When it comes to defensive stats, none of them mean anything. How many tackles your team makes doesn’t correlate with shots allowed. How many interceptions, how many fouls your teams make, and how many passes they block. All basically have zero correlation to shots allowed.
Even how many “defensive actions” a team takes. No matter how I add them up or try to make them fancy with location information, they don’t correlate to shots allowed. You know what does? Clearances. Oh yeah, and possession. The more possession you have, the fewer shots. So, tackles are nice, interceptions and blocks are cool, but doing more of them doesn’t mean you will concede fewer shots.
When I add up Tackles (won), Interceptions, and Passes Blocked (TIPs) and adjust for possession (specifically for possession allowed), I get a slight correlation.
The problem with Tackles, Interceptions, and Passes blocked is that there is very little variation among teams. Even adding all three together, the variation is just 11 events per game between the most “active” defensive team (Huddersfield last term) and the least (Man City). And, and I know I have said this many times before but it bears repeating, there is NO CORRELATION between possession and the number of defensive actions a team takes. None. It doesn’t exist. Less than 0.07.
So, really, possession is the only thing that seems to correlate with number of shots against. The better your possession, the more likely you are to have a good defense.
I also need to point out that tackles high up the pitch are not defense. They correlate with offense. As much as we love a high pressing team and as much as we love a player sliding in to win the ball away, that isn’t really defense. Defense begins and ends with your team having the ball.
As much as we all like a good tackle, it’s much more important to control the ball in possession. Which I think you all already knew.
So, defensive actions don’t really correlate with fewer shots. However, some teams are more or less active with their defending, which is a stylistic thing. If I adjust for possession, here are the most active defensive teams:
The chart above is this season. Man City are the most actively defensive team when they are out of possession in the Premier League. Southampton are 2nd and Liverpool are right in the middle of the pack this season. Arsenal are 7th this season with a value of 78, which is a slight change from last season where they had a value of 71. I use “value” because this isn’t like “tackles per minute out of possession” or anything, it’s just a calculation. Like xG is just a calculation and doesn’t really represent real goals.
Last season’s chart is above. There shouldn’t be any surprises here: Liverpool, we know from all of the data, have made fewer tackles (etc) than last year. This is part of a strategy (I think) by Klopp to help preserve his team.
Arsenal have moved up a bit this year. Opposition possession stats are almost exactly the same (46%) but we are attempting slightly more tackles and interceptions.
What I take away from all of this is that possession is significantly more important both to the offense and defense of the team. The top possession team is City with 64%. But just below them are Chelsea, United, and Liverpool all of which have roughly 58% of the possession. The two teams outside of that group are Arsenal and Tottenham, who only dominate about 54% of the ball. And then below them are all of the other teams who basically don’t dominate possession at all. The only outlier is Southampton who actually control 51% of possession (so far) this season.
I looked at this data set to see how Emery has changed Arsenal from Wenger’s annus horribilis and he has changed the team, slightly: T.I.P.s are up over last year. But the real leap forward isn’t going to be when Xhaka starts trying to head the ball off another player’s foot (pashun tackle) but when the team plays with the same level of commitment in defense and also controls possession better.
P.S. because someone will ask: xGa is a worse fit in some charts and better in others. For example, the R^2 for “TIPs adjusted for possession” with xGA is 0.266 but xGa and possession is a better fit with an R^2 of 0.481 whereas with shots and possession the R^2 is just .407. I still wouldn’t call these “good” fits but when it comes to any defensive stats, the key is always possession.
Wrong! Poseidon is the best defense!
Was he possessed?
Other than W/L/D and goals scored/conceded what is – forced to pick – the one stat which you want improved?
Mine is Lichsteiner running a 40.4/40. Yes, running the 40 in 40.4 seconds.
Possession.
Explain. We used to pass it to death. We were a possession based pass and move team for years under Wengerball and we came away largely empty of trophies (yes, I loved the FA Cup in May, but…)
I do wonder if some alliance between possession and pressing would be worth looking into, rather than one or the other.
One naturally feeds into the other, no?
Possession is nine-tenths of the law
I was waiting for someone to say that 🙂
It is one of the most uttered lines from Clive Tyldsley, a particularly annoying football commentator for ITV.
Good analysis. Logical, I suppose that one way to defend is to stop your opponent getting the ball to do damage.
I know. That commentary made it into one of the EA Sports’ Fifa games, which made me think at the time that this was a common footballing usage.
Hey, on Ramsey, I came across this graphic.
https://twitter.com/Reunewal/status/1082275740228505600
He is a better player than many gooners give him credit for. Unless we do some world class scouting, Ramsey’s skillset isn’t going to be easily or cheaply replaced.
Ever Banega doesn’t look so hot, does he?
And glad to see Fabregas justifying my undying love.
There is passive possession which is sideways passing and there’s incisive possession that starves the ball, releaves pressure and can build attacks. City are the masters at it through Guardiola.
Liverpool prefer to be more incisive with their style of play.
Arsenal I think want to be like Liverpool but lack the personnel. We lack players to set the tempo as Unai wants. To be protagonists.
I feel his squad will need two years to be moulded but without transfer budget backing for crucial signings I presume I feel he won’t be able to ever put his stamp on the team as Guardiola,klopp, pochettino, Eddie Howe and Marcus Silva have done for theirs.
Love watching wolves. Quite entertaining.
Agility/mobility/acceleration is key in the system that unai wants to implement. Our first 13-14players on the list do not all meet the criteria.
Hope he moulds young players or cheap buys to punch above their weight or uses a variety of tweaks for maximum output.
“Agility/mobility/acceleration is key in the system that unai wants to implement. Our first 13-14players on the list do not all meet the criteria.”
I agree, I think this is the future of the sport. You see it in the NBA and NFL too, teams who get younger, longer and faster thrive. Arsenal were built on Wenger’s technique first ethos and will take time to reconfigure in this age of Kylian Mbappe.
Wenger’s ball was to always have more possessions in the other team’s final third. … and to always win it back asap . He was always way ahead tactically footballing wise. …
You’ve given us a lot of cool data to chew on here.
My first thought as a salty, crusty correlations sceptic is this: http://tylervigen.com/spurious-correlations
I’m not a statistician but I’ve learned enough to be mistrustful of even the most straightforward looking correlations, particularly (apologies to Tim) when conclusions are represented in ALL CAPS, as in, this cannot exist because R^2 = 0. Not true, from what I know, because you can change an R^2 pretty easily by fitting your model differently, for example by eliminating outliers. Also, simply looking at a different range of data might give a totally different answer. I don’t expect a high level of statistical rigor on a football blog but I would discourage jumping to broad conclusions on the basis of this kind of math.
My second thought is that the possession to shots correlation is blindingly obvious and probably exactly correct, but it’s way deeper than simply having the ball is good. Yes, if you completely deprive the other team of the ball, they can’t shoot, but that’s really hard to do for 90 minutes. And if you only allow two shots but they are both big chances, was it still worth all the sterile possession and safe passing between CB’s? What if all your possession doesn’t actually create any chances for your team? That, in very basic terms, was Arsenal’s problem for about a decade under Wenger: lots of sterile possession without incision at one end, giving up few but extremely high quality chances at the other. Sure, Pep’s Barcelona could keep the ball all day and their opponents hardly ever shot the ball, but that doesn’t help teams who don’t employ about 5 of the world’s best technicians in their team. And what about a team like Atletico who only possess the ball 48.5% of the time and allow 11.4 shots per game but still have lowest xGA in the league (also known as peak Mourinho ball)? Maybe they are just as unique as Barcelona in their own way but it does add to the complexity of the question.
Lastly, I know the correlation with xGA isn’t as good in your data, but shot quality has to matter in this discussion. I’d much rather let a team get 15 shots from outside the penalty area than even one from inside the 6. To me xGA is a more meaningful variable to measure than shots when looking at what types of actions on the pitch contribute to defensive efficiency.
I think it’s funny that you start with “spurious correlations”, suggest removing outliers, and then move on to using the outliers as an example of why my correlations are spurious.
As I said in the PS, the xGa is actually a STRONGER CORRELATION TO POSSESSION than the shots data.
Yes, in one or two matches a season, teams who play with possession will be caught out – teams will also be caught not playing the ball in to the box fast enough and there will be 0-0 draws. I’ve literally been there and seent that. And yes, of course, there are Leicesters and Atletico Madrid. Outliers don’t disprove anything. The point of stats is that over the long haul, the possession stats will pay off. The other point, the bigger one here, is that tackles are irrelevant. Interceptions are meaningless. If you want to look at any stat to see how poorly a defense performed, look at clearances.
Hm, I don’t think I made myself very clear. I am trying to contribute to the discussion as opposed to making you mad (you sound angry, sorry if that’s a false presumption).
The first point I tried to make is that I don’t trust r^2, whether it’s high or low, because it’s too easy to refit the model and so much depends on the nature of the variables and their context. Outliers are just one example of something you can easily modify to change an r^2. Basically, I don’t like to see absolute conclusions being drawn based on r^2 and this view is not unique to you or your analysis in any way. The second point is that I do actually think the correlation you showed is real because it makes so much logical sense. I just think xGA would be a better dependent variable to track.
The last thing about Atletico is simply to give an example of a team that does cede possession but still defends at a high level. I have no idea if they are a statistical outlier or not but they’re not the only ones to do so; most of Italian football is played this way, I believe, and most PL teams play this way against higher echelon opposition. So the question to me is not WHETHER possession results in less shots, the question is WHEN is possession football “worth it”? In other words, when does it benefit a team to control the ball as opposed to controlling the space? Clearly, everyone would love to play like City or Barcelona and win quadruples but I can easily recall dynasties built on the opposite approach (AC Milan in early 2000’s) or a more mixed style (Zizou’s Real Madrid perhaps). So to me it’s not as straightforward as the correlation suggests: possess the ball more and therefore defend better. If you have Xavi, Iniesta, Busquets and Messi then yes, hog the ball by all means, that’s your strength. But what happens if you have Ozil, Ramsey and Mkhitaryan and you try to play like that? You might achieve 75% of that mastery, but is that enough? Enough for what? Conversely, if you have Maldini, Gattuso, Cannavaro et al then it definitely makes more sense to control the space and then slice people open with Shevchenko at the other end. But what happens if you try to do the same with inferior players like Smalling, Matic and Ashley Young? That’s the real mystery of football, to get the in-between or lower echelon cases right. Does this correlation help us solve that? Genuine question.
Possession ….. umm ….. I cannot remember how often I saw Arsenal having (say) 70% possession and 25 fruitless shots at goal — yes, that is not a contradiction in terms, both are possible within the time frame of a game — I am sure many of you will have endured watching games like that — and then still lost to a team with little possession and and then winning, with one fortunate, speculative scoring shot on our goal goal.
Stats are great — but football, indeed all sports, are about the human input, the determination not to give up, the spirit to fight back under intense pressure, to guts it out, and keep going until the final whistle.
That’s why I love football, to yell support, or to pump my fist in ecstasy, or to become engulfed in a black cloud of disappointment — and as for the stats — not so much. 😏
Dr
WhoGooner,I think we are coming to similar conclusions but from deferent perspectives.
Statistics have the ability tome people’s eyes roll, and for them to nod off to sleep, so I am not going to get into more than basics by saying a definition of statistics is the collection, presentation, analysis and interpretation of the data.
In my experience, it is amazing how often the ‘interpretation’ of the data is not given sufficient weight, in a given context, and becomes subjective, and possibly irrelevant, so that the conclusions drawn are questionable.
Not that I am suggesting for a moment that the data interpretation has not been given a microscopic examination in todays Post.🥺
‘deferent’? what the hell is that? ‘
different’ – that’s the word — you, yes, you – middle finger of the right hand – wake up!!
Reading the scandal sports rags this afternoon, I note that Rambo will soon be gracing the Italian team Juventus, at a rumoured £140k per week, (£7.3m p.a.).
Has anyone a theory as to how that can be a sensible strategy from Arsenal.
OK, it is rumoured that his current salary £100k per week (?) or £5.2m p.a. is causing problems under the FFP rules, altho now that Arsenal have been returned to being a private limited company, under the Companies Act [UK], I am not sure how any of us can know that.
He is a quality player aged 28 y.o. and at his prime, he must have a transfer value of (say) £40m. Any incoming replacement would presumably cost half that, and would expect a salary of (say) £50k per week (£2.5m p.a.) makes the math a little head scratching/.
Missed transfer out value – £40m; transfer value in – £20m; less annual salary reduction (£5.2m – £2.5m) = £2.7m p.a.
Net transfer costs to club: £60m – less annual reduced salary costs ££2.7m = £57.3m.
Any ideas?
Here’s a clue as a starter, as everyone seems to be working – except me – there is a sneakily misleading bit of maths above!! Can’t help it!😎