Fortuna:Liga 2022/23 Team Preview Series: Behind the quirky metrics (2.0)

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 themselves, how could they be interpreted and what do they effectively mean. You’ll notice I’ve done a lot of overthinking and thinking over.

Beware, this may all look familiar at some places, but it is indeed a thoroughly updated version of the initial introductory piece I did in 2021. I’ve expanded the range of metrics considered for goalkeepers, only a third of the original metrics considered for attacking midfielders (for instance) have survived, etc.

Now, I’m not going to do this every year. That being said, however, a large-scale overhaul was kind of the plan all along — if you read the 2021 piece, I noted at various spot that I’m going to return to them and re-consider once I have more time. Last year was a bit of a hot-needle pilot version of everything due to my 2020 Euro commitments only leaving me with a precious little window to work with. This year I’ve actually had the time to pause and think.

This means I can’t really compare 2021 and 2022 pizza charts. It’s not that the previous version of my model would suddenly be horrible; I still stand by it to a large extent. It’s mostly practical — you can’t compare even slightly tweaked metrics (like switching between “per 90 mins” and “% success rate” bases).

Next year, the goal is to do as little re-thinking as possible to avoid these limitations and start tracking year-over-year dips and improvements properly.

Anyway, let’s get to it. Just like last year, I’m leaving out some straightforward metrics (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 real stuff. I’ll point out new modifications, too.

Goalkeepers

The one position that has undergone the greatest revolution. A much-needed one. First of all, I didn’t like the GK pizza chart being by far the least populated. While others included 15-16 metrics, this one only had 11.

I was not having it.

This year, all radars shall feature exactly 16 metrics, hopefully divided into 4 groups going by what they are supposed to express together. (If we can find a fourth colour, that is.) Some mini-models are actually fed by up to 18 metrics, giving us the overall percentile per role, but radars will be consistent.

The four groupings of metrics for goalkeepers read as follows:

Shot-stopping qualities

• Prevented goals: You should be familiar with this one already. This metric incorporates what is commonly known as post-shot xG, usually shortened as xG² or xGC (expected goals conceded) — the younger of two xG methods that only considers shots on target (ie. shots that can actually turn into goals) and accounts for stuff like shot power and shot placement (ie. stuff the traditional xG method typically doesn’t control for). This post-shot xG then gets turned into the so-called “prevented goals” metric that follows the simple equation of expected goals conceded-minus-conceded goals, divided by the number of starts (to get a 90-minute average). Crucially, this year I’m not including penalties and don’t count own goals among goals conceded, giving the stat a fresh look that truly reflects one’s shot-stopping (whereas earlier it played in randomness’ favour more).
• High-danger saves: This stat has remained intact, and once again does include penalty saves for the benefit of the GK. Generally, any shot that would — per xGC — result in a goal scored in at least half of cases (value of 0,5 xGC) is potentially credited to the goalkeeper as a high-danger save. As ever with arbitrary cut-offs, you could easily argue there’s no difference between 0,49 and 0,51 shots (and you would be right) but we need an objective framework to complement our fundamentally subjective view of football and we wouldn’t gain anything by allowing for tolerance. (Then you could say there’s no difference between 0,44 and 0,46 shots etc.)
• High-danger situation save percentage: This metric effectively replaces the original “non-penalty saves from inside the box”, and for two reasons: 1) The previous metric is, fundamentally, a by-product of opportunity. If you’re moving behind a stellar defence, you’re inevitably going to rack up less high-danger saves, and it’s not necessarily your shortcoming. Three best goalkeepers in this particular area, after all, flashed their gloves for relegation candidates. That doesn’t render the stat meaningless, but we need to be more inclusive. 2) I had established at the beginning of my 2021 preface that not all shots are created equal, yet then I treated them as such by lumping both harmless shots from the perimeter and close-range shots from centre of the box together in one metric. Inconsistency. Bleh. Consider this a lesson learnt on my part. Thus, to limit the scope even further, I’m partially incorporating xGC — in turn completely transforming the playground. Think of this as a semi-scientific approach to the age-old “He doesn’t save anything extra” complaint about a mediocre goalkeeper.
• Steals: This metric effectively replaces the original “overperformed xGC” where I looked at how often (expressed as % of all starts) did a goalkeeper overperform the expectations. That metric was flawed for one major reason: I was actually docking games off the goalkeeper’s total when he didn’t face a single shot (ie. didn’t have anything to outperform, of course). Besides, the playing field wasn’t quite level — goalkeepers were getting the same amount of credit for saving 1,35 and 0,05 goals above expected, which is obviously a nonsense. Hence, here we only look at truly memorable performances— both in positive and negative sense of the word. A goalkeeper gets credited for a “steal” whenever he prevents at least 0,81 goals (the equivalent of one penalty saved — again, we need an arbitrary cut-off and this felt suitable), but also loses a point for any game he just about “threw away” by allowing at least 0,81 goals above expected. I don’t care if he actually stole/threw away the game in the end, mind, because a goalkeeper can’t tangibly influence what his strikers do upfront. Also, this is one of the very rare cases where I don’t convert a stat to per-90-minute basis, as it wouldn’t make much sense (or achieve anything).

Consistency and dependability

• Prevented goals from outside the box: I believe this is self-explanatory. For one last time, we reach for xGC values but only take the sum made up of shots taken from outside the box and deducting mid/long-range goals.
• Loss leading to shot: This 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 per Wyscout rules, otherwise it’s rendered a “harmless” loss, which is yet another case of a much-needed arbitrary cut-off. As I said last year already, my bigger problem with this metric is: 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? Let’s assume it’s down to the goalkeeper’s psychic powers.
• % share of preventable goals: This is where my subjective view comes to play, but since it’s me and only me considering things all the time (with my lens that could be challenged but remain the same), the subjectivity doesn’t cause too much harm. As for what it means: simply how big a portion of goals conceded I deemed to be “saveable”. It effectively replaces the original “soft goals conceded” metric where I used the arbitrary cut-off of 0,3 xGC value, which has however proven to be way too high over time. As I delved deeper into the actual tape in 2021/22, I often had to write to Wyscout and annoyingly ask them to re-consider their data input leading to some odd xGC values. They would typically stand by their model, which is totally fine by me, but… I’d rather rely on my own set of eyes in this case.
• Grave errors leading to goal: Here, I don’t mess around with some vague “preventable” and “saveable” adjectives. Nothing like “he could’ve done better on this one” suffices anymore — I only count the real fuck-ups of which one goalkeeper accumulated a maximum of four this season. Then the number of starts matter, of course. And just for the record, these grave errors are also part of the sum of the preventable goals — it’s just that not all preventable goals are a result of grave errors. It doesn’t work both ways.

Sweeping and distribution

• Success rate of actions far outside the box: Another new one, designed to measure sweeping qualities of a goalkeeper. By a “successful action” I mean anything from a pass while building from the back to a clearance while intercepting a counter-attack, as long as they keep/regain possession for the said team. By “far outside the box”, I more or less mean the area of the pitch below where a GK usually doesn’t operate. I don’t consider anything happening inside the attacking half, since that usually suggests desperation mode towards the end of a losing game. Not the point.

Yeah, I know; no wonder I delegate all graphic work to someone else…

• Long pass % / final third pass % / low misplaced pass: I lump all passing metrics together as there’s not much to add really. A long pass is “a ground pass longer than 45 meters or a high pass longer than 25 meters” per Wyscout Glossary, while not more than 5 passes to final third even get attempted per average game — it’s a really long kick, you know. Arguably the most intriguing stat is the last one where I account for any goalkeeper’s pass ending inside his own half, directed to opponent or past the sideline.

Damage control

• Aerial duels success rate: I’m only interested in success rate to not automatically punish goalkeepers who tend to be more cautious (but also consistent in their efforts). Simply put, when you do engage in an aerial duel, you’d better be 100% sure about it and, ultimately, 100% successful at it (hint: only 8 out of 24 goalkeepers considered were indeed 100%).
• Interceptions / left line to claim or punch: These two sound like they overlap, and they do indeed, but not too greatly to leave out one of them. While the latter only accounts for instances where goalkeeper comes out to deal with a cross (always successfully!), the former also includes balls played outside the box or claimed passes cutting across the six-yard box.
• Clearance from the centre of the box: This can be a punch or a kick into the stands, simply any solution to a mild or major problem that arises in the most heated area of the box — and not by the goalkeeper’s own doing. The area, by the way, roughly corresponds with the no. 2 on this scheme.

Side note: Keep in mind that this set of metrics is not necessarily positive. If a goalkeeper does a lot of damage control (hence shoots up in the average percentile for this particular area of goalkeeping), it typically suggests he’s also highly erratic and very unpredictable. Case in point: both Jiří Letáček and Tomáš Grigar score high(est) on damage control but low(est) on consistency and dependability, with many of their “control” decisions leading to damage.

Centre backs

While I very much dip into the unknown with goalkeepers, like I did last year, the centre back territory is a much more familiar one to navigate, so I won’t sort through all stats and all categories of them (you’ll see them on pizza charts). Still, it’s worth revisiting a couple of the not-so-straightforward metrics, what do they mean and what’s the thinking behind their inclusion.

• Defensive duel: This 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).
• Loose ball duel: It occurs when no team controls the ball so it could be considered “in possession” (these are typically the crucial “second balls”).
• Controlled clearance: Apart from clearances from centre of the box which I got into earlier, I also track those clearances that are “controlled” (expressed as sort of a success rate in %), ie. those that result in the team in question retaining the possession 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.
• Forward pass played into half-space: Oh yes, your hipster friend’s term— of course it makes a return in 2022. You always find the half-space here, and exploiting it is a noted weapon, so hitting them regularly is welcome.
• Chance/goal-creating actions: This is basically any creative contribution that doesn’t lead to the centre back getting credited with an assist — ie. either a secondary/terciary contribution to a goal, or a primary/secondary contribution to a wasted goal-scoring chance per my personal notes. It’s important to remember that not every chance has a secondary contribution, and that I don’t consider every single involvement in the lead-up to be a “chance-creating action”. It must truly be vital for the chance to present itself, not just a sideways pass on the half-line. It’s a fairly rare occurrence for a centre half, and so Lischka (12), Kadlec and Martinec (both 10) are the only 3 centre backs to have made their way into the double digits (excl. Hancko or Vraštil who’ve acted as fullbacks, too).

Fullbacks

There’s not much new to introduce with the rest of the defenders either, but I put more emphasis on proactivity and dynamism on/off the ball, further underlined by the first two metrics below, and some clever calculation expressed in the third metric that I’m particularly happy with, to be honest.

• Successful loose ball duels: I’ve already introduced this stat earlier with centre backs, but here’s an important distinction: while centre halves get compared in terms of success rate (%) because there’s generally more loose ball duels occurring in their space, with fullbacks (and defensive midfielders) I’m more interested in how many of those duels they win per 90 — highlighting their engagement level on top of what success % tells us.
• Offensive duels in final third won: This kind of sounds like dribbles, but it’s a tad different. You typically get tagged for an offensive duel even when receiving and shielding the ball under pressure, and you win it even when the ball bounces away for a throw-in (in your favour, that is) or you get fouled for a set piece. Either way, while some of those wins are only partial ones, they do gain your team some valuable attacking zone time.
• % of crosses blocked early: Even in the inverted wingback era, crossing remains a big part of one fullback’s game which is why most public models or databases include crossing accuracy. I myself had considered “accurate crosses to centre of the box” last year, but I won’t be doing so any longer, and I’ll explain why: it’s too random. Is it necessarily the fullback’s fault when he delivers the ball into a danger area only for no one else to take the hint and make the run? It usually is not. Is it, however, the fullback’s fault when his delivery gets blocked before it even enters the penalty area, potentially kick-starting a dangerous counter? Typically it is, yes. So that’s what I’m looking for here, basically. I don’t care if your cross doesn’t quite connect as long as it can feasibly lead to a lucky bounce or even an own goal.

For good measure, here’s a reminder of the last year’s trio of introductions:

• Progressive run/pass: 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 picks the ball up 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.
• Accurate stretch pass: This type of a pass is not readily available on Wyscout so I cherry-pick it manually, defining it as any accurate pass that either goes in behind the line (forcing defence to re-shuffle / 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 — they stretch the opposition and buy your team time and area to hit them.
• Accurate cutback in danger zone: Effectively forming one part of a half-space, the “cutback zone” is another noted danger area of the pitch where you’d ideally want your modern fullback to function and do damage. I don’t necessarily limit myself to cutbacks from inside the box only, but they do have to be directed to centre of the box to qualify as valid threat.

source: efotbal.cz

Defensive midfielders

Just a little tweaks here and there, as well. We begin with what’s rapidly becoming a household metrics not just in this space, but just about anywhere.

• Possession-adjusted interceptions: Interceptions are generally well understood, but they mean next to nothing without adjusting for context. If your team always controls the flow of the game, ie. retaining possession, how the hell are you supposed to intercept anything? And why should you suffer for it? Enter the “possession-adjusted”. What Wyscout people do here is to pretend 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, even the guys on dominating teams can think of landing in the 90-100 percentile — like Lukáš Kalvach did in 2020/21, for example.
• Progressive passes allowed to go past: 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). Thus, 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 (and only re-appears with wingers to illustrate their level of pressure on fullbacks).
• Diagonal pass into final third: A new addition, and a CDM variation of the fullback’s “stretch pass”. It doesn’t necessarily have to travel on a diagonal trajectory, but it should come reasonably close for it to count. This type of a pass neatly accompanies the broader progressive pass, the more vertical through pass and the more game-breaking smart pass to give us a fairly complex picture of one holding midfielder’s passing range.

Attacking midfielders

Almost no stone left unturned. This is another position where I wasn’t too happy with the initial setup, hence leaving just five metrics intact and introducing eleven new ones. Let’s sort them out by categories again:

Prime offensive output

Along with expected assists (xA), I’ve identified three metrics to make up what I call prime offensive output — the cutting-edge spirit of one’s play.

• Success rate of actions inside the box: This is basically any successful execution — be it a pass, a shot (only those directed on target count as a success) or a duel of any kind. I’ve opted for success rate to not disadvantage those who — not just through their own doing, or lack there of — don’t get to influence the game from inside the box too often.
• Chance created for himself: This can either be interpreted in the more straightforward sense — you get into a great shooting position by beating a man and setting yourself free on your own — but also in the less selfish way, so to speak — by going for a quick one-two, neat give-and-go or even winning an aerial duel before making yourself available for a pass again. This is a sub-category of chance-creating actions I track myself manually.
• High-danger shot: Make no mistake, while we are interested in post-xG when it comes to goalkeepers and their high-danger situations, here we only look at pre-shot xG to see who gets into high-danger situations in the first place. Finishing will only become a matter of interest once we get to centre forwards (while tweaking the approach a little bit); with attacking midfielders, the sole ability to find those sweet spots inside the penalty area warranting an xG value of at least 0,2 tells us all we need. It’s usually a decent pre-condition for a fine offensive output — unless you’re Pavel Šulc, of course, who’s actually topped the charts as a dependable outlier.

Ability & tools to gain danger zone

I once again compare the respective foul-drawing records, but that’s not all!

• Offensive duels won top of the box (%): Originally, I’d be happy with successful offensive duels in final third normalized per 90 minutes. This year, I wanted to limit the scope significantly — once again, I’d rather consider the success rate to not discriminate, and I’m also focusing exclusively on true danger area entry points that, give or take, include:

• Inside / outside channel fed: A smart attacking midfielder should be able to pick out runs in a timely manner, and that’s exactly what this metric is supposed to describe. It looks for long-ish passes that either feed an overlapping run on the outside, or an underlapping one on the inside (threading the half-space) — both while penetrating the danger zone.
• Deep completed pass / cross: Last year I combined both types of deep completions into one metric, now I thought it’d be interesting to see and compare them separately to measure one’s versatility. For the sake of a reminder, here’s where any “deep completion” must occur to count as one:

per Wyscout glossary

Ball progression

• Accurate smart pass: This shouldn’t be an unknown entity to you anymore. I let Wyscout lead on this one: any “creative and penetrative pass that attempts to break the opposition’s defensive lines to gain a significant advantage in attack” it is then, though I’m not convinced consistency-wise.
• Meters gained via run / passing play: This is a new, but straightforward one. Ball progressors can sometimes be divided into two groups — those who run with the ball to gain precious meters for their teams, and those who prefer passing it to achieve the same. The very best ball progressors — like Petr Ševčík — do both at a very high level. Important to note, though: even this metric is normalized per 90 mins, it’s not a sum total of meters.
• Meters gained per ball loss: This one takes a look at ball progressing from a different, but no less important angle — if you do attempt to move the needle often but also lose the ball frequently, you’re not going to look too fantastic in this column. This is very much what set Stanciu apart.

Proactivity & discipline

That last metric is already balancing proactivity and discipline in a way, but I highlight three even more fitting metrics as part of this category. Foul differential (ratio of fouls won and conceded normalized per 90 mins) and balls recovered in final third should require no detailed introduction, but…

• Long-distance hopeful shot: Oh boy. He did go there after all. The ultimate beating stick of the “xG folks” and he reaches for it by himself. Yes, the more you try your luck the worse you look here, but that’s not to say you should never do it. If there’s an open lane and/or your team struggles to break down the opponent’s low block, pulling the trigger is sometimes the right move. That’s also why I don’t mind a mid-range outside-the-box attempt; I only count those from roughly 30+ meters. (And guess what, Stanciu was the 2nd best CAM despite being the worst culprit in this sense — it’s just one of 18 metrics fed into the model, after all.)

Wingers

I’ve covered all of the (new) ground together with other player roles, with the only slight distinction here being that I also look at primary chance-creating actions separately. However, there’s a few older ones that deserve a refresher:

• Successful penalty area entry from open play: This is what it says on the tin, really, just want to specifically bring your attention to the “open play” bit (this distinction goes for deep completions, too; only xA generally includes set pieces, as Wyscout doesn’t offer a way how to easily filter them out), and also the fact you can crack the penalty area in possession of the ball, as well, so it doesn’t exclusively capture passing/crossing ability.
• Accurate cross field passes (~50 meters): This may mostly suit a pondering type of a tucked-in winger (it doesn’t quite, though, and you can take the next picture as a hint), which is fine because there are other metrics that will put a straight-line runner in pole position instead. Either way, this is any successful attempt to switch it up, to make the opponent shift from one side to the other and gain advantage, tracked by myself.

source: fkjablonec.cz

Centre forwards

For one last time, it makes sense to provide lowdown on one category at a time:

Attacking the box & poaching

For the proverbial fox in the box, there’s still the specially extracted xG of all non-penalty shots from inside the box only (combined and converted to a per-90 basis) and I stick to the success rate of actions taken inside the box, introduced earlier, but as far as finishing goes, we’ve got to take a deep dive…

• Quality added to finish (xGS-xG): Most data analysts would tell you finishing is too random to spend too much time — or any time— on dissecting it. It’s tempting to agree, since the very reason why Plzeň have just won a shocking title was all the same a big reason why Plzeň bombed a year earlier. Per this metric, Jean-David Beauguel made for the 8th best finisher this year (excluding penalties), a wild swing from being the 11th worst last year when he missed the 2nd most high-danger chances. It’s exactly for this unpredictability and virtual impossibility to detect any longterm trends that most data analysts readily give up on sizing up finishing ability altogether. I refuse. First of all, this particular metric still tells us a valuable thing or two even if we shouldn’t read too much into it. If your individual sum of xGS — expected goals scored; ie. post-shot xG — far exceeds the sum of xG — expected goals; ie. pre-shot xG — it can effectively mean (one of) two things: the striker in question 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. There are absolutely trends to be found along this line in particular.
• Second of all, I think we can collectively do better. One little part of my attempt to do so is constructing this very simple luck index of sorts where I take away non-penalty goals scored from the sum of xGS and woodwork-hitting attempts — with the highest luck index then suggesting the striker in question had some bad luck to rue along the way; be it for only a few inches missing every now and then, or the opposing goalkeeper standing on his head to deny him all too often. This little aid helps us to explain why Milan Škoda or Daniel Vašulín got stuck far away from the double digits (after all), or why Jablonec had to fight off a real relegation threat (with both Martin Doležal and Jan Silný combining for roughly 7 goals scored below expectations). It’s not all-encompassing, but it’s something. And it’s definitely better than just looking at pre-shot xG and actual G discrepancy.
• Finally, I decided to ditch the original “high-danger chances missed” metric which, undoubtedly, tilted the field too heavily against the strikers on strong teams who had better service, and hence more opportunities to miss. Instead, I’m now looking at % of high-danger shots put off target which does a much better job in measuring one’s unease in front of the goal. To be clear, any shot that gets blocked (on the way to goal) or saved by the goalkeeper doesn’t work against the striker as part of this metric — it was an OK job done as far as giving him at least some chance to celebrate. Meanwhile, missing the target in a high-danger situation (at least 0,2 xG) simply never reflects well on the striker in question, so no slack cut there.

Target man qualities

This set of skills has been widened considerably, and so — along with the mainstays such as pass to final third accuracy (%), dangerous set pieces won, expected assists and deep completed pass — it now includes the following:

• Back pass accuracy (%) in danger areas: With this new metric, I’m looking to capture neat little lay-offs in areas right around/in the box where they can feasibly lead to a direct goal-scoring chance. I’ve gone with accuracy to, additionally, size up one’s first touch and soft technique. At first sight, the results are mixed: Chorý looks great, Beauguel looks bad; Cicilia looks outright fantastic, Tecl or Papadopulos look outright awful.
• Aerial duels won in/around the box (%): Undoubtedly another big part of a target man’s skillset, this metric was sorely missing from my last year’s model. I may throw-in the occasional inside own box success rate to demonstrate one’s usefulness in his own end, on defensive set pieces, too.

Proactivity, efficiency & discipline

The last grouping of metrics ranges from offsides and foul differential (discipline) through meters gained per ball loss (efficiency) all the way to balls recovered in the final third and the only bigger addition:

• Balls recovered via ground duel engagement: While balls recovered in the final third can come along in different forms and shapes — via astute or completely random positioning, via an aerial or ground duel — this metric designed to (also) measure one’s proactivity isn’t limited to any particular area of the pitch, but is limited in terms of the means (only ground duels).

Most metrics referred above are directly taken or sourced from Wyscout database.