Football World


youngsports 2016. 6. 15. 08:01



Let’s start with a scenario involving Team A and Team B. Ready? Here it goes …

Team A: 52% possession; total shots 18; corner kicks 7

Team B: 48% possession; total shots 14; corner kicks 5

These are the statistics we normally see during football broadcasts.

Do these numbers tell you anything? Can you tell who won the game?

Well, Team A was Brazil and Team B was Germany in the 2014 World Cup semifinal. Germany advanced, thanks to a historic 7:1 victory. Yet according to the stats, looks like it was a close one.

Hegeler and Reinartz during their time at Bayer LeverkusenSINSHEIM, GERMANY - NOVEMBER 25: Stefan Reinartz of Leverkusen celebrates with teammate Jens Hegeler after the Bundesliga match between TSG 1899 Hoffenheim and Bayer 04 Leverkusen at Rhein-Neckar-Arena on November 25, 2012 in Sinsheim, Germany. (Photo by Dennis Grombkowski/Bongarts/Getty Images)

Jens Hegeler and Stefan Reinartz during their time at Bayer Leverkusen on November 25, 2012 in Sinsheim, Germany. (Photo by Dennis Grombkowski/Bongarts/Getty Images)

Football statistics that don’t reflect what happened on the pitch suck, it’s that simple.

Counting stuff and dumping the raw data in front of the audience without any further analysis, is what TV has been doing for decades. There are niche shows and websites where tactics and stats get broken down, but data needs to be processed and evaluated. Analytical content therefore goes online after the games ended. The casual TV audience doesn’t care anymore at that point and sites like and aren’t exactly mainstream to begin with.

Big websites like Kicker, SportBild or Spox are all about transfer rumours, slideshows and interviews. Analytics articles are long, hard to get and simply create less traffic than clickbait stuff like: “Pogba liked FC Barcelona on Instagram. Transfer imminent?”.

In 2015 Stefan Reinartz (retired in 2016, formerly Bayer Leverkusen, three caps for Germany) and Jens Hegeler (under contract at Hertha BSC) identified this problem.

Both are/were defensive midfielders. Throughout their careers they became annoyed that media outlets and coaches evaluated their performances with tools and methods that don’t reflect the actual value they added to their team. A “shutdown” cornerback like Richard Sherman can be a star in the NFL thanks to interceptions, broken up plays and tackles. “Lock down NBA defenders” like Bruce Bowen and Dennis Rodman can prove their worth with steals, blocks and rebounds. Football has goals and assists, that’s it.

Defensive midfielders who can’t score like Ballack or hit free kicks like Pirlo, are the unsung heros in football. They do crucial work that doesn’t show up in any stat category. In the example above, Vidal wins the ball from Juve’s Pogba and hits Coman on the flank. Vidal’s play was the key to that goa and also prevented a dangerous counter attack, but Coman and Müller are the only ones who earned meaningful stats on that play.

Why are the usual stats no good?

Well, the usual stats aren’t bad per se and do contain information. But they are all about quantity, not quality.“Possession” for example means nothing in a vacuum.

A lovely through ball by Thiago takes only a split second to perform. That pass has a negative impact on Bayern’s possession rating, even though it might send Lewandowski on a breakaway.

Another example that proves that our old metrics don’t mean anything.

Another example – from the recent Champions League semi-final – proving that our old metrics don’t mean anything.

“Total shots” doesn’t mean much either, a close range shot by Lionel Messi counts just as one shot, just like a desperate effort from 35 meters out by Heiko Westermann.

And “corner kick totals” have a different value to each team. Tiki Taka teams like Spain or Barcelona often don’t even send crosses into the box from a corner, while physical teams like Liverpool or England heavily rely on set pieces.

“Completed pass percentage” is also flawed if you don’t assess the risk/reward scenario of every single pass. During the 2015/16 Bundesliga season the pass completion rate of Mats Hummels (85%) was worse than Sokratis’ numbers (88%). Is Hummels the inferior passer of the two?

There are some more  advanced metrics around like Expected Goals (ExpG) and Total Shot Rate, but to be honest they might be too complicated for the average viewer.

And I actually see a huge flaw in ExpG, since the stat suggests that a shot from a certain area of the field has a certain chance of going in. Well, everybody who’s played football himself, knows that weather, pitch quality, physical condition or even the type of shoe you wear can alter a shot greatly. There are no identical situations in football.

Let’s not forget that up to 30 million Germans watch Die Mannschaft play, so you need a metric that is easy to explain and understand, since grandma’s and kids watch too.  
In the USA the majority of the public knows what a sack or a rebound is, while I know few people here in Germany who know what ExpG is.

What does that new metric do?

All of Impect’s metrics are based on the idea that every play needs to be assessed in terms of “quality”. A team can only score, if it can move the ball past defenders (“Gegner überspielt,” or “defenders beaten by the pass”). Beating defenders adds quality to a play, the more defenders you take out, the higher the quality of a play. Can’t argue with that.

The big breakthrough here is that “beating opponents” is an objective stat that can assess quality without relying on subjective opinion, you either beat the defender or you don’t.

The defending team obviously tries to prevent all that, so a low number in the “Überspielt werden” (getting beaten) category tells you how good a team is at defending.

To analyse every play as a separate entity and adjust for quality is genius. 
In this video (German only, but the illustrations speak for themselves) you can get an idea how it works.

It all comes down to the number of defenders in your way before and after a certain play, this logic can also be applied to dribbling events, turnovers and interceptions as well. In the above Vidal example, he wins back the ball in a situation where Bayern was pretty vulnerable, that counts for more than an interception with six Bayern teammates behind him.
The total of “opponents outplayed” during a game is called “Packing-Rate”.

Impect further determined that beating defenders is more valuable to an offense than getting the ball past pressing opponent strikers. The number of deep defenders taken out per game is called “IMPECT”, just like the company.

Why are those new metrics better?

Let’s take a look at the Brazil vs. Germany game again, but this time through the eyes of Reinartz and Hegeler’s system:


The Packing-Rate metric shows that Germany took out 61 more players than Brazil. A 15% swing in favor of Die Mannschaft, yet that doesn’t explain a 7:1 butt-whipping. The Impect numbers are more conclusive, Germany was much better in beating defenders than the Selecao.

Stop this clip at 5:35 and look at the Brazilians standing between Kroos and Müller, you can identify center backs Dante and David Luiz pretty easily (hint: big hair).

When a pass gets completed behind opponent’s center backs, it creates a dangerous situation almost every single time. The CB’s are the last line of defense, getting past them is the “money play”.

Evaluated by old metrics, Müller would earn a “completed pass” here. Kroos’ lovely ball would also show up as “one completed pass”. Klose gets all the glory, while Müller and Kroos don’t get much love from the stat keeper.

You see, the old metrics simply don’t work; instead, let Impect tell you the real story of this particular play.

kroos to müller

Impect’s metrics don’t care that Klose converted from close range. The “heavy lifting” on that play was done by Kroos and Müller – Klose only had to punch it in.

Klose gets a +1 for securing Müllers pass, that’s it. Any replacement level Bundesliga player can score from that spot, but only few players in the world can hit Müller in stride like Toni Kroos did on that play. Big difference.
Kroos earns +4 in “Gegner überspielt” for his pass to Müller, just like “TM13” gets +4 in “Überspielen ermöglichen” for securing this valuable through ball. 
Müller receives an extra +1 for his pass to Klose that took Marcelo out of the play. 
Brazil’s defense “earns” -5 in terms of “Überspielt werden”.

The MVP of that sequence clearly is Müller, who found a pocket of space, ran into it while staying onside, created a passing lane for Kroos (+4), controlled the pass and dished it out to Klose (+1). Finally, there is a stat that can really show you who deserves credit.

Additional Examples

Hummels to Reus:

Hummels and Reus receive a +6 in Packing each, since Hummels’ pass and the run by Reus were equally valuable. It kinda works like a completion in American Football, where the QB and WR both get credit and produce stats on the same play (yards, catch/completion). The pass by Hummels was great, but the running lane and timing of Reus was just as great. If Hummels misplaces the pass or Reus mistimes his run, the play is dead.

hummels reus

Irrelevant pass:

This is a pass that doesn’t show up in the Packing stats, it totally counts as a “completed pass” using the old methods though. Impect disregards all passes that go backwards or don’t beat any defender. It’s a correct assessment, a pass that doesn’t travel forwards, doesn’t help you score. Can’t argue with that. At the same time the sample size that needs to be analysed gets a lot smaller, which speeds up the compilation and analysis of data.

carvajal and javi

Even Messi’s greatness can be expressed in numbers:

This stunning dribbling sequence gives Messi a +4 in Packing-Rate. What’s so great about this dribbling and the goal is that Messi still had three defenders in his face when he scored. It’s hard enough to earn a +4 in Packing with a pass, Messi can do it on his feet and still convert a tough shot.

It was a legendary goal that actually consisted of two world class plays:

Play 1

messi step 1

Play 2:

messi step 2

However, despite Messi’s transcendent skill, Impect’s 2015/16  MVP of all European leagues was actually Toni Kroos, not Messi.

Kroos beat 85 defenders per game (!), the average of all defensive midfielders is at 28. OK, you probably didn’t need fancy metrics to know that Kroos is very good at passing, but there never was a number that could tell us how much value Kroos really adds to Germany and Real Madrid.

Toni Kroos Real Madrid

Can these stats tell me who to bet on?

No, I’m sorry dear gamblers. I literally “wouldn’t bet on it”.

While the sample size Impect has analysed so far proves that their system is more accurate than anything we’ve seen before, it still can’t predict outcomes – it’s descriptive, rather than predictive. All we know is that the team who scores better in the “Gegner überspielt” statistic, has a 86% chance to not lose the game.

But Impect metrics can totally tell you who played “better” and “should” have won or not lost, which could be huge on many levels.

Those 14% of teams who lose despite winning the Packing-Rate battle, can now point to bad luck and individual errors. Coaches might save their jobs once they can present statistical evidence that their team gets unlucky each week. Guys like Kroos or Xhaka might make as much money as strikers one day. And GM’s finally have a tool that helps to prevent expensive transfer busts, since talent evaluation will be revolutionized. Speed and overpowering physicality of a player can be lost through injuries and age. The ability to play a valuable pass doesn’t go away so quickly, ask Xavi and Pirlo. Those guys played each other in a Champions League final at 35+ years, while über-athletic players don’t age that well. Now clubs can find out if a prospect can play valuable passes or if he is simply faster and stronger than the opponents he faces.

Brad Pitt (L) and Jonah Hill (R) in a scene from Moneyball.

Brad Pitt (L) and Jonah Hill (R) in a scene from Moneyball.

What’s Next?

The stats Reinartz and Hegeler came up with are nothing less than a quantum leap for football analytics, so naturally there is high demand.

Die Mannschaft, Bayer 04, Borussia Dortmund and RB Leipzig have already licensed the technology. That client list proves that Impect is on to something. Roger Schmidt, Joachim Löw, Thomas Tuchel and Ralf Rangnick are among the smartest and most innovative coaches in the game today. If all of them like something, it’s legit.

During the Euros 2016 German #1 network ARD will use Packing-Rate in broadcasts. According to former Bayern star and ARD employee Mehmet Scholl, the data can be compiled so quickly that Packing numbers in real time are possible.

And even more important, Packing is easy to explain and won’t scare away the casuals.

Unfortunately, it’s unknown if there are plans to reevaluate past World Cups, Euros and Champions League games. We also have no clue whether an open access database will be available to the public at some point or if this technology will be kept behind closed doors.

CR7 vs. Messi: who's better? Debates like this could be a thing of the past.

CR7 vs. Messi: who’s better? Debates like this could be a thing of the past.

All I know is that Impect’s stats can be combined with metrics like expected goals by the internet crowd, this will produce even better numbers that can be tailored to each position or player.

Lahm or Maldini? Ronaldo or Messi? Zidane or Netzer? Luis Suarez or Klinsmann?

Many historic “Who was better?” arguments could be settled for good.

And even if you are a traditionalist who is sceptical about “all that nerdy stuff”, don’t worry!

Games like Dortmund at Liverpool (3:4) will still happen and prove even the most advanced prediction models and algorithms wrong. But for those who look at football through spreadsheets, Impect’s new numbers will be a revelation.

The only thing that worries me, is that once advanced metrics really take off, clubs like Real and Bayern will sign away the best analysts and tech guys from clubs like Dortmund and Villarreal. Unfortunately an “information war” can be won with money.

So don’t expect advanced metrics to even the odds between the “rich” and “poor” like it did in baseball for a while. The datacenter in Barcelona’s La Masia will be bigger and better than the one Darmstadt 98 will have, I’m pretty sure about that.

Anyways, I guess the “Moneyball” era in Europe’s top football leagues has begun.

It was about time.