|
Brown
Feb 16, 2023 13:33:54 GMT
Post by tachyon on Feb 16, 2023 13:33:54 GMT
Indeed. I sincerely hope this is the not the reason for that celebration. *IF that is really the reason for the glasses celebration Brown's from Halifax. Huddersfield is five miles down the road on the 501 bus. Probably why he does a bit of a turn when he scores against them.
|
|
|
Post by tachyon on Feb 16, 2023 10:29:38 GMT
Jacob Brown vs Huddersfield in a hugely important, almost must win game.
2nd in presses. 3rd in progressive runs. 1st in touches in the opposition box. 2nd in featuring in passing sequences that resulted in a goal attempt. 3rd in interceptions. 1st in duels won. 2nd in completed dribbles. 1st in defensive duels won. 1st in successful defensive actions. 2nd in attacking duels won. 2nd in successful attacking actions. 2nd in sprints. Linked up as a passer & receiver as a wide right, a central striker and as an option around the cenre circle. Equal 1st in goals scored. Equal 1st in goal attempts. 1st in xG....by a country mile. 1st in penalties won. 1st in best celebration.
|
|
|
Post by tachyon on Jan 26, 2023 10:12:38 GMT
Where's he play: Five seasons as an left sided/wide, inverted attacking mid, frequently popping up in the half space channel, two more as a touchline hugging winger. As a goalscorer: 84% right footed, zero threat in the air, 0.17 non penalty xG per 90 in the Championship (8145 minutes). Smallbone's running at 0.08, Powell in the Championship is 0.32. Averaged 5 free kicks a season, so mostly open play attempts. Likes a shot from range, so lots of low grade efforts. Here's his shot map. (Waste's a lot of possession with poor quality shots?) Attachment DeletedCreativity: Around 0.17 expected assists per 90. Isn't too reliant on dead balls, 75% of his expected assists in the Champ were from open play. (compare Baker who runs at around 50/50). As a creator an upgrade on Smallbone, slight downgrade on a fit Powell. Last played in the Championship in 19/20.
|
|
|
Post by tachyon on Jan 25, 2023 9:28:04 GMT
Sarkic in the Championship with Birmingham. He faced 83 shots/headers on target. Conceded 29 goals (one was an own goal). Here's his save/goals map. Attachment DeletedPost shot expected goals was 25.7 (an average Championship keeper would have allowed 25-26 goals, OG's excluded). That's a shot stopping underperformance of around 8%, which just sneaks him into the top 20 of keepers who played at least 10 games in the 2021/22 season. For 2021/22 his shot stopping was below Davies and Bonham, but above Bursik and he hasn't yet done anything to justify the hype from the Birmingham fans ("best keeper in the league" was borderline delusional :-)) Good distribution.
|
|
|
Post by tachyon on Jan 13, 2023 18:53:49 GMT
Butland was crap in the Championship is becoming another one of those myths that become true on here. This is spot on. Butland was excellent in our first season in the Championship. A worthy player of the year. He prevented around six more goals than an average keeper would have. Second season was when his numbers fell off a cliff.
|
|
|
Post by tachyon on Jan 13, 2023 9:22:29 GMT
Out of interest, do you know what his stats are like for England? It isn't generally available, but you can scrape the play by play from match reports to get an idea and most games are available in full or highlight form. He hasn't faced too many attempts on target, (most games are big mismatches) but he's been a bit unlucky with a OG's & some virtually un-savable shots, also a penalty. He's neither stood out, nor looked a liability. He occasionally looks a bit rooted. His League One stints show him in a more favourable light at that level.
|
|
|
Post by tachyon on Jan 12, 2023 10:17:44 GMT
Delap 2022/23 xG/90 0.23, xG/attempt 0.10, Goals/90 0.21
Jacob Brown 2021/22 xG/90 0.29 (25% higher), xG/attempt 0.12 (15% higher), Goals/90 0.36 (78% higher).
Maybe try him up top.
|
|
|
Post by tachyon on Jan 12, 2023 9:48:15 GMT
You see this is the problem with over-relying on data. You've taken one data point that you read once on here, over one specific time period, which showed Bursik was performing below average for shot-stopping, and on that basis you've disregarded any other factors (other time periods, mistakes, performances for other teams, cross-claiming, control of area, distribution, decision-making, and all the other nuts and bolts of goalkeeping that are not saving shots) and you've allowed yourself to make up your mind that he is "a poor keeper at this level". Not to mention his continued involvement in the England setup - they use data as well you know, and I should imagine a lot more extensively than one statistic from one narrow time frame. To defend my keeping analysis. It's not one data point, it's 157 Championship data points, spread over 4500 minutes (his entire Championship career) described by over a dozen pre and post shot parameters, along with the keeper's pre shot positioning. They've been compared to every on target shot/header faced by every other Championship keeper since the beginning of the 2014/15 season. That's around 34,000 comparative data points, again each with a full set of player and attempt parameters. The conclusion is that in each an every season, shot stopping ability hasn't moved. It has been amongst the worst in the league and it has been consistently so. Compared to a league average keeper, we have conceded an average of an extra goal every four games with Bursik. That is one heck of an obstacle to overcome. His cross claiming is 10% above league average (you can't say the same for Bentley, should he arrive), his distribution is largely determined by the manager's preference. (Initially he went long more often, then MON wanted fewer long ball, & AN is back to going long more often). He's a perfectly average sweeper keeper, if called upon. All of these are an essential part of the keeping skillset, but they don't trump shot stopping. Are there easily identifiable problems? Yes, it is in the data clear as day. Have they been addressed either at Stoke or within the England set up (your faith in their extensive use of advanced data may be somewhat misplaced). They don't appear to been addressed or perhaps the skillset at Championship level isn't there to move the dial.
|
|
|
Post by tachyon on Jan 12, 2023 8:12:05 GMT
[quote author="shakermaker" source="/post/7684665/thread" timestamp="1673458993"I don't really understand those Goals Prevented stats, they don't make much sense to me.
Taking last season's stats into account for example, Bentley's Goals Prevented rate is 1.03. Is that per game? If so, that would be 1.03 x 38 = 39.14, but he conceded 56 goals.
[/quote][/b] It's his expected goals on target (52.4) divided by the actual goals he's allowed (51). Greater than 1 is above average, below 1 is below average.
|
|
|
Post by tachyon on Jan 11, 2023 17:44:35 GMT
Yeah, he doesn't come out great in Opta's 'Goals Prevented' stat.
But no idea how much of a meaningful metric it is, and how much stock to put in it. It can be very noisy in small samples. Here's what his 2022/23's shot placement looks like. Attachment DeletedHe's let in most of the difficult to save ones and a couple of the easier ones.
|
|
|
Post by tachyon on Jan 11, 2023 16:41:31 GMT
He's been in the Championship since 2016/17, largely as a starter until this season.
Overall a very, very slightly above average shot stopper.
He's allowed a whole 1% fewer goals than you'd expect from an average Championship keeper over those seven seasons.
Best season was 17/18 when he prevented 10% more goals than average based on the placement/shot type of the on target attempts he faced.
Worst season was the current one, where he's allowed 9% more goals than an average keeper.
He claims crosses at well below the league average.
|
|
|
Post by tachyon on Jan 6, 2023 17:41:32 GMT
I am not scoring points here, Mr(s) Tachyon, because whatever I might think, I accept that XGs is the currency of the day and your explanations are very helpful.
However, whilst teams can of course get lucky, it doesn't follow that Reading finished where they did that season through luck. They might or might not have had a better than average slice of luck - I don't think it is possible to measure, let alone know. But it absolutely has nothing to do with luck that their keeper had such a terrific season. Nor does it follow that XGs was a reliable predictor for their following season.
I guess I would warm to XGs more if there were a corresponding measure to reflect how effectively a team prevented goals. I suspect there must be such a thing. Is that how expected GD is measured?
Meanwhile here's one for the youngsters - goal average, anyone?
[/quote][/b] Luck was probably a flippant term because it implies a lack of skill, which obviously isn't the case for any professional team or player. But the alternative term is "random variation centered around the true talent of a team or player". Which tends to check out any interest. Performance is a combination at true talent & random variation. Sometimes the latter is positive & helps a side & sometimes it's negative and hinders them. But it doesn't stick around, you have little control over it and all you can say is it is likely to become less extreme over time. (It's why you can flip a coin (true talent 0.5) and sometimes get a run of heads or tails. True talent is unchanged with a fair coin, but variance can run hot or cold at times. Al Habsi allowed 19 fewer goals than you would expect an average keeper to allow in 16/17....which was an insane level of over performance. The previous two seasons combined he allowed just one fewer than expected. He couldn't have done it without a huge slice of.....positive random variation. xG is a good guide to a player's true talent. Any over or under-performance is correlated to the team or player's positive or negative random variation (or luck). Reading's 16/17 season was the biggest outlier in the xG era (that's why I picked it), but every natural process (which includes football) is prone to random variation & every prediction is improved by accounting for its presence (along with a measure of true talent). The Championship is just a massive experiment in how randomness impinges on actual outcomes over a (short term) single season, where true talent is kept relatively stable, outside the three relegated sides and their parachute payments. Love the idea of bringing back goal average. Go for it!
|
|
|
Post by tachyon on Jan 6, 2023 8:54:48 GMT
Also tachyon.. as you seem to be around, can you give practical (or theoretical) examples of what you mean by your second line about predicting "future goals"? Over how long? A team or player's xG changes all the time so what period are we looking at to make a more accurate prediction (and predicting for how long)? Dunno if that lay talk makes sense, I'm just after more real-life evidence of how it "works" in the way you say. My guess/how I'd test it... Look at a bunch of teams (10?) who are scoring way under xG and the same who are scoring way over. Then look at whether they score more goals in their next 5 than they did in their last 5. I bet the first group on average sees increased scoring and the second group a decrease. Could put this on the table with some flutters on the sports betting! 30 teams who under perform their Championship xG most in a season, pretty much hit their xG in the following season. 30 teams who over perform their Championship xG in a season also pretty much hit their xG in the following season. Outcome regresses towards their underlying process. 5 games is unfortunately rife with statistical noise. Mean regression is long term. It's almost meaningless to talk about "clinical" or "non clinical" teams. You're almost always seeing random variation, not repeatable skill. If a team wants to do better it has to improve their underlying process(which xG measures), not rely on the transient trait of finishing that is way, way over valued as a factor.
|
|
|
Post by tachyon on Jan 6, 2023 8:17:48 GMT
Roughly four whole repeated seasons to get most teams close to where their performance merits them to be. The table always lies for quite a few sides. Can you clarify what you mean by "repeated seasons" here? You mean simulations of the exact same season with the same xG? Just a season that has you playing a round robin schedule consisting of four home and four away games against the same side. So 184 total games for each team.
|
|
|
Post by tachyon on Jan 6, 2023 8:14:06 GMT
Roughly four whole repeated seasons to get most teams close to where their performance merits them to be. The table always lies for quite a few sides. I am struggling with the notion that the table lies (other than in circumstances where there is skulduggery to influence the outcome of games). However, I am prepared to be persuaded if there are examples of where it does lie. Teams get lucky. They sometimes winning lots of games by a single goal, sometimes these games turn into draws. With a bit of negative variance, draws turn into narrow losses. My fav example of a side that went on a prolonged hot streak, but without the underlying quality of process to sustain it was Reading in 16/17 when they finished 3rd & lost the playoff final in a pen shootout. Their goal diff was +4 (which wasn't great), they won 18 games and a playoff game by a single goal margin, but their expected goal difference was -16. Al-Habsi had a season that no keeper has had before or since. He wouldn't have repeated it even if he'd stayed. They were the 20th best side by xG in 2016/17, even though they finished 3rd. Next season they finished 20th.
|
|
|
Post by tachyon on Jan 5, 2023 15:08:18 GMT
People completely under-estimate what a long term sample in football is. I’d argue that even a full season is prone to some wild deviations; hence my pushback against the cliche’ “the table never lies”. Roughly four whole repeated seasons to get most teams close to where their performance merits them to be. The table always lies for quite a few sides.
|
|
|
Post by tachyon on Jan 5, 2023 6:25:21 GMT
No need. All to need to know is previous xG predicts future goals better than previous goals does. The "All that matters is the result" mantra is wrong. The "xG is shite" brigade are wrong. There's no debate to be had. xG has won.
|
|
|
Post by tachyon on Jan 4, 2023 22:04:19 GMT
. Does xG take into account how well struck or placed a shot might be (re your comment about opposing keepers making good saves) or does a scuffed shot straight at the keeper that "my granny" could save have no effect on xG? As a former keeper I am struggling to recall any games this season where I have left having been impressed by the saves made by the opposition keeper. (I am getting on a bit so my memory might not be all that. ) Football analytics has always had a problem naming metrics. (Hockey named theirs after players & bloggers....which is arguably much worse) Expected goals or xG as quoted in these posts refers to everything and everywhere that happens before a shot or header takes place. On target xG that I've mentioned in my post includes everything that happens before and after the attempt. So that does include how well the shot/header was struck, where it did/would cross the goal line, what's the keeper's line of sight like, was it deflected (a keeper killer) etc. It's also call post shot xG, PSxG. I called it xG2 for a year or two and still do for old times sake. Keepers are judged on xG2, not xG.
|
|
|
Post by tachyon on Jan 4, 2023 21:09:32 GMT
xG is the foundation of modern football analytics. It underpins every player/team model that is used in the game today.
It's a big part of why Liverpool won the PL & UCL, Brentford bought low & sold high and how they persuaded less clued up sides to pay over the odds for their players.
It's a soothing antidote to the idiocy of Ian Darke and his pundit mates claiming "he must score" & "this game should be 6-4" (usually at half time).
Fundamentally it looks at chances and the likelihood they are converted (nowhere near as high as people think), rather than just goals.
There are 10 times more chances than goals in a season, so by looking just at goals (the binary outcome of a chance), you are chucking away 90% of the data at the very start and taking a statistically noisy subset. Not a good idea.
Main takeaways.
Getting into good scoring positions is a repeatable skill. How frequently you convert those chances is much more prone to randomness.
How does it (broadly) apply to Stoke 2022/23.
We've scored 26 non OG goals from around 32 xG. Our on target xG is 27.5 xG. Short conclusion. We're not making the most of our location based chances (our on target xG is below our xG). And when we are on target the opposing keepers are making some good saves (our on target xG is above our actual goals scored).
At the other end of the pitch.
Our xG allowed is 32 and our goals allowed is 32 (discounting OG's).
But, our on target xG allowed is only 25.
So likely scenario is that the opposing strikers aren't getting the ball on target as well as they might (perhaps credit our defenders for that), but when they do get the ball on target our keeper(s) aren't making a great fist at saving them even with an average level competence. We've had the odd, consistently awful shot stopper of late.
|
|
|
Post by tachyon on Dec 28, 2022 13:21:58 GMT
Brown's instincts were spot on for the far post header. First he used his smarts to find space at the far post. Then he used his strength to get his header, not only on target, but in the sweet spot where goals are scored more frequently than other placements. Here's the outcome of every on target, far post header in the Championship since 2014. Attachment DeletedSticking the ball towards the near post, as Brown did is the best strategy. 65% of far post headers don't trouble the keeper. They either miss the target, get blocked by a defender or hit the woodwork. Aiming centrally is the poorest strategy. They are easier to save or clear. Going back from where the cross came from are difficult to save, but they lead to more attempts that totally miss the target. Brown turned a 0.36 xG chance into a 0.78 on target chance. The keeper pulled off an excellent save, but it reiterates how good an attacking header of the ball Brown is
|
|
|
Post by tachyon on Dec 22, 2022 16:40:35 GMT
tachyon do you have any stats which show our win rate with and without Souttar over the last few seasons? Or better still win/lose/draw and average goals conceded/scored and cleansheet % Stoke's goal difference was 0.02 per 90 better with Souttar than without last season. So virtually no difference. With/without stats are popular, but often hugely flawed. They work fine in the NBA where lineups change within the game quite frequently and scoring is relentless. They don't really work in football, where scoring is rare and lineups tend to be more consistent. Souttar's obviously very good, but I wouldn't be using our record prior to his 2022/23 return and in the three games after as a huge piece of evidence.
|
|
|
Post by tachyon on Dec 21, 2022 10:13:25 GMT
Currently 5% chance that Stoke go down. Their median final position is 14th with 61 points.
Most likely points won by the 22nd team is 50 (18% of simulations).
Range for the final relegated team is a low of 41 points and a high of 57.
Thanks tachyon. So if I read all this right we have a 5% (1 in 20) chance of going down and a 10% (1 in 10) chance of making the playoffs which equates to a 2.5% (1 in 40) chance of going up. Basically in all probability we are staying right where we are. Yes that's a fair assessment. Around two in every 1000 sims we come second, but that doesn't really move the promotion dial once you round up the numbers :-)
|
|
|
Post by tachyon on Dec 21, 2022 8:52:56 GMT
What about staying up? I'm thinking 50 someone will go down with 50 points Currently 5% chance that Stoke go down. Their median final position is 14th with 61 points.
Most likely points won by the 22nd team is 50 (18% of simulations).
Range for the final relegated team is a low of 41 points and a high of 57.
|
|
|
Post by tachyon on Dec 20, 2022 17:52:11 GMT
Points needed to finish 6th to 3rd (playoffs) depends on;
1) Current distribution of points already won by each team. 2) Distribution of team quality (however you chose to measure it). 3) Remaining fixtures and venues.
Also a reminder that to finish 6th, all you need to do is have more or the same points/better goal difference than the team who finishes 7th (with a few caveats).
Currently, but subject to change, there's a 9% chance the 6th place team will need to beat just the 66 points won by the 7th placed one.
It's a 17% chance they need to beat 67 points.
It's a 22% chance they need to beat 68 points. Most likely current outcome.
It's a 17% chance they need to beat 69 points.
It's a 16% chance they need to beat 70 points.
It's a 9% chance they need to beat 71 points.
It's a 4.5% chance they need to beat 72 points.
Most likely scenario is you'll need 68 points and a better GD to make the playoffs.
Stoke currently make the playoffs 10% of the time.
|
|
|
Post by tachyon on Nov 21, 2022 10:48:01 GMT
|
|
|
Post by tachyon on Nov 10, 2022 19:09:37 GMT
We were offside once on Tuesday in the play by play.
Delap in the 43rd minute. The flag was against him, not Gayle.
|
|
|
Post by tachyon on Nov 2, 2022 15:12:30 GMT
tachyon do we have a graph which shows xG over 2022 so far? Be interested to see how we are improving. Main viz is the last 10 games. Blue is xG created, Orange is xG allowed. Mini viz along the bottom is the last 40 games. So that takes in the 2022 calendar year. Stuck in a rut.
|
|
|
Post by tachyon on Nov 2, 2022 9:57:17 GMT
I'll do the defence in a couple of weeks. Giving the shot stopping a bit of a break :-)
|
|
|
Post by tachyon on Oct 12, 2022 13:09:02 GMT
That all makes sense, but Gawa’s posted stats for the Sheff Utd game give Wilmot’s XG as 0.6 and Delap’s as 0.5. I don’t think either of them had another shot, so safe to conclude that those are the XGs for their two goals. Wilmot had two attempts. The goal (0.49 xG) & the weird sliced shot in the first half following a corner (0.05 xG). Attachment DeletedBen (experimental361) does the xG that Gawa originally quoted where BW's two shots > LD's one. Ben's xG stuff is for the Press Association, I think he models xG by scraping match reports and using key words from that source (header/shot/penalty area/six yard box, that kind of thing).
|
|
|
Post by tachyon on Oct 12, 2022 9:25:18 GMT
Both were big chances, but Delap's had a bigger pre shot xG than Wilmot, not smaller. View AttachmentBarring 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). What do you mean by failed to register an xG. Does that chance not get included in any metrics then? His goal registers as an xG. The near post cross just prior to his goal doesn't because he didn't get a touch. If he had it would probably have been a 0.4 ish xG. The non xG chance does register on a non shot xG model for the player who crossed it (can't remember who), both for carrying the ball into a dangerous area and crossing it into one. That's another level of analysis.
|
|