Team-by-team 2021/22 season preview series: What each quirky metric means?
Below are some quick explanatory notes on what’s the thinking behind the use of some of the advanced metrics you’ll be seeing in the previews itself, how could they be interpreted and what do they effectively mean.
I’m leaving out some straightforward ones (like fouls conceded in defensive third or blocked shots), but most will appear here with at least some sort of background I consider vital for you to be aware of when reading the stuff.
Please note that while I sort all players meeting the 900-minute threshold into 7 categories (goalkeepers, fullbacks, centre backs, wingers, defensive midfielders, attacking midfielders, centre forwards), some players such as Patrik Brandner have their playing time split between two positions depending on their usage as long as they meet the threshold in both (winger and centre forward in this case), while some players will only fall in one category despite being used elsewhere (I’ll pinpoint those cases in the text accompanying pizza charts). Also, some metrics are common for more positions, so I only elaborate on them once (usually where it’s more relevant).
Goalkeepers
I hope we can agree right off the bat that facing 5 shots from outside the box is not the same as facing 5 shots from the edge of the 6-yard box. That is exactly why I’m not using the most common metric — save percentage — to measure the quality of one goalkeeper’s performance. To separate the best shot stoppers from the worst, then, we’ll rely on a different set of metrics that shall better account for shot quality faced. First off, we’ll make use of Wyscout’s post-shot expected goals model, otherwise known as expected goals conceded (xGC) or xG² in some other circles too (11hacks are using this term), turned into the so-called prevented goals metric that follows the simple equation of CG-xCG, divided by the number of appearances (to get a 90-min average).
Now, what’s post-shot xG (xGC, or for strikers xGS — expected goals scored):
In a nutshell, post-shot xG corrects for many common complaints against the more traditionally used pre-shot xG which only takes into account the location and the type of the finish in question (whether it followed a cross or a pass, whether it was headed or not, and so on). This sort of xG model really only tells us something about the raw quality of chances created, fullstop, whereas many would lament that it doesn’t control for who’s finishing and how well. That’s exactly what post-shot xG brings to the table: it crucially accounts for the placement and power of the shot, only working with shots put on target.
Hence, the value of post-shot xG is often at least a double the pre-shot xG value. For instance: the common expectation, I find, is that a partial 1v1 with the goalkeeper finished from the top of the box (so not quite a breakaway, but a clean look at the goal) comes close to 1 goal expected — because that’s what we expect if we’re cheering the striker on, right? — when in reality, it usually falls somewhere inbetween 0,15 and 0,3 in terms of pre-shot xG. However, if the striker finishes in line with our expectation, the post-shot xG value can jump all the way to 0,6–0,75 (ie. it’d result in a goal in up to 75% of cases), which sounds more like it, doesn’t it? Basically, the only time pre-shot xG > post-shot xG is when the finishing player scoffs his shot from a good position.
Now, my approach to prevented goals is very straightforward, but it may not be the best (so I’ll look to tweak it in the future). As it stands, I account for penalties (always carrying xGC value of 0,81) as well as own goals (0 xG/C) which admittedly doesn’t feel right, but penalties are also a noted way how to improve your numbers (right, Florin?) while own goals are sometimes down to goalkeepers’ mistakes, so it also wouldn’t feel right to automatically absolve them of all blame. Again, I’ll look to tweak the model a bit, but only a little; now it still works because the rules are the same for everyone and that’s cool.
If this bothers you — meaning you’re likely not a fan of arbitrary metrics and would like to take a bit more ad hoc approach to stuff — then kindly look away, because the following two metrics are of the same nature.
Using the post-shot xGC, I’ve also designed simple metrics of soft goals conceded and high-danger shots saved with arbitrary cut-offs for what is a soft goal and what is a high-danger chance. The result of my trying: a high-danger chance would have at least 50% chance of going in based on the model (ie. carrying a value of at least 0,5 xGC) while a soft goal would carry a value of 0,3 xGC or less, which from my experience both pass the eye test as suitable thresholds, even if the 0,2 xGC difference seems tiny (it’s not really).
Now, onto the limits: obviously, there’s a vast difference between shots of 0,05 and 0,29 xGC values (and hopefully, with more time next summer, I’ll look to distinguish between “soft” goals and “fucking terrible” goals to concede), much like there’s virtually no difference between shots of 0,29 and 0,31 xGC values. But we’ll have to live with that. This is both the drawback and the gain of the arbitrary cut-offs, in fact, because what we are trying to achieve here is to set an objective framework whereby to judge a goalkeeper; and there’s naturally a great added value in that, because a subjective framework is… well, us — our memory, our experience, our opinion. We don’t need stats to replace that, do we? We could use stats complementing/correcting it, though.
We’ll have more fun with xGC as we proceed, with other metrics I use being overperformed xGC, expressed as a % of all games played a majority of (ie. simply looking at how often the said GK concedes less goals than expected), and average xGC value of a shot let in (this is where penalties and own goals don’t skew the outcome, finally; I’m simply taking all non-penalty shots that turned into actual conceded goals and averaging their xGC values).
To close out the shot stopping section, we return to a save percentage — of sorts. Here, I’m only working with non-penalty shots from inside the box which should, in theory, tell us something about a goalkeeper’s positioning as well as reflexes. Sometimes, when particularly relevant/interesting, I’ll throw in the exact opposite (save percentage for shots from outside the box), or even the odd high-danger chance save percentage (because, again, the likes of Kolář will otherwise never perform truly greatly simply for facing less high-danger shots), but I don’t have this data for all the goalkeepers out there (yet), sadly.
Now, onto the more straightforward metrics: long pass or pass to final third accuracy is self-explanatory, while pizza charts also feature how big a portion of all passes played by the said goalkeeper is taken up by those on long distance (ie. travelling more than 25 meters). This metric is just for your information, mind; I don’t factor it into the overall percentile as I realize it’s more a product of team’s tactics/dominance than the keeper’s own doing.
A loss leading to a shot is an interesting metric in that it accounts for both misplaced passes but also lost aerial or ground duels. The follow-up shot needs to occur inside the next 20 seconds, otherwise it’s rendered a “harmless” loss, which is yet another case of a needed arbitrary cut-off. My bigger problem with this: what if the loss is truly staggering and awful, yet the opposition is even worse and passes its way to a corner flag instead of shooting? This stuff doesn’t show up here, only the punished — but oh well.
Finally, a set of metrics measuring one goalkeeper’s proactivity. Interceptions and left line to claim/punch sound like they overlap a fair bit, and they do indeed, but not too greatly. While the latter only accounts for instances where goalkeeper comes (way) out of the goal to deal with a cross, the former also includes balls played outside the box or stopped passes cutting across the 6-yard box. Finally, with aerial duels, I’m only interested in success rate to not automatically punish goalkeepers who tend to be more cautious. Simply put, when you do engage in an aerial duel, you’d better be 100% sure about it and 100% successful at it (hint: almost half of regular starters were not 100%).
Centre backs
While my approach to goalkeepers is as unrefined as it is non-traditional, hence the length of the previous section, this one is going to be much shorter.
A couple of important notes on metrics used as part of my CB model, however:
- A defensive duel is any duel undergone vs opposition in possession excluding aerials who have their own category (and I only consider those undergone in the defender’s own penalty area for my model’s purposes);
- A loose ball duel occurs when no team controls the ball and would be considered “in possession” (these are typically the crucial “second balls”);
- I limit the traditionally used clearances to those made in centre of the box, which roughly corresponds with the area number 2 on this scheme;
- I also track those clearances that are “controlled”, ie. result in the team in question retaining possession of the ball post-clearance; this is just a bit of fun, really, because it’s often random — and in many situations conceding a corner kick/throw-in is the best-case scenario anyway — but I also believe it can tell you something about one’s awareness/peripheral vision;
- Finally, one specialty: your hipster friend’s favourite “half-spaces”. You always find them here, and exploiting them is a noted weapon, so I’ve decided to track every accurate forward pass played into half spaces;
Fullbacks
An important skill of every modern fullback — you’ll be told by your favourite analyst — is to be an efficient ball progressor, so the million-dollar question is: how long does a run/pass need to be for it to be considered “progressive”?
It depends on where it starts and ends, in fact. If it occurs fully within own half, it has to move the ball at least 30 meters closer to the opponent’s goal; if it crosses the half-line (ie. starts and ends in different halves of the pitch), only 15 meters suffice; and if it already starts in the attacking half, a mere 10 meters are considered enough to make a difference for this purpose.
Important thing to note with the last type of a progressive pass/run: it doesn’t only work vertically but also horizontally, so even when a player takes the ball near the corner flag and dribbles with it all the way to the near post, he’s technically fullfilling the criteria and thus completing a progressive run.
Notes on some other non-traditional metrics used as part of my FB model:
- For fullbacks specially, I don’t track passes into half spaces but rather the so-called stretch passes which are defined by here as any accurate pass that either goes in behind the line (forcing defence to re-shuffle or adjust positionally at least slightly), or across the field towards the opposite flank (above all forcing midfield to do the same). Both types count as equal;
- Effectively forming one part of a half-space, the so-called cutback zone is another noted dangerous area of the pitch where you’d ideally want your modern fullback to function and do damage. So, I’ve decided to additionally track all accurate cutbacks in (those) dangerous areas.
Defensive midfielders
Interceptions have become one of the household stats, but the trouble with them is the good ol’ context I’ve mentioned above. If your team always controls the flow of the game, ie. retaining possession, how the hell are you supposed to intercept anything? Fortunately, we do have a way how to control for context in this particular instance, through the so-called possession-adjusted interceptions metric. What it effectively means that Wyscout people do the math pretending the possession game was equal in every matchup: so if the ball is, on average, in play only 60 minutes per game (I know, depressing), the stat is normalized for 30 minutes of possession.
Only that way, some extremely smart-moving players like Lukáš Kalvach are done proper justice. While in a normal world, Kalvach intercepted an opponent’s action an uninspiring 5,76 times per game, in a possession-adjusted world he was a beast — intervening an elite 9,39 times per game.
Another metric to measure one’s positioning qualities: progressive passes allowed. This one is a bit tricky to get hold of, because it simply tells you how many progressive passes were completed in the said player’s zone (ie. central areas here), so it’s always a collective effort, but you’d expect your holding midfielder to be responsible for cohesion and organization of your team down the middle, which is why this metric shows up on this board in particular.
Attacking midfielders
A little disclaimer to get us underway on a spicy note: I hate key passes.
Their definition is vague and subjective (a “clear goal-scoring opportunity”, as Wyscout puts it, is a definite grey area we strive to avoid) and their tagging by Wyscout in particular is suspect (by their own definition, it’s supposed to be an action that creates a chance “a teammate in turn fails to convert”, yet I’ve found actual assists sprinkled all over), so I will be consciously avoiding this term and this ready-made metric. After all, you can be a threat without your teammate pulling the trigger, which is why I introduce you to so-called deep completions, or simply passes/cross completed inside/into penalty area and its immediate surroundings. To be precise, any connections made in this zone:
Nevertheless, we won’t be taking shot attempts out of the equation altogether, which is also why I consider shot assists, but only those made from inside the box or just outside of it, at the very top of the penalty area. I think that — along with expected assists that always correspond with the xG value of a shot created — just about covers the set-up qualities of one player.
But for an attacking midfielder to top my charts, he’s got to have a more complex influence. He must know how to attack the box, too (through successful offensive duels in final third or successful actions inside the box), he obviously needs to know how to progress the ball effectively, his foul differential shouldn’t be a minus-anything and while he’s expected to lose the ball more than he recovers it, he should ideally not drop too far below 0.
And then there’s the proverbial X-factor; here measured by the number of dangerous set pieces won (any foul won in the attacking third counts equal, with no special bonus points for penalties because, hey, every tackle/push is judged by the same measuring stick regardless of where it occurs, right? Right…?!), and accurate through balls along with completed smart passes which are defined as “creative and penetrative passes that attempt to break the opposition’s defensive lines to gain a significant advantage in attack”.
‘Nuff said.
Wingers
There are many types of wingers and none of them is any worse or any better than the other, so there was a whole lot of ground to cover for my model. Ultimately, there are metrics designed to suit straight-line runners — like accelerations with the ball (at least 10 meters long per Wyscout) — but there are also metrics designed to suit the more pondering, playmaking types — like accurate crossfield passes (simply any successful attempt to switch it up, to make the opponent shift to the other side and gain advantage, tracked by myself). At the end of the day, though, you likely want your winger to crack the penalty area one way or another, which is why we have successful penalty area entries — be them via a run, a pass, or a cross, anything will do as long as we are talking about open play (set pieces factor in as part of xA).
My favourite thing is to measure a winger’s discipline — let’s call it the “Potočný clause”, shall we? Here I’m not only interested in the above-mentioned foul differential, but also in how often the player in question pulls the trigger from outside the box (rarely the perfect option). I express this via % of all non-penalty shots taken, and the good thing about this metric is that it speaks to one’s discipline but also willingness/ability to attack the box.
Centre forwards
Together with goalkeepers, this is where I let run my imagination riot most.
For all outfield positions, I create five groups of “neighbourhood metrics” that speak to one or two aspects/phases of one player’s game (eg. attacking the box, proactivity and discipline, build-up contribution etc.). For strikers more than any other outfield positions, these groups (well, 3/5 of them) cover various types of centre forwards — again, with none of them being superior.
For the proverbial fox in the box, there’s the specially extracted xG of all non-penalty shots from inside the box only (combined and converted to a per-90 basis), then a number of shots put on target from centre of the box (in text sometimes accompanied by success rate which takes away the sample factor), and finally % of actions in the opponent’s box successfully executed.
For a sharp sniper (who may or may not be a fox in the box, too), there’s the number of big chances missed (those would be carrying at least a value of 0,25 xG) as well as the good old fashioned shooting percentage (% of all shots put on target) along with a true new invention: quality added to non-penalty finish. This last metric once again works with the post-shot xG; specifically it looks at how does a sum of xGS (expected goals scored) compare to a sum of pre-shot xG which you encounter the most.
Ideally — and most commonly — an individual xGS should beat xG comfortably. That’d mean the striker A) has a fine shooting technique so his shots tend to have a decent chance of beating the keeper, B) doesn’t miss the target too often, because a shot off target doesn’t get assigned any xGS value.
Now, you could do numerous fun exercises with this, getting different perspectives. You could simply take away xG from xGS (and perhaps divide it by the number of appearances, as per usual). Or you could take away actual goals from xGS to pretty much find out which striker has suffered the most from quality/above-average goalkeeping throughout the season. That’s what I sometimes look at too, but for my model, a % of quality added/detracted works the best. Simply take both sums and see by how much the xGS one is greater/smaller than the xG one, which limits the damn sample size factor.
Finally, for a target man whose main calling card is his hold-up play, I’ve come returned to the above-mentioned metrics like successful penalty area entries, xA or deep completed passes, but also I want to see how accurate are the centre forward’s passes into the final third. Generally speaking, I don’t care how often and how deep he drops (or how often he receives the ball; because that’s not under his control), but when he does, I want him to spread the ball well.
Just like with wingers, I’m also interested in striker’s discipline — here expressed by how often he’s caught offside but also how often he resorts to low xG shots. Those would carry a value of 0,03 xG at most, which basically gives you only those attempts taken from the perimeter/way outside the box.
