|
Post by tachyon on Mar 31, 2018 9:45:03 GMT
Charlie's potential assists are mostly crossed balls onto heads and set pieces. Our attacking creativity has shrunken, particularly after Arnie's departure. Combined with the awful defensive process, a relegation campaign was inevitable. This is what we're up against when we play Arsenal on Sunday. Attachment Deletedcheers for the heads up on SIMCA, it looks a neat package, PCA particularly is currently big in finding similar types of players from their advanced data. These plots are just done in R or excel before they go over to the app team for potential inclusion on later versions.
|
|
|
Post by tachyon on Mar 30, 2018 11:20:46 GMT
|
|
|
Post by tachyon on Mar 30, 2018 11:20:13 GMT
|
|
|
Post by tachyon on Mar 30, 2018 11:19:11 GMT
|
|
|
Post by tachyon on Mar 30, 2018 11:18:31 GMT
Here's the kind of statistics club analysts look at when evaluating players and their performances. It's a plot of so called expected goals and expected assists for each of our seasons from 2014/15 corrected for time played. Quick explanation. (skip if you're not interested) Players are credited for creating chances for others,(whether or not they are scored) and getting on the end of chances, (whether or not they score them). So if player A creates a one on one that's typically scored 60% of the time for player B, player A gets 0.6 expected assists and player B 0.6 expected goals, regardless of the outcome of the chance on the day. Idea is the outcome usually approaches the process in the long run. If you're getting on the end of chances, but not scoring, eventually reality will trend towards opportunity. (Although we do appear to have signed the biggest outlier in Premier league history). Our creative players are towards the top of the plots and our potential goal scorers towards the right. Players towards the top and the right are doing a bit of everything. Penalties are excluded. Here's 2014/15. The rest to follow. Attachment Deleted
|
|
|
Post by tachyon on Mar 28, 2018 21:00:06 GMT
Here's the shot stopping performance ratings for every Premier League keeper in 2017/18. First column's the average number of goals an average keeper would expect to concede given the quality and quantity of attempts each keeper has faced. Second column is how many each keeper has actually allowed. Basically, good shot stoppers at the top, average ones in the middle and those having a poor year at the bottom. Attachment Deleted
|
|
|
Post by tachyon on Mar 28, 2018 19:47:29 GMT
[/quote]The sum of shots saved and goals conceded would be an interesting stat because it would a good measure of how poor the defense in front of him has been. The ratio of saves to total shots on target would say how efficient he has been as a GK.[/quote]
Based on the quality of the 159 on target attempts Jack's faced, an average Premier League keeper would most likely let in 49 goals (excluding own goals). Jack's let in 49. So his shot stopping is bang average this season.
De Gea's expected total is 38 goals allowed. He's let in just 21. He's this season's top performer.
Next up's Pope. 22 Actual against an expected 30.
Hart's got the worst shot stopping record this season. Expected to let in 27, he's allowed 33.
|
|
|
Post by tachyon on Mar 28, 2018 16:08:32 GMT
Here's a graphic of all the on target attempts Jack has faced in the Prem this season. Oldest on the left, most recent on the right. The height of the bar represents the difficulty of the potential save based on shot or header location, how hard it was hit, where on the plane of the goal it would have entered the net, if it was deflected, had swerve etc. So an expected goals of 0.1 would result in a goal around one time in ten, 0.9 is a goal 9 times out of 10. Saves are in blue, goals are in red. Attachment DeletedJack's shot stopping is around the league average after a recent spurt of positive variance. OG's by convention aren't included.
|
|
|
Post by tachyon on Mar 16, 2018 17:44:13 GMT
|
|
|
Post by tachyon on Mar 15, 2018 14:14:24 GMT
A naive question: What difference does Mark Hughes appointment at Southampton make to the relegation maths. or does the model take no account of such human factors? Most managers get sacked because their side's performance levels have been below expectations *and* they've been a bit unlucky. The performance levels tends to persist after a new manager is appointed, but the luck tends to become less extreme. So there often appears to be a "new manager" bounce. Our model uses underlying performance indicators, rather than just goals and results and therefore our ratings aren't excessively influenced by random runs of luck, good or bad. Bottom line, we have accounted for Pellegrino's sacking & Hughes' appointment.
|
|
|
Post by tachyon on Mar 15, 2018 12:54:59 GMT
In addition to the "Like" "Dislike" buttons at the top of the page, we should also have a "I Don't Understand" button or "WTF" for short. which bits are you struggling with?
|
|
|
Post by tachyon on Mar 15, 2018 10:05:56 GMT
The Everton game is a potential tipping point. If we win our likely chance of relegation falls to around 36% Draw and we end up slightly worse than we are now at 56%. (The current most optimistic bookie has us around 54%) Lose and we are likely to hit around 67% to drop to the Championship. For those who prefer to avoid numbers, in a recent poll a 36% chance was perceived as being best described as "probably not", "unlikely" or "we doubt". 56% was perceived as "about even" and 67% equated to "probably", "we believe", or "very good chance".
|
|
|
Post by tachyon on Mar 4, 2018 18:19:37 GMT
cheers, mate. we're around 50/50 for Stoke after Brighton/Arsenal and the bookies go around 53% for us to go down. Implied bookies chance is around 31% for Palace to drop, 32% Huddersfield and 23% Southampton. I think 538 are a bit knee jerk in incorporating really recent form. Our model takes a weighted view over the last 38 matches. Here's our crib sheet for expected average points we think will be won from the remaining games including this week. Red games are the hardest, blue the easiest. Attachment Deleted
|
|
|
Post by tachyon on Mar 3, 2018 20:47:03 GMT
We're slightly better than a coin toss to stay up after Southampton. So a (very) slight improvement.
|
|
|
Post by tachyon on Feb 27, 2018 8:19:32 GMT
The 5/1 about Southampton to go down implies the bookmakers think it will happen around 17 times out of 100. (Slightly less, once you account for their margin).
The Southampton game is indeed huge.
If we lose our chance of relegation rises to 62 times out of 100.
If we draw it stays at around 50 times out of 100.
And if we win we only go down around 33 times out of 100.
The bookmakers are currently going 54% a Southampton win, 26% the draw and 20% a Stoke win. They've arrived at those figures because Southampton are at home and they also think they are a better team than Stoke. I'm slightly more optimistic, making it 51/25/24.
PS Palace's rating has been reduced to reflect their current injury situation.
|
|
|
Post by tachyon on Feb 26, 2018 22:30:20 GMT
Re run 10,000 simulations for the remaining matches after the weekend and we go down in 51 out of every 100 trials.
So it's virtually a coin toss.
37 points is when our survival becomes more likely than not. We can stay up with as few as 34 points, but it's hugely unlikely and we can go down with 39 points, but again that's unlikely.
WBA go down in 85 out of every 100 sims, Swans 48/100, Huddersfield 32/100, Palace 23/100, Newcastle 20/100, Soton 19/100, WHU 9/100, Brighton 8/100, Bournemouth 4/100, Everton 1/100, Watford 1/200, Burnley 1/1000.
|
|
|
Post by tachyon on Feb 15, 2018 9:06:02 GMT
According to this BBC article 38 points is more than likely enough to stay up whilst 40 points will certainly be enough.
35 points is the new 40.
www.bbc.co.uk/sport/football/43049564
"So how many points will be needed?
Using the same formula, Gracenote says 38 points "is more likely than not" to be enough, giving a 62% of staying up, but anything below that and it is a less than 50% chance.
"At this stage, it is certain that 40 points will be enough to stay up and that a team getting 34 points will be relegated," says Simon Gleave, head of analysis at Gracenote Sports."
We're slightly more optimistic than Simon. We think 37 points is the point where survival becomes more likely than not. Unfortunately each team has its own personal survival target and likelihood and Stoke's poor goal difference only gives us a 57% chance of staying up if we get exactly 37 points. The really bad news is our most likely final points total as of now is just 35 points and although we could stay up with only 35 points, that scenario only happens 15 times out of every 100 occasions we limp to 35 final points. It's all summarised from the Stoke perspective here.
|
|
|
Post by tachyon on Feb 12, 2018 11:37:29 GMT
[/quote]Sorry but that just doesn't tally at all. Pulis WAS able to attract attacking quality, he just didn't know how to use it. He spent £10m when that was still a lot of money on Crouch and didn't know how to use him, he neglected the wingers, he didn't know what to do with Adam or Owen, he over-used Walters. He'd created an effective team that was quite good to watch and spent good money to dismantle it.
It unravelled for Hughes in the end but if his process was so much worse why did it deliver more goals and higher league finishes for the first few seasons on a lesser net spend?[/quote]
We'll have to agree to disagree :-)
Crouchie was 30 when we bought him, two years past peak age, Owen was 32 and had a history of injury. Adam was bought for his dead ball delivery, if not his penalty taking.
Tony bumped along at a goal a game and never let the goals allowed stretch past the 1.4 per game we posted in our first Premier league season.
Hughes averaged 1.1 goals per game in attack, but lost the defensive process. 1.4 to 1.5 to 2.2 when he was sacked.
The underlying chance creation was even worse, despite the "encouraging" finishing positions.
The board perhaps mistook positive variance for a sound, underlying process. But it has been largely downhill since before the end of the 2014/15 season, with the occasional break to the surface to take in air.
Still ~40% we stay up, so not all lost yet.
|
|
|
Post by tachyon on Feb 12, 2018 9:24:44 GMT
Was that our 'solid attacking process' under Pulis that saw us finish 92nd and 91st respectively for goals scored in the entire pyramid in his last two seasons? If you are unable or unwilling to attract big money attackers (min £35 million) you have to rely on a solid defensive process and eek every last drop out of attacking strategies that require the least amount of skill. Which means set plays. Pulis' genius was that he managed to produce the occasional high quality opportunity (in terms of likelihood of scoring, rather than quality of build up). Seyi's double chest trap from two yards from a Rory flick on against Arsenal, for example. Aesthetically, not great, but a high chance of being converted. Pulis at Stoke, Palace and WBA managed to keep the the defensive process around 1.2/1.3 goals per game in matches where points were likely. Even if your attacking process was only around 1 goal per game, you had a chance. A solid process. Hughes was much more prone to playing a poor man's version of the game played by the wealthier clubs, but without the expensive talent to implement it. Much easier for opponents to defend and much easier for them to beat us if our defensive process is neglected, which it has been. 1.8 expected goals per game allowed is where we're at this season and where we've been heading under Hughes. Survival is really, really difficult with those kind of numbers on the defensive side of the ledger and a variety of mid price misfits fueling you attacking process. Overall, a poor process and relegation shouldn't be a surprising outcome.
|
|
|
Post by tachyon on Feb 12, 2018 8:20:45 GMT
[/quote]Was that our 'solid attacking process' under Pulis that saw us finish 92nd and 91st respectively for goals scored in the entire pyramid in his last two seasons? [/quote]
Don't recall saying "solid attacking process", just "solid process", the combination of attacking and defensive actions. Pulis always made enough from niche attacking strategies (Rory, set plays,) but always has majored on the defensive side of the ball.
|
|
|
Post by tachyon on Feb 12, 2018 8:05:31 GMT
How is that calculated Tachyon? Goals are relatively rare in football, ~2.5 per game, so instead clubs measure themselves based on the quality of the chances they create and allow. If you concede a penalty, that's "worth" 0.8 of a goal because penalties are converted around 8 times out of 10. If you take a shot from long range, that's worth 0.03 of a goal because long range efforts are converted around 3 times out of 100. If you are allowing more chances than you are creating once the quality has been accounted for you'll end up with a negative expected goal difference (that's what I've plotted) and that's where we have been gradually heading since 2014/15. In short, it's a measure of your side's process, rather than their outcome. And while the outcome might outstrip you process in the short term (you finish 9th, but only had a process that deserved 13th), your actual results tend to eventually reflect the quality of you process. One fallacy in football is that "luck" and randomness evens out over a season and you get what you deserve. That's not the case, a season needs to be at least four times as long for that to start to apply. Lee Grant had a purple patch when he replaced Butland and that papered over our declining process for a while. There are lots of other examples of this. Newcastle finishing 5th in 2011/12, but with a poor process (16th next season) Swansea 8th under Monk, poor process, Monk sacked three months later.(Variance boosts you up and then variances takes away) Reading 3rd in the Championship last season, terrible process, currently 18th. Palace no goals, no points after 5 games, decent process, now 15th. Cardiff and Bristol City, bottom half finish in 2016/17, but both had good attacking and defensive processes, both currently in the playoff positions. We always had a solid process under Pulis, much less so under Hughes. I can't speak for Stoke, but many Premier League teams use this methodology to check on their own well being and use it to strengthen in vulnerable areas, even if the actual results aren't currently reflecting any weaknesses. There was a one day event for clubs in London last week where these topics are discussed, virtually every Premier League club was represented (Liverpool sent half a dozen) + Barca, PSG, MLS, many Championship teams, Derby for example................There was at least one Stoke supporter there ;-)
|
|
|
Post by tachyon on Feb 11, 2018 21:40:55 GMT
I've been thinking about this for a while now. When it started to become clear that the wheels were falling off big time and some fans first started to turn against Hughes around the middle of last season, one of the first things his supporters always said when his position was questioned was 'he led us to three 9th place finishes in a row give him more respect' Fair assessment. This is what our overall attacking and defensive process has looked like since 2014/15. Attachment DeletedThe black horizontal line gets you about a 9th finish. We've spent much of our time below it and have progressively declined. To have missed such a prolonged and fairly obvious trend and taken the finishing positions at face value is.....unfortunate.
|
|
|
Post by tachyon on Feb 11, 2018 9:48:00 GMT
Here's the shot maps for Stoke and Brighton yesterday. The size of the circles represent the quality of the chances created. Shots taken from where Shaq equalised from are typically scored 3 in a 100 attempts, hence the 3% value attached to his shot in the popup. Adding up all of the figures for all of the attempts gives you the so called "expected goals" for each side. This is the same xG figures that are put up after the matches have been shown on MOTD. Attachment DeletedOverall our xG was 1.61 compared to Brighton's 1.08, but half of our total was the penalty. Same old same old, we just aren't creating any decent open play opportunities.
|
|
|
Post by tachyon on Feb 11, 2018 9:12:39 GMT
Updated survival chances after that penalty. Go down, 54 times out of 100 simulations. Stay up 46 times out of 100. Most likely final points total 35 (11 times out of 100). Probably shouldn't do this but *if* Charlie smashes the penalty home *and* we hold on for three points, we'd stay up 67 times out of 100. I'll do a more detailed one after all the week's games have been played.
|
|
|
Post by tachyon on Feb 9, 2018 18:10:22 GMT
[/quote]An interesting article tachy, may I ask a couple of questions: 1. You say there is a 44% chance of beating Brighton, what is the confidence interval on that projection? 2. Are you able to go back a few weeks and determine the chance of a) Swansea beating Liverpool, b) Swansea beating Arsenal, and c) Swansea beating Liverpool and Arsenal?[/quote] Swans beating Liverpool had an implied probability of 0.08 or 8% or 11/1. For Arsenal it was ~14% or 6/1. Combined probability of them winning both was 0.012 or roughly 83/1. Over the last 11 completed seasons, they've been about 980 games where one team has had less than a 14% chance of winning the game. The implied probabilities suggest that 94 should have been successful and 92 were. So the model is well calibrated. An 83/1 chance sounds unlikely, but if you have enough opportunities for something as seemingly unlikely as this to happen, sooner or later it will. Unluckily for us Swansea had a boost of positive variance at an inopportune moment. They probably deserved the Arsenal win (they created chances worth 2.2 expected goals and allowed 1.1 xG), but ran their luck vs Liverpool (only created 0.5 xG and allowed 2.7 xG). Attachment DeletedHere's the xG shot maps for Swansea's two shocks and our last three games. The size of the circles denotes the quality of the chance. Bigger circles are more likely to be scored. We aren't pulling up any trees in the attacking half. T
|
|
|
Post by tachyon on Feb 9, 2018 9:32:43 GMT
Part of my day job is to model football matches from play by play data and use those models to make match & season long predictions.
You can only make projections based on the likelihood, rather than the certainty of events happening. On Saturday, it's generally accepted that Stoke have around a 44% chance of beating Brighton. (My model and the bookmaker's odds concur).
So the most likely single outcome is a Stoke victory, but it is slightly more likely (56%) that Stoke will win either one or zero points. The same probabilistic assessment can be made for every other Premier League game being played this weekend and every game due to be played until the end of the season.
You can simulate the remainder of the season using these derived match probabilities and a random number generator to produce actual match outcomes. Points "won" in these virtual matches can then be bolted onto the current actual points won/goals scored etc to simulate a final table.
The powerful aspect of this approach is that the process can be repeated thousands of times to create a range of possible scenarios, along with how likely each scenario is to occur.
I've just run 10,000 simulations of the remainder of the season and Stoke go down in 4,400 of these simulations. So the chance of us getting relegated prior to the Brighton match has around the same likelihood of us winning the Brighton match.
In short, prior to Saturday, we are a slightly biased coin flip to survive.
Such simulation can help to answer other questions.
The most likely final points total won by the 18th placed side is currently 35 points (23%), centered around a low of 31 and a high of 39. That's not quite our survival target though, because our goal difference is poor and in many of those scenarios we already finish in the bottom three, so aiming to beat the survival cutoff point for that single simulation merely improves our final position, but still leaves us in 18th spot.
It some simulations, we gain a maximum of 39 points.....but are still relegated and conversely we very occasionally stay up with just 33 points. Our most common survival total is 38 points and our most common relegation points total in these simulations is 34.
There are nine teams currently with double digit likelihoods of being relegated. WBA, Swansea, Huddersfield, Newcastle, Southampton, Palace, WHU, Brighton and ourselves. We play five such sides in the run in, Swansea play six and WHU just three.
Such games are viewed as "6 pointers". It is preferable, given the choice of one win from two games to defeat Brighton and lose to say Man City, rather than vice versa.
Unsurprisingly, if we took zero points from these five relegation six pointers, we are relegated in 97% of such scenarios. If we improve to take five points from these five games, out survival rate is still only 40% and it is only when we take six or more points from a possible 15 that we become more likely than not to stay in the Premier League.
We survive 80% of the time we take nine points from our rivals and 11 or more points from these games all but ensures survival.
I'll try to up date these projections as the probabilities collapse to actual results.
|
|
|
Post by tachyon on Jan 14, 2018 13:34:26 GMT
They're rolling 6 game averages, it's just that Barca are that much better than most of La Liga.
|
|
|
Post by tachyon on Jan 14, 2018 10:34:13 GMT
One more chart. This is used as a quick health check to see where there's room for the team to improve. Attachment DeletedIt's colour coded. Blue is above average, the darker the better. Red is below average, the darker the worse it is. First column is the average expected conversion rate based on the quality of chances a side is creating. Chances in the six yards box are higher quality than long range shots from 30 yards. League average is 10% and we're league average. That column's paired with the third column, the number of chances we've created this season. We're below average here. Second column is the average conversion rate of chances we're allowing, again based on their quality. 12.3$ is the worst in the league, tied with Liverpool!. They get away with it because of their attack and they've only allowed 168 attempts on goal to our 307 (fourth column). In summary all red is BAD, mostly blue is GOOD. footnote. Burnley allow lots and lots of low quality chances......that was us under peak Pulis. Read more: oatcakefanzine.proboards.com/thread/275712/stoke-statistically-2016-17?page=1&scrollTo=5876009#ixzz549bFXrJc
|
|
|
Post by tachyon on Jan 13, 2018 18:55:25 GMT
Here's another problem we have. It's not a dig at the particular player because keepers are inevitably under the spotlight and due to the partly random nature of scoring events, their success in saving attempts can vary considerably. Attachment Deleted This plots Jack Butland's rolling average save rate compared to his expected save rate. You first model how likely a shot is to go in based on shot location, shot type, power, placement, whether it was deflected etc and then you compare these numbers to how many goals the keeper actually concedes. Jack's not having a great season. If a keeper is saving more than you'd expect based on the quality of the chances his defence is allowing, you'll see the blue actual line below the orange expected line. That's obviously what you'd want. Unfortunately, the opposite has been true for large parts of this season, although Jack's numbers have improved of late. We're hopefully just emerging from a defensive perfect storm.
|
|
|
Post by tachyon on Jan 13, 2018 13:48:21 GMT
Potter's a risk coming from a relatively low profile league, where innovative tactics tend to work well. Ostersunds use analytics and their league positions have been under pinned by solid fundamental stats (xG etc), but it is easier to get a competitive advantage in weaker leagues. His Europa run will have raised his profile & I suspect he'll have clubs queuing up for his services, but pitching into a relegation battle with only half a season to go, doesn't immediately scream "good deal" for either Potter or us.
|
|