This is more how Ozil came to be used by germany - arguably at his midfield peak in 2016 I think his performance vs France in tandem with Kroos was out of this world - but in a game Germany actually lost. Sadly germany lacked any decent 9 by that stage
Look, I’m not saying very improvement we see is down to Emery. But he has changed the way we play. That’s clear from watching the games and the stats. This has helped some players and hurt others. I think it’s fair to say we have more support for our CBs with the TWO holding CM players vs the one under AW. Also, the pass back the keeper seems encouraged, adding another outlet. While this means perhaps fewer accumulated chances, it means we were within striking distance and rarely chasing a result in games except maybe the first one. That is to say, the game state generally has favored efficient and patient play. Iwobi is actually completing passes at a lower percent this year, but most people think he’s playing better. Judging by how Emery applauds an audacious attempt, I can only guess iwobi has been encouraged/instructed to attempt more risky passes/moves. Xhaka is auditioning for CM of the year whereas last season we were ready to send him packing. Also, watch the first several home games from the kick off—first pass is a long diagonal to try to get Auba behind. We never did stuff like that under AW. We were usually possession based. Yes, I know we are ahead of our xG. I wasn’t ignoring that, I was pointing out the shot selection is careful so far. What annoys me about the Orbinho posts is that the insinuation is that we have been lucky this season (or unlucky under AW) simply because we’ve scored on fewer chances/xG. This is ridiculous. A shot is not a coin flip! It is not a random event. It can be (and has been for us) a carefully crafted event. Some players and teams will be better or worse at this craft. Over time the models will update to reflect your performance or your performance will regress to the model, but that doesn’t mean any deviation from is down to luck. Also, while the efficiency may regress to the model, it doesn’t mean the results will change. When David De Gea stops more goals than expected I don’t see people saying he is lucky—I just watched him save United yet again from juve. If Messi has a game where he scores 4 goals on 4 shots I don’t see headlines saying he is lucky.
We're in agreement on your last point. Lacazette and Aubameyang are efficient strikers. Both have finished slighty more than 20% of their shots over their entire careers. It would be interesting to see whether or not their goals scored exceed their xG over the course of two or three seasons. But I'm still not comfortable with the fact that Arsenal are in the bottom half of the table in shots per game. It's beginning to look like Mourinho-lite.
It’s cool to be cautious or uncomfortable...I get it. But all the regression to the mean doomsday talk is a real bummer. As we’ve discussed before (at least I have) we’ve had plenty of strong attacks that just haven’t resulted in a shot. I personally think it’s either coaching or players being patient and waiting for a shot they are prepared to execute well. Lacas Liverpool goal is a good example. He could have taken the pass from iwobi and shot first time. XG probably would have been >0.3 or so. But Allison is out quick which xG models usually don’t account for. So real xG is probably much lower. Laca passes on this shot and circles back for a shot against the flow of defenders. It was a 0.1 xG shot. So he traded a 0.3 for a 0.1. We’ve done this a lot this season. Also, since you asked: Nice article. You may want to run your model on Laca though. While in France 2014-17 he had 22% shot conversion rate with 43.87 xG vs 56 scored (penalties removed). Nearly a 1.28 ratio and about the same as Messi over the same period (1.25). Via understat.— Tom Jones (@Arsenihilist) October 11, 2018 If you follow oh_that_crab on Twitter he tracks our passing (final third, through balls, etc) and assigns a passing value added score. By those metrics along with touches in the final third we are solidly in the top 4 on attack this season. So let’s just enjoy having a few players who are firing them in.
I wonder if we're creating chances that don't result in a shot. Cutbacks or level passes in the box. They are tough passes to make that create high quality chances.
Interesting about Lucas Torreira: https://statsbomb.com/2018/11/alvaro-morata-lucas-torreira-and-evaluating-imperfection/ Key quotes: The opposite phenomenon is going on with Torreira. Torreira looks like a strong signing for Arsenal. He’s done a great job at bringing stability to the middle of the pitch. He’s made midfield partner Granit Xhaka look less like a defensive liability, covered for the fullbacks and just generally been an all around calming and effective presence in Arsenal’s midfield. That’s a skill that’s especially salient for a team whose midfield has consistently gotten overrun for a generation. He’s also not a particularly creative passer. First, the good stuff. He’s ninth in the league in interceptions adjusted for possession with 1.96 and tackles adjusted for possession with 2.95. If your looking for Arsenal’s ball winner, it’s Torreira. These are all great and important and necessary skills for Arsenal to have in midfield. They need these things. They need them so much that it’s absolutely reasonable to overlook Torreira’s lack of creativity on the ball. But, that doesn’t mean that Torreira’s lack of creativity isn’t a real part of his game. He only notches 4.86 deep progression per 90, well down below the team’s creators in midfield. On the other hand, with a 90% completion rate he’s the most consistent passer on the team. Combine the two and what you have is a player who is very conservative on the ball. His most frequent passes were 11 to Shkodran Mustafi, 11 to Granit Xhaka (playing leftback, which admittedly makes everything weird) and 10 to Rob Holding. The week before against Leicester City, in a much better attacking performance, the pattern was the same with 15 to Mustafi, 12 to Xhaka and 11 to Bellerin. None of this is bad. There are lots of very successful defensive midfielders who profile this way. Win the ball back, settle possession with safe passes, be an option to rotate possession, and don’t make stupid mistakes. That’s the recipe for a very good player. It’s also somebody who needs to be surrounded with passing to succeed.
And as a result he scored us a goal! Which earned us 1 point, while taking 2 away from LFC. If he had tried that 0.3 xG shot, he surely wouldn't have scored (because of the keeper's quickness as you point out). So the judgment, smarts, then exquisite skill of a specific player made all the difference in this case. Something that stats struggle to capture.. I don't know if this was your intention, but it seems you highlighted why obsession with xG can get a bit silly at times. It has some value in the aggregate, but it seems like the new toy that everyone wants to play with... a bit too much? <awaits onslaught of hate>
The point of a player like him is to give the ball to the creative players which I feel like he does (Xhaka and ozil and passes wide) it is only his first half of a season too and we are in a system he was t playing in before (someone told me sampdoria played a diamond like Uruguay) U guys get way too bogged down in stats. Your trying to turn a game with constant variables into a shit bore fest like baseball (each event and movement is in order I.e. pitch then hit or strike) same as nfl Why are u guys obsessed with data? Football has so many variables. Look at lamelas chance vs city, does xg take into account tottenhams pitch was an nfl reduced bog? It really is making the conversation here stale, long walls of text copied and pasted from other people. How about we all write our own opinions, discuss the game and maybe cite a small amount of ‘data’ to prove an argument
There is room for both. Some people don’t need stats to see that we aren’t creating enough great chances. We are just converting well. Some see it. Others want to look at the stats. To each his own. If one of those stat types wants to look into something interesting, I wonder how many of our attacks come from defensive players compared to offensive. My eye says that it may be deceptively high and probably is the underlying problem, but maybe the stats would prove that theory wrong.
As you can see by the fact that I generally rely on stats other than expected goals, I may be a stats geek, but I'm not an expected goal geek. However, based on our discussion, I believe the expected goal model would be better if it included the player taking the shot as a factor, e.g. If Lacazette or Messi exceed the xG by a factor of 1.25 during the previous season then the expected goal number, based on the location of their shots should be multiplied by 1.25.
This was something that came up a lot when i was working in content marketing for an AI company back in the early days. In general stats does not seek to capture the unique situation. The whole point is to create a model. But then the trouble is there is always pressure for your model to be more and more specific so the models get more and more complex to try to provide the ultimate guidance for specific situations. e.g. should we lend money to this customer? Humans use semantic reasoning rather than stats. So we don't ask what happened the last 1000 times Auba shot the ball. We ask what happened in the situations like this that I can remember. The advantage the AI has is that it can remember what happened every time Auba shot the ball and every excruciating detail. So it can merge statistics and semantics. What happened on average from every time Auba shot from this location, but also what happened on the occasions most like this? This was called "No SQL" - basically there is no model. You look for patterns like humans do. Anyway this is way too anorak but when you say stats can't capture the chaos and beauty of the moment, you are describing the holy grail that IBM Watson and those guys are chasing Finally - lots of what we see in analytics is based on the past but the real money is in prediction.
Sigh. Here goes. You've earned it this time. Need to address this out of order. xG is just one stat dude, and as far as things go it's only a starting point in analyzing performance. Like I've said before, all it does is capture shot volume and shot locations, and it's useful as an average. The whole point of using stats is to factor all of this in because it typically averages out.
I seriously doubt Arsenal use xG for a lot: their data is probably much more granular and the analysis is much more robust. The problem with this is that the only people that have consistently exceeded xG models by 20% or more over multiple seasons are Lionel Messi and Lukas Podolski. So basically the best player ever and an idiot savant who was really good at scoring from the inside left channel with his left foot. That's it.
This. The real value is in figuring out how to analyze incomplete information. Analytics, or the process of drawing intelligent conclusions from reams of data, is worth billions.
It's really silly - we're all watching the same games, and most of us think that we aren't playing particularly well. In this case, the underlying data matches up with what everyone sees. Worth following Scott Willis on Twitter - he's done some interesting worth with concepts like Passing Value Added. Statsbomb had this thing called xG Chain a while back, but then figured out that it wasn't particularly useful.
Yeah, no surprise here. I pretty much know which people are going to have a go at me. Interesting that you ignore the bigman who was a tad more critical than me?
I started out designing reports for a living, before I went to law school. If you remember those articles in newspapers about how horrible labor tracking in Amazon's warehouses was, I rolled out probably 6-8 such implementations a year (and speaking as someone who understands what the articles talk about, those articles seem like awful exaggerations). You just have to be comfortable with data and numbers to realize they're an incredibly useful tool. That footie has so many variables means that data becomes more useful (and footie analytics really is in its infancy - baseball analytics modeled park and field effects and things like weather impacts over a decade ago): the number of variables makes it that much harder for someone to draw conclusions simply with their eyes, because you can't possibly see everything. It's also really useful for me when watching games, because I'm on my phone a third of the time, so I'm naturally going to not see things, and viewing a game on TV means you miss a ton of stuff off the ball anyway.
This is a WAY better reply to my my post than what I got from the other dude. It was informative, insightful, and perhaps most of all civil. +1
Working my way down the list. The issue I have with you (and bigman) is that you guys are essentially throwing stuff at the wall to find issues with models. That's not the point of modeling real world interactions: they're not supposed to be perfect replicas of reality, they're supposed to be useful approximations that we can look at in a pinch. The question becomes how accurate a model in terms of error rate and skew. Raising individual anecdotes tell us nothing about that.
No, some of us just think the endless talk of stats is OTT and has become monotonous. As a fan, stats have their place in the discussion for sure. We've all been referencing them for decades. But it's gotten out of balance. I guess fantasy football is a lot to blame for that trend? It's gotten to the point where many fans (around here) are not able to smell the roses, to share "individual anecdotes", to revel in a thing of beauty, which much of the time is more analog than digital. Pretty soon, instead of spending 1/3 of your time on your phone, you'll go to 1/2, and then why not just go the whole 9 yards and not watch the game at all? Just watch real-time stats. During a match, when I spend too much time in the MD thread or looking at BBC.com, I invariably regret it. It's a trap that's easy to fall into. And I don't even do FB or Twitter! I always end up saying to myself "stupid, you should've just watched the game more closely". In fact I said that to myself today, although I had some more legit reasons that pulled me away this afternoon.
Are you sure you want to be the arbiter of what an appropriate level of analytics discussion is? For those of us with young children and busy jobs, it's not necessarily practical to say "just watch the game". There's always something else going on - I can't remember the last time I gave an Arsenal game my undivided attention (maybe the FA Cup game I caught with Yoss at Brewhouse in January?). And even if you did, you still wouldn't see everything. I like analytics because they enhance my understanding of the game, and I see a lot more when I do pay attention to the games (like, for example, Iwobi's impact in the left half space). We all enjoy the game differently, and its silly to whine that people are talking too much about the different ways they perceive and are exposed to the game and the information they like.
The reason MebeSajid has a go at Northbank and not bigman is that he is afraid of what bigman represents: Bigman has consistently over-performed his xPost numbers, to the tune of a 27% GoodPost conversion, well above what his spelling and syntax numbers would suggest are possible. His ShitPost numbers are very high, probably highest on the forum, but he routinely converts ShitPosts into likes which other posters just aren't able to do at the same level. The current statistical posting models we have about posting seem to be broken by bigman's "idiot savant" levels of efficiency, and this scares Mebe. What if the models can never show why bigman is so good at what he does? Mebe has looked into the bigman void, and he sees--no equations coming back. He hears nothing except a brummie voice asking him who he'd rather shag: Eddie Howe or Hirving Lozano. The void has stared back into Mebe.