|
Post by tachyon on Oct 12, 2022 9:21:10 GMT
Is the simple conclusion here that WBA have been unlucky? Or terrible at finishing? 1) Playing from behind in 9? games. Opposition may drop into a low block, allowing WBA to rack up lots of low quality chances. So in short game state has played a part. The good thing from WBA pov is they are creating some chances, they aren't totally terrible and conceding first isn't a persistent trait. Here's their xG shot map. They are taking a lot of speculative, small xG attempts from out wide and outside the box. 2) Post shot xG would tell you if they've been finishing poorly (haven't looked) or if opposition keepers are playing out of their skin. (Either tend to be transient, rather than repeatable).
|
|
|
Post by tachyon on Oct 12, 2022 8:25:20 GMT
I’m fully on board with the concept of Expected goals but really have to doubt the application when Wilmot is given a higher XG than Delap for the Sheff Utd game. Both were big chances, but Delap's had a bigger pre shot xG than Wilmot, not smaller. Attachment DeletedBarring penalties, LD's goal is the biggest chance we've created all season. Chances from crosses have alot of outcome bias associated with their xG. When everything goes right, (pace of cross, player/keeper positioning, contact etc) they look easy. Two minutes previously, LD had a near post chance where he was pressured & he failed to register an xG by a whisker. We try not to overfit to the specific chance because that is generally more predictive. Include post shot data (placement, power, specific opposition pressure etc, which does overfit to the actual chance) and Delap's so called xG2 rises to around 0.95. The post shot xG of all three goals was 0.65 (Wilmot), 0.57 (Jags) and 0.95 (Delap).
|
|
|
Post by tachyon on Sept 18, 2022 8:51:03 GMT
Does that mean we were lucky not to lose? One way to look at the xG for a single match is with a simulated xG timeline that estimates how likley it is a side is winning, losing or drawing at any time in the game based on the quality of chances that they have faced or created. Here's yesterday. Attachment DeletedQPR created all the chances in the first half hour, but they weren't high quality ones, so it most likley that the game is still all square at 0-0 (75%). Stoke had their best spell from 30th min to the interval. Again not high quality chances, so although there is a possibility that either side leads, a stalemate is still the most likely current outcome (60%). QPR have the game's best chance (Dunne) on 53 mins (xG 0.37). Along with everything that has gone before, it's a 50/50 chance QPR now lead, 35% it's still level and a 15% chance Stoke have taken and held a lead, even though they been out chanced. QPR have another decent scoring opportunity on the 76th min, Chair (0.28 xG), that boosts their chance of leading to around 65% and drops Stoke to 10%. After that the game meanders to the end with a couple of low quality chances per side and the final probabilistic outcome is 65% a QPR win, 25% a draw and 10% a Stoke win. In terms of expected points, QPR would average a return of 2.2 points from that performance and Stoke would average 0.56. We got outplayed, but nicked a point (the second most likely outcome). On another, more likely day we lose and on another, much less likely day, we take all three points.
|
|
|
Post by tachyon on Sept 16, 2022 11:01:15 GMT
What would be Browns comparison compared to the average championship finisher? JB for Stoke in the Championship. 20 non penalty goals from 17.1 xG from 147 goal attempts over 6847 minutes. Clinical finisher, but you can't rule out a bit of positive variance. So he's probably an average finisher on a bit of a hot streak.
|
|
|
Post by tachyon on Sept 16, 2022 10:13:03 GMT
Thanks for that. Actually slightly surprised but he’s still more than 3 goals under xG whereas Brown was over 3 more goals than xG.
Just trying to understand the second to last paragraph. If there’s a 26% chance that an average PL finisher scores fewer than 23 goals from MBD’s chances, does that mean there’s a 74% chance they score the same or more?
Wouldn’t that suggest he was underperforming and in the lower percentile bracket of finishers in the PL? Or have I misinterpreted? [/quote] If you simulate every goal attempt for MBD in the PL, you get a distribution like this. If you add up the percentage likelihood that an average PL player scores 24 or more goals, you get 67%. So it's more likely a player scores more than MBD. But MBD's clumsy 23 goals might be just down to random variance. You can only start to become more certain that he was an under-performing finisher if the red bar for actual goals scored falls to around 18 or fewer goals.
|
|
|
Post by tachyon on Sept 16, 2022 8:42:24 GMT
Tachyon, any chance of Dioufs xG stats when he played for us? Yes, sure. He deserves alot of credit. His first season he was genuinely elite. Got on the end of lots of high quality chances and also was fairly creative for a player who was primarily a scorer. Here's his shot maps by season and xG summary. Attachment DeletedPer 95 and quality tailed off a bit after that, but he was still mostly a useful PL level contributor. In the PL he scored 23 non penalty goals from 26.3 xG, A slight under performance, but distribution of chances also matter (he was limited to some low quality chances in his dozen or so wing back outings in 2017/18) and 23 goals was the third most likely goal total based on his xG in the PL. There's a 26% chance an average PL finisher scores fewer than 23 goals from MBD's chances, so he was a decent and certainly not a poor finisher. Nice to see him smacking the goals in after his messy Championship seasons.
|
|
|
Post by tachyon on Sept 15, 2022 18:46:00 GMT
Your stats see a good player, my eyes see a pretty poor one. I don't suppose either are 100% accurate. Subjective player assessment is also subject to scrutiny. You want two things. Reliability. Everyone comes to broadly the same conclusion. Obviously that's not going to happen on here for a variety of reasons. But you can sort of test the second requirement. V alidity. Do the conclusions of, for want of a better word, the Oatcake scouting staff tally with an independent, objective rating system for that player. A variety of objective ratings for last season for player 9393 (JB's Opta ID) ranked him pretty highly, between 3rd - 6th best Stoke player. Eye test ratings for strikers very often coincides with a numerical approach, (much more than defenders). So if you see a poor player, but the numbers say he's around the 3rd to 6th best player on the squad, you have to come up with some pretty compelling and extensive reasons why there's a disconnect. "Isn't"
|
|
|
Post by tachyon on Sept 15, 2022 9:53:50 GMT
Not actually true. His goals are entirely consistent with the difficulty of the chances he and his teammates make for him. He scores as many easy chances as the xG of those chances suggest he will and he scores as many difficult chances as the xG of those chances suggest he will. There's no statistically significant difference, either here or at Barnsley or at both combined. All of football uses xG, which is cool :-) It is true. Isn't.
|
|
|
Post by tachyon on Sept 15, 2022 9:52:17 GMT
Not actually true. His goals are entirely consistent with the difficulty of the chances he and his teammates make for him. He scores as many easy chances as the xG of those chances suggest he will and he scores as many difficult chances as the xG of those chances suggest he will. There's no statistically significant difference, either here or at Barnsley or at both combined. All of football uses xG, which is cool :-) That's not actually an effective rebuttal as it only deals with the difficulty issue and not the time to think issue which from the unstatistic eye of a spectator he does indeed seem to fuck up much more often than execute as intended, Stats are great but there does still need to be a role for the eye of the beholder. Most players spend the whole 90+ mins thinking. It doesn't just kick in when the ball arrives at their feet. So "thinking time" is a construct of fans. Time between the ball arriving at a player's feet & time to them taking a shot/header might provide evidence for this theory (were it not largely a fabrication). But it doesn't add any insight into how likely a player is to score, either as a group or this particular individual.
|
|
|
Post by tachyon on Sept 15, 2022 9:23:39 GMT
He does fuck them up. He scores the more difficult chances where he doesn’t have to think. Not actually true. His goals are entirely consistent with the difficulty of the chances he and his teammates make for him. He scores as many easy chances as the xG of those chances suggest he will and he scores as many difficult chances as the xG of those chances suggest he will. There's no statistically significant difference, either here or at Barnsley or at both combined. All of football uses xG, which is cool :-)
|
|
|
Post by tachyon on Sept 2, 2022 6:56:48 GMT
Bonham's faced shots on target in the Championship since 2014/15 that are worth just over 20 post shot expected goals. He's conceded 18 (+ an own goal). That makes him 10% better than an average Championship shot stopper, but he's only faced 60 odd attempts, so the uncertainty is fairly wide. He's saved around seven shots that were more likely to go in than not and missed out on around five that are more often saved. Fielding's 1% better than the average Championship shot stopper over 350+ goal attempts spread over 5 Championship seasons. But he's 34 and 150 days old and his last meaningful minutes were in 17/18. Here's the shot plots for both. Attachment Deleted
|
|
|
Post by tachyon on Aug 18, 2022 17:39:39 GMT
Easily the most reliable Championship xG values available :-) With these numbers, you win the match 60% of the time (if you're Middlesbro). On average M'b take 2 points from the game and Stoke take 0.8 points.
|
|
|
Post by tachyon on Aug 17, 2022 9:06:22 GMT
|
|
|
Post by tachyon on Aug 17, 2022 9:04:02 GMT
For those interested in the NJ tenure here's a breakdown of just his league games. Attachment DeletedFirst graphic highlights his final ten games, with his entire Stoke career in the mini graphic towards the bottom. The second plot highlights what George was talking about in the NTT20 pod. Stoke wern't terrible, we were so-so, but the results were brutal, often single goal defeats. Final graphic our xG differential was -0.9, but our actual goal difference was -18. That's just the footballing gods having a laugh, big time.
|
|
|
Post by tachyon on Aug 17, 2022 7:37:18 GMT
Ahh the old expected goals chestnut, a stat designed to make teams that aren't picking up points feel like they have been hard done by. The 30 PL teams since 2014/15 who most over-performed their xG in the first half of a season averaged a 27% over-performance. The 30 Pl teams since 2014/15 who most under-performed their xG in the first half of a season averaged a 24% under-performance. Imagine each side's surprise when the combined over-performance and under-performance in the second half of the season for the respective groups was just +1% and -1.1%. Guess their good luck/bad luck largely ran out. Pretty smart chestnut :-)
|
|
|
Post by tachyon on Aug 16, 2022 17:32:50 GMT
Problem is, seems most shots on target go in. Less pressure on the shooter - easier to shoot accurately and with power. True, which is why these are factored into the modelling.
|
|
|
Post by tachyon on Aug 16, 2022 17:30:46 GMT
There's some logic to Tachyon's stats but, really, are three or four games enough to go on? Sample size is a factor. What the likelihood x number of goals are allowed from 10 on target shots having a individual post shot xG values of a,b,c,d,e,f,g,h,i,j. If that likelihood is very small, the likelihood of your subject being a below average keeper is quite high. in isolation, it sets alarm bells ringing and you start to look at other things the data might tell you, such as positioning, technique etc. If you have data from previous seasons, you use that as well. Team data is weighted from the previous 46 games, so a couple of matches just moves the baseline numbers slightly.
|
|
|
Post by tachyon on Aug 16, 2022 15:38:39 GMT
Single game or single shot xG provide a more informed version of what happens in a match. "Brown should score that" is replaced by " a shot from that location is scored 18% of the time". Less opportunity for ill informed bias/opinion to prevail. Single shot xG does vary (all models are slightly different), but in the aggregate over more games you do get a general consensus. (despite the marketing wars). We regularly use xG for a couple of bits of content. View AttachmentTop two graphics are Stoke's 10 game rolling xG, first under Rowett & Jones, then under MO'N. Blue areas are when we've created more xG than we've allowed (good process-better results are likely to follow, although that's not a given. "Randomness" plays a huge part in single games). Orange is where we've allowed more than we've created (poor process-results are likely to be poor). When the blue line dips, we've been having trouble creating chances. When the orange line rises, we've struggled to prevent chances. Bottom graphic is an alternative league table. I've simulated every goal attempt, in every game so far, 10,000 times, added up the points in each simulation to calculate the most likely current position, based on what's happened in each game. Stoke's most likely current position is 7th, the xG for (inc pens) & xG against are listed. The fcst position is where we are most likely to finish, based on the actual points we've won so far and how well I think we will do in the remaining matches (based on the rolling xG plots for us and the other 23 teams). That's 11th if it's difficult to read. xG is a very flexible, unbiased, non narrative driven tool that virtually every team uses. Brilliant, thanks. I have a few questions really but I’ll restrain myself to one for now, about your fcst value… you say that’s “based on the rolling xg” - Watford have performed poorly so far this season (exp 20th) *and* presumably their recent rolling xg can’t be very good considering they finished 2nd-bottom of the Prem last year… so how do they end up forecast as champions? Doesn’t seem to make any sense.. or are the values weighted differently for the Prem or something Yes, there's an exchange rate based on the historical performance of relegated sides from the PL in their first season in the Champ. xG created increases, xG allowed decreases. Can't remember off hand what I'm currently using. Here's Watford & Norwich, last season in the PL. Plotted to the same scale, so the former were quite a bit better than the latter. Attachment DeletedFor teams like Burnley who have their unique brand of financial hell, I take the underlying numbers, but also take a consensus from our traders and look at the games as they are played to see if there's any systematic error in out rating.
|
|
|
Post by tachyon on Aug 16, 2022 15:00:23 GMT
I’ve got an expected shags (XS) of 4.2 but usually end up with an XS of 0.8. It’s all claptrap. Got the double up! The chances of that in consecutive post is nearly 100/1 :-)
|
|
|
Post by tachyon on Aug 16, 2022 14:56:11 GMT
Incidentally though tachyon , are the stats in the OP correct? From what I checked we were about 8th in the xG table. Norwich (nowhere on that infographic) were top, and Bristol City (5th here) were rock bottom! Single game or single shot xG provide a more informed version of what happens in a match. "Brown should score that" is replaced by " a shot from that location is scored 18% of the time". Less opportunity for ill informed bias/opinion to prevail. Single shot xG does vary (all models are slightly different), but in the aggregate over more games you do get a general consensus. (despite the marketing wars). We regularly use xG for a couple of bits of content. Attachment DeletedTop two graphics are Stoke's 10 game rolling xG, first under Rowett & Jones, then under MO'N. Blue areas are when we've created more xG than we've allowed (good process-better results are likely to follow, although that's not a given. "Randomness" plays a huge part in single games). Orange is where we've allowed more than we've created (poor process-results are likely to be poor). When the blue line dips, we've been having trouble creating chances. When the orange line rises, we've struggled to prevent chances. Bottom graphic is an alternative league table. I've simulated every goal attempt, in every game so far, 10,000 times, added up the points in each simulation to calculate the most likely current position, based on what's happened in each game. Stoke's most likely current position is 7th, the xG for (inc pens) & xG against are listed. The fcst position is where we are most likely to finish, based on the actual points we've won so far and how well I think we will do in the remaining matches (based on the rolling xG plots for us and the other 23 teams). That's 11th if it's difficult to read. xG is a very flexible, unbiased, non narrative driven tool that virtually every team uses.
|
|
|
Post by tachyon on Aug 16, 2022 13:13:50 GMT
Right footed, played almost exclusively for Oxford as an inverted winger in L1. Split starting time as an inverted winger at Brentford (38% of time), natural winger (20%) & RB (20%). the rest was spread between a variety of LB/midfield positions. Spent around a third of his touches in the defensive half for Brentford. Created 0.12 expected assists per 90 from open play, 85% of the chances were to feet. Compare that to TS's 0.11 xA, 58% to feet and Tom Edwards 0.07 xA, 50% to feet. So he's as productive as TS & more productive than TE and doesn't need a big target man. Here's his shot map at Brentford, mainly to show that he's a comfortable on each flank and also turns up centrally and very occasionally in the six yard box. His xG is also 0.12 xG/90, TS is 0.02, TE is 0.01 He's got a much bigger eye for goal, admittedly from a more advanced position. High volume tackler, with a below par success rate, who regains possession from recovering loose ball or interceptions. Probably an upgrade on TS and better than any alternative given the immediate need.
|
|
|
Post by tachyon on Aug 16, 2022 12:39:36 GMT
Taychon last season pointed out the Bursik is one of the lowest performing Keepers across all leagues for saving regular shots, think it was low 50% odd. This is the post which I'm referring to: Bursik Shot PlacementShot placement for all attempts on target faced by JB in Championship. Blue circles are goals, orange saves. Bigger the circle, the more difficult the "save". Replace JB with a Championship average keeper, simulate every attempt 10,000 times and you only concede more goals than JB did once every 100 simulations. Sub optimal for all concerned. (site's image upload is playing up). Reading the above left me with the impression that maybe I was harsh on him last season. I went into analyst speak. It says that an average keeper would perform worse the JB just once in every 100 trials. Going back to the sim, there's a 96% chance you would have conceded fewer goals than JB did if he had been replaced by a league average shot stopper. So the most likely conclusion is JB is a below average keeper who has performed in line with that premise concerning his shot stopping abilities.
|
|
|
Post by tachyon on Aug 16, 2022 11:35:43 GMT
Shot location for and against in the Championship to date. It's not that unlikely that a team turns 4.7 xG into just 3 goals or fewer. Raw xG is only half the story. Distribution of low quality attempts and connected shot events also play a part. View Attachment Our post shot xG (which just measures the likelihood that on target shots will result in a goal is 4.3 for (3 goals scored) and 2.4 against (5 goals allowed). So that's saying their keepers did about a goal better than average at saving our shots and Bursik did about 2-3 goals worse? Or does that include defenders blocking shots too. It's entirely keeper dependent, no blocks included. Oppo keepers have made some good saves (Gayle vs Blackpool springs to mind). That's most likely going to become less extreme going forward. I don't want it to sound like a witch hunt, but at our end we've had a problem for the best part of a season and a half that's as clear as day to see in the data & it hasn't been addressed.
|
|
|
Post by tachyon on Aug 16, 2022 9:03:48 GMT
Shot location for and against in the Championship to date. It's not that unlikely that a team turns 4.7 xG into just 3 goals or fewer. Raw xG is only half the story. Distribution of low quality attempts and connected shot events also play a part. Our post shot xG (which just measures the likelihood that on target shots will result in a goal is 4.3 for (3 goals scored) and 2.4 against (5 goals allowed).
|
|
|
Post by tachyon on Aug 15, 2022 10:24:46 GMT
That statistic there on the defence is crazy though that there would only be a 2% chance of conceding that many goals. And most worrisome about that is that it can't even be attributed to just one player on the back line either. Unfortunately, post shot xG, is attributable to a single player. Left side combinations have been our xG for driver.
|
|
|
Post by tachyon on Aug 15, 2022 10:05:37 GMT
The attacking process is consistent, if not slightly improved from previous Championship seasons.
3 or 4 games is a tiny sample size. Three actual goals or fewer from around 7 xG happens around 10% of the time. (Strip out Morecambe and it happens 34% of the time, so 8 Championship teams would have our level of league under-performance merely by chance after 3 games).
Also under-performing our attacking xG hasn't been an issue in the previous 123 MON league games, so you're most likely looking at random variation. You can't alter what you can't control.
Our attacking xG is up, not down & poor conversion is likely just "noise".
Defensively, there is more of an issue.
We're allowing around the same quality and quantity of chances for a typical MON team, but it's just a 2% chance we concede 5 or more goals from the post shot xG of 2.54 (for on target attempts).
That's a recurring issue since 20/21, we're keeping chances well contained, but often they aren't being saved at anywhere near Championship level performance.
|
|
|
Post by tachyon on Aug 8, 2022 16:53:14 GMT
What was the xg of the header they missed about 4 yards out dead centre? 0.52 xG. Volleyed, head high cross, with unmarked player central, a couple of yards out & close enough to get a head on to it to register a scoring chance. Very slightly more than half of those chances are scored, the rest fly wide & off target. Two other data companies who run xG for the Championship got 0.44 xG and 0.61 xG. Fotmob's 0.8 xG looks the outlier.
|
|
|
Post by tachyon on Aug 8, 2022 10:27:35 GMT
A couple of points that have been raised post Blackpool.
Ratings generally relate how much you've helped the team to do well, but that's position & role specific.
It's easier to "score" points as a forward compared to a defensive mid.
So you adjust for position.
Average ratings for Championship strikers is 6.6, for attacking mids it's 6.4 and for central defenders it's 6.2.
WS got a rating of 7.12 on Saturday, corrected for position, that's 1.5 standard deviations above average.(very good).
The whole left side functioned well as a unit.
Wilmot pushed into the opposition half more than the right side did, he led the team in adding value with non chance making progressive passes. (Rating 2.8 standard deviations above average, also very good).
WS sat between the lines, creating space and an outlet/target.
Tymon turned Wilmot/Smallbone's progressions into high value/high volume chances.
Laurent slide across to provide defensive assistance & more defensive bodies when needed. (He had a low raw rating of 6.6, but it's difficult to accrue merits as a defensive mid and he too scored 1.5 standard deviations above average for his position....very good).
And Jacob Brown turned 0.5 xG into 0.8 on target xG (clinical) and created 0.4 expected assists. (2.4 SD above average, beginning to look like we played quite well).
The right side didn't function as well offensively, but was a solid defensive unit.
Flint played predominately on the right of centre, (presumably to help out Taylor) only occasionally stepping into the left side.
Taylor wasn't as advanced as Wilmot & his passing wasn't as valuable.
Baker didn't get any progressive runs or passes going and sat in front of Taylor.
Gayle dropped into Blackpool's left side to press their deep ball carriers.
Aside from a potentially unlucky ricochet in the Stoke box at 0-0, it was a controlled performance.
Stoke "won" the xG battle 2.31-0.91
|
|
|
Post by tachyon on Aug 3, 2022 6:49:15 GMT
Alright, I'll give you that one. With set pieces stripped out, that's 0.8 goal making crosses or through balls per game. A good way behind the likes of RWB thoroughbreds such as Todd Kane, Sorba Thomas and James Bree. I've never once put Tommy in the same category as those lads. But Tommy is the next best of the rest after them for in-play goal making chances. I wasn't exactly bowled over by Clarke's signing either and if anything, he appears a retrograde step from Tommy. By the way, I guess you'd be surprised that Josh made only 0.7 in-play goal making chances last season - less than Tommy! I’ll go with stats provided by tachyon not some nonsense you’ve made up thanks I seem to remember Smiths goal scoring chances created at something under 0.2 from open play and Tymon’s considerably higher. Smith was an OK WB, but with definite deficiencies, especially compared to Tymon, who was one of the Championship’s standouts in 2021/22. Tymon expected assists were almost exclusively from open play & 20% higher per 90 than Smith’s total xA. Tymon created higher quality chances per attempt, he crossed from high value positions to high value positions. Tymon didn’t hug the touchline, he came infield, he took players on (liked a nutmeg) and was efficient when he did. He got into the box much more frequently and he got on the end of better quality chances & his xG/90 was twice that of Smith. Tymon's ball progression via passes & carries were excellent. Smith unloaded inaccurate longballs more often. A third of Smith’s xA was from set pieces and the quality per chance created from open play was lower than JT. He hugged the touchline, rarely ventured in field, crossing origin & destination weren’t as good as JT. He didn’t take players on & was very inefficient on the rare times he did. Smith’s tailing off (30.3 years), Tymon is kicking on (23.2 years).
|
|
|
Post by tachyon on Aug 3, 2022 6:41:02 GMT
Love you input on here. Remind me again please how or why the xG is calculated and how it is now considered a useful pre-game indicator of the likelihood of a final score? It’s process verses outcome. XG measures the process, how likely an event is to happen in the long term based on historical data of what happened in the past over a large sample of similar scenarios. The outcome is how often it did happen in the short term. Outcome doesn’t always tally perfectly with process. It’s like tossing a coin 10 times and not always getting five heads and five tails. You can have a run of negative or positive variance that makes your scoring/conceding look better/worse than your underlying process, but going forward, your outcome will tend towards your process. Rolling xG is a measure of a side’s underlying wellbeing that is more predictive of future actual performance than using past actual performance. Long term, you get, more or less, what you deserve. Short term you can end up 3rd with a so so process, like Huddersfield last season.
|
|