, Ellafella said: Yes; inevitably there’s human error involved. But it’s only a small proportion and once you start getting 10000 observations you’ll find the “big data” stuff is predictive. And therefore very useful. Being deliberately facetious I'll be 200 years old by the time Derby have 10000 observed shots! More seriously I've worked with statistics for 40 years and Project Managed several (successful-my words) computer systems with statistical analysis at their heart, I agree, with enough data and some good AI you can squeeze a lot of info out of apparently nothing. The issue I have with expected goals is not only do I think that the human error rate is quite high (given the guy is inputting tackles, fouls, shots, saves, possession etc) but the chap(ess) at another game is probably calibrated completely differently. Also over time with teams evolving you never get enough consistently good data. A simple example: Look at the BBC stats after a game then compare it to Google stats or Sky stats. The variation is often significant. The only thing they ever agree on is Derby shots on target--NIL Link to comment Share on other sites More sharing options...
Ellafella Posted January 1, 2020 Share Posted January 1, 2020 4 hours ago, FindernRam said: Being deliberately facetious I'll be 200 years old by the time Derby have 10000 observed shots! More seriously I've worked with statistics for 40 years and Project Managed several (successful-my words) computer systems with statistical analysis at their heart, I agree, with enough data and some good AI you can squeeze a lot of info out of apparently nothing. The issue I have with expected goals is not only do I think that the human error rate is quite high (given the guy is inputting tackles, fouls, shots, saves, possession etc) but the chap(ess) at another game is probably calibrated completely differently. Also over time with teams evolving you never get enough consistently good data. A simple example: Look at the BBC stats after a game then compare it to Google stats or Sky stats. The variation is often significant. The only thing they ever agree on is Derby shots on target--NIL Yes I was just making a general point about large numbers of observations and the relatively small proportion of error. Your issue about inter-rater reliability is a good one and I think you’ve got a good point about xG. I have yet to get a clear understanding about what it’s measuring really but clearly it’s a crude index of how many goals a team or playing might reasonably have scored given the “chances” they had. That then places the emphasis of what is a “chance” but I suspect by-and-large you can get enough inter-rater agreement to make its validity sufficiently high. For example Waghorn should have scored after 15 seconds v Charlton. Many stats companies actually use video of games for recording purposes enabling cross-checking rather than a single coder at a live game. As you’ll be aware statistics are not perfect when used in isolation but multi-variate statistical modelling when you use indicators for predicting the probability of future events are altogether another matter. Link to comment Share on other sites More sharing options...
Millenniumram Posted January 1, 2020 Share Posted January 1, 2020 8 hours ago, Tamworthram said: Whilst reluctant to raise this topic again (temporary relief from the Rooney hysteria) but, perhaps further proof that often, these stats are nonsense. According to the BBC, based on XG, Man City should be top (reality is they're 14 points adrift) and Newcastle should be bottom (they are 12 points clear). Also, Southampton should be 5th. They’re not meant to be used that literally though. They give an indicator about how well or not a team is playing, that’s it. There’s reasons as to why teams are/aren’t getting results that aren’t picked up by XG. But sometimes they show that a team is overperforming/underperforming, and that can predict things changing. For example, Southampton were really struggling, but now they’ve shot up the league, potentially getting the results their performances deserve. The opposite may happen to Newcastle, they’ve hardly been performing brilliantly, you may well see them drop down toward the end of the season. Link to comment Share on other sites More sharing options...
DRBee Posted January 1, 2020 Share Posted January 1, 2020 28 minutes ago, Millenniumram said: They’re not meant to be used that literally though. They give an indicator about how well or not a team is playing, that’s it. Not true - they try to measure the performance of a team using a limited aspect. They have come up with a construct "expected goals" , which in itself has limited value when set against all the other things that contribute to what might be called a good performance. Even that is to assume that this characteristic (i.e. expected goals) can be given a reliable numerical value. Psychologists eventually decided that IQ tests only measured the ability to pass IQ tests rather than measure the 'general intelligence' the tests were supposed to measure. Perhaps it will prove to be the same with 'expected goals'. I wonder what Jurgen Klopp would respond if told Man City are playing better than Liverpool according to 'expected goals' . I think we already know. Link to comment Share on other sites More sharing options...
Recommended Posts
Archived
This topic is now archived and is closed to further replies.