Analysis
Packing metric football: counting the opponents you bypass
Packing counts how many opponents an action takes out of the game. What qualifies as bypassed, why weightings differ, and the four ways the number misleads.
By CricketTaken EditorialPublished Analysis18 min read
Two passes in the same match travel exactly eighteen yards. The first goes sideways between a pair of centre-backs and changes nothing about anybody's position. The second goes forward through a gap between a holding midfielder and a centre-back, and four opponents who were in front of the ball are now behind it. Completed passes: one each. Pass accuracy: identical. Every conventional column treats them as the same event.
The packing metric in football exists because that equivalence is absurd. It counts the opponents an action removes from the play, so the second pass scores four and the first scores nothing, and a player who repeatedly finds the second kind of pass separates from one who does not. The idea is easy. The counting rules underneath it are not, they differ between the people who implement them, and almost every misreading of a packing figure comes from assuming the version you have in front of you is the version somebody else is quoting.
Two passes of equal length, and only one of them costs the defence anything
Begin with what the metric is trying to capture, because the counting rules follow from it.
A defending team is an arrangement of bodies between the ball and its own goal. Its value at any instant is roughly the number of players it has in that arrangement and how well organised they are. Every one of them is a body that has to be beaten before a shot is possible.
An action that leaves four of those bodies behind the ball has reduced the defence from, say, nine outfield players between ball and goal to five. It has done this instantly, and the four bypassed players cannot undo it without turning, sprinting and recovering, which takes several seconds they may not have. The defence that remains is thinner, more stretched and more likely to concede.
That is the currency the metric is denominated in. Not distance, not completion, not territory. Bodies removed.
The consequence is that a five-yard pass can be worth more than a fifty-yard one, which is the single most useful thing the metric does and the thing no distance-based measure can reproduce. A short ball played between two centre-backs into a striker's feet bypasses both of them. A long diagonal that travels sixty yards to a full-back standing in space bypasses nobody at all, because everybody it flew over was already wide of the direct line to goal.
Where the idea came from, and what it was reacting against
The metric was devised by two German professionals, Stefan Reinartz and Jens Hegeler, in the period after the 2014 World Cup, and the company they founded to sell it, Impect, has since been absorbed into a larger sports technology catalogue.
The starting complaint was specific and it came from inside the game rather than from a statistics department. Players and coaches looking at their own post-match sheets kept finding that the numbers described a match they had not played in. A midfielder could finish with ninety-four per cent pass accuracy and have contributed nothing, because ninety per cent of his passes went sideways or backwards to a teammate under no pressure. Another could finish at seventy-eight per cent and have twice put the opposition defence in serious trouble.
Pass accuracy, in other words, is a measure that improves when you do less. Every player knows how to raise it: pass shorter, pass safer, pass away from the goal. A statistic that can be improved by playing worse is a statistic that will eventually corrupt behaviour, and in youth development it demonstrably did.
The response was to change what gets counted. Instead of asking whether the ball reached a teammate, ask what the pass did to the opposition. That reframing has a pleasing property: it cannot be improved by playing safer, because a safe pass beats nobody and scores nothing. The incentive now points at the risky action, which is the direction football wanted the incentive to point.
The reframing also explains why the metric was collected manually for years despite the cost. Answering "what did this pass do to the defence" requires watching the defence, and event feeds of the era recorded the ball rather than the twenty-one players around it.
The packing metric in football: what a single action is worth
The mechanics, in the simplest form.
Take the ball's position immediately before the action and immediately after it. Draw the set of opponents who were between the old ball position and the goal being attacked. Draw the set who are between the new ball position and that goal. The difference is the count.
An opponent who was in front of the ball and is now behind it has been bypassed. He scores one. Add them up and the action's packing value is the total. Award it to the player who made the pass or the dribble, and on most implementations award the same figure to the receiver as well, on the grounds that his movement and control were half of why the action worked.
Two immediate consequences follow from this construction and both catch people out.
A dribble counts. Carrying the ball past two opponents produces the same bypass as passing it past two. The metric was designed to value ball progression by any means, and one of its early attractions was that it credited the carriers whom passing statistics ignore.
A backwards pass can score zero and nothing worse. Playing the ball back puts opponents in front of it again, and most implementations floor the value at zero rather than recording a negative. That choice is defensible, since a reset is not a mistake, and it is also a hole that we will come back to.
The count is a raw number of players. It has no units and no natural scale, which means an isolated packing figure is uninterpretable without a comparison group. A total of forty in a match is meaningless until you know what a player in that position usually posts.
Who counts as bypassed, and the geometry that decides it
Here is where implementations part company, and where a figure quoted without its rules becomes untrustworthy.
The naive rule is a straight line: an opponent is in front of the ball if he is nearer the goal being attacked than the ball is, measured along the length of the pitch. This is simple to compute and it produces obvious nonsense. A right-back standing on the opposite touchline, forty yards from the ball, is nearer the goal in that sense and is counted as bypassed by a pass down the left. He was never involved, he could not have intervened, and beating him was free.
The corrections divide into two families.
Corridor rules. Only opponents inside a defined channel around the line from the ball to the goal are eligible. The channel might be a fixed width, or it might widen as the ball gets further from goal on the reasoning that a defender has more time to slide across from distance. This removes the far-side full-back and it also removes some genuinely beaten players who were closer than the corridor allows.
Reachability rules. An opponent counts only if he could plausibly have intervened, which is estimated from his distance from the ball, his facing direction and the time the ball spends travelling. This is more faithful to the idea and much more expensive, since it requires positional tracking rather than event data alone.
The two produce different numbers for the same match, and the gap is largest exactly where the football is most interesting: crowded central areas near the box, where a corridor rule may exclude a defender standing three yards outside the channel who was in fact the man who should have made the tackle.
There is a third variation nobody mentions. Some implementations count the goalkeeper among the bypassable opponents and some do not. Since a pass into the box in front of the keeper does not beat him and a ball played across him does, the choice changes final-third figures by a little and changes the ranking of the players who provide the last pass by more than a little.
Opponents bypassed and defenders bypassed are two different numbers
The metric's second layer is where most of its predictive value lives.
Bypassing a striker who has jogged back to the halfway line is a real thing and a cheap one. He was not going to make the tackle, he is not part of the defensive block, and beating him alters the defensive structure barely at all. Bypassing a centre-back twenty yards from goal alters it enormously.
So the family splits. The broad figure counts every opponent left behind. The narrow one, generally described as defenders bypassed, counts only the players occupying the rearmost defensive positions at that moment, which in practice means the back line, the goalkeeper and often the deepest midfielder. Published descriptions of the original commercial implementation define the defensive group by position relative to the goal rather than by shirt number, so a full-back who has pushed forward may not be in the group and a midfielder who has dropped in may be.
The distinction matters because the two numbers correlate weakly. A deep midfielder in a possession side can post a large broad total by playing forward past the opposition front two all afternoon while barely touching the defensive group. A wide forward who receives inside the full-back three times a match can post a small broad total and a large narrow one.
If you are being shown one number and told it describes creativity, ask which one. The broad figure describes progression volume. The narrow one describes penetration, and penetration is what precedes goals. The related idea of manufacturing a spare man to create exactly those penetrations is set out in the piece on building overloads and using them.
Where the credit goes, and why the receiver gets paid too
Handing the same score to the passer and the receiver is unusual among football metrics and it is deliberate.
A line-breaking pass requires two things to happen at once. Somebody has to see the gap and weight the ball into it. Somebody else has to be in the gap, at the right moment, having moved there against a marker's shoulder and having timed the movement so as not to arrive early and close the space himself.
Attribute the whole value to the passer and you make an accounting error the size of half the action. Attribute it to the receiver and you make the same error in reverse. Splitting it, or duplicating it, keeps both contributions visible.
This does mean team totals are not a simple sum of player totals, and anybody constructing their own version needs to decide whether they are double counting on purpose. It also means the receiving figure becomes a rather good description of a specific skill: occupying the space behind a defensive line and being available there. Forwards who drop between the lines and second strikers show up strongly on it, and their passing numbers may be modest.
The asymmetry to remember is that a receiver cannot generate his own figure. He needs a teammate capable of finding him. A forward with excellent movement in a team with no line-breaking midfielder will show a low receiving total that describes his teammates rather than him. The marking systems that make those spaces harder or easier to occupy are covered in man-marking against zonal marking.
What the count refuses to see, and why each omission was chosen
A metric is defined as much by its exclusions as by its rules, and these are the categories that sit outside it on purpose.
Whether the bypassed players mattered. Four opponents beaten in the centre circle and four beaten on the edge of the box score identically on the broad count. The narrow variant partially fixes this by restricting the eligible group, and no version of the metric contains a term for how dangerous the resulting position is.
Time. A defence broken open for half a second, because the receiver's touch is heavy and a covering player arrives, has been broken open in the arithmetic exactly as much as one broken open for four seconds. Duration of the advantage is where much of the real value lives and none of it is recorded.
Off-ball work that created the gap. A striker who pins two centre-backs by holding his position, allowing a midfielder to receive between the lines, has done a large share of the work and appears nowhere. Only the two players directly involved in the action are credited.
Defensive intent. A team pressing high has deliberately put nine players in front of the ball, which makes it enormously easier to bypass several of them at once. A team defending in a deep block has almost no players in front of the ball to bypass. The metric reads the first as a defence being torn apart and the second as an attack achieving nothing, when the tactical reality may be the reverse. The counting rule for how aggressively a side presses is set out in the piece on PPDA and what pressing intensity actually measures.
The pass that was not played. As with every count of completed actions, a player who ignores the line-breaking option four times a match is invisible.
Each of these was excluded for a reason. Including danger would turn the metric into a possession-value model and lose its distinctive character. Including time would require tracking data throughout. Including off-ball contribution would require a theory of credit that nobody has agreed on. The exclusions are the price of a metric simple enough to compute and explain.
Worked example: one invented attack, action by action
- Starting shapeThe defending side sits in a mid-block: two forwards, a band of four in midfield, a back four and the goalkeeper. The attacking team has the ball with its left centre-back, ten yards inside its own half. Nine outfield opponents and the goalkeeper are between the ball and the goal being attacked.
- Centre-back to right centre-back, squareNobody is bypassed. The ball has moved twenty yards across the pitch and every opponent remains in front of it. Packing: 0. This is the pass that conventional statistics score identically to the one four steps below.
- Right centre-back carries forward eight yardsOne of the two opposition forwards is drawn towards him and is left behind as the carry continues past his shoulder. Packing: 1. The carry counts on exactly the same terms as a pass, which is one of the metric's better features.
- Carrier to the holding midfielder, played into feetThe second opposition forward is bypassed. Packing: 1, running total 2. Under an implementation that counts the goalkeeper as bypassable this changes nothing yet, since the keeper remains in front of the ball.
- Holding midfielder turns and plays it between the two central midfieldersThe pass beats both of them and also the opposition left midfielder, who had tucked inside and sits within the corridor. Packing: 3 under a wide corridor rule, 2 under a narrow one that excludes him. This single implementation choice is the difference between a running total of 5 and one of 4.
- Receiver, between the lines, slides it past the centre-back into the strikerOne defender bypassed, and this one belongs to the defensive group, so it scores on both the broad count and the narrow defenders-bypassed count. Broad running total 5 or 6. Narrow total 1, and that 1 is worth more than the preceding five put together in terms of what usually happens next.
- The striker is closed down and lays it backBackwards. The centre-back who was bypassed is now in front of the ball again, and most implementations record zero rather than minus one. The structural damage the attack caused has been partly undone and the metric does not say so.
- Where the attack finishedA broad total of five or six depending on the corridor, a narrow total of one, and no shot. A team-level packing figure built from many such attacks will be dominated by the cheap early counts and will barely register the one action that actually opened the defence.
A wholly invented six-action attack against an invented defensive shape of ten outfield opponents plus a goalkeeper. It shows which players the count picks up and which it does not. No real match, team or player is described, and the totals are constructed for legibility.
The four ways a packing number misleads
The long ball problem. A goalkeeper who kicks it seventy yards over everybody has, by the letter of the rule, bypassed the entire opposition team. If the ball is headed clear immediately, the defence has suffered nothing and the total says otherwise. Implementations mitigate this by requiring the receiving player to control the ball, or by discounting actions with low completion expectations, and the mitigations vary. A packing figure from a very direct side needs checking against how often its long balls were actually retained.
The undone action. Because a backward pass floors at zero rather than scoring negative, a sequence that breaks a line and then gives the ground straight back nets out positive. Over ninety minutes a side that repeatedly penetrates and retreats accumulates a total that describes an attacking performance it did not have.
Bypassing players who were never in the way. The far-side defender, the forward tracking back forty yards behind the play, the winger who has not bothered to recover: each is a free unit of packing under a permissive geometry. The players who inflate their totals this way are usually the ones playing against disorganised opposition.
Volume through role. A deep central player in a possession side has more opportunities to play forward past two opponents than a striker has to do anything at all. Raw totals rank by opportunity before they rank by quality, which is the same trap every counting metric in football sets.
None of these make the metric worthless. They make raw totals worthless, and they make per-action rates, positional context and the narrow defenders-bypassed variant the numbers actually worth reading.
- 44Player A, opponents bypassed per 90
- 3Player A, defenders bypassed per 90
- 44Player B, opponents bypassed per 90
- 11Player B, defenders bypassed per 90
Two entirely invented players over one invented season, constructed so that an identical headline figure conceals two completely different contributions. Nothing here is measured and no real player is described.
Packing rate, and normalising for how much of the ball you had
A raw total answers a question about volume, and volume is mostly a function of minutes, position and how much possession the team had. The derived versions are where comparison becomes possible.
Per action. Divide the bypass total by the number of passes and carries attempted. This gives an average structural gain per involvement, and it reorders the table dramatically. Deep midfielders with enormous raw totals fall, because most of their involvements are square balls. Wide attackers and second strikers rise, because a higher share of what they do goes forward through people.
Per possession, or per hundred touches. A slightly different normalisation with a similar effect, useful because it does not penalise a player who carries the ball a lot and passes rarely.
Per opponent available. The most sophisticated adjustment and the least commonly published. Since a high defensive line puts more players in front of the ball, an action's difficulty depends on how many opponents were in the way to begin with. Dividing the bypass count by the number of eligible opponents converts the figure into a share of the defence beaten, which is a fairer comparison between a team that faces high pressing and one that faces a deep block.
The rate versions also behave much better across leagues. Raw totals inflate in competitions where defensive organisation is weaker, and a scout who compares a raw figure from a lower division against one from a stronger league will reach a conclusion the data does not support. Rates are not immune to this, since a defence that is easier to break makes each action cheaper, and they are considerably less distorted.
- Forward passes past the opposition front two47%
- Long balls into wide areas22%
- Carries out of the first line19%
- Passes through the defensive line12%
An entirely invented season total for an invented deep midfielder, broken down by the type of action that produced it. Constructed to show how much of a large raw figure can come from cheap actions. Nothing here is measured and no real player is described.
Show the numbers
| Item | Value |
|---|---|
| Forward passes past the opposition front two | 47% |
| Long balls into wide areas | 22% |
| Carries out of the first line | 19% |
| Passes through the defensive line | 12% |
A final note on sample size. Bypass counts are noisy at match level, because a single long ball can add a dozen to a total. Nothing below a large block of matches should be used to describe a player, and single-match packing graphics are entertainment rather than evidence.
The defensive figure nobody puts on a graphic
Every bypass has a victim, and the metric can be pointed backwards.
Count the times each defender is left behind the ball, and you get a description of how often a player is beaten. Do it for a defensive unit and you get a description of how often a structure is broken. This is far more informative about defending than tackles and interceptions, both of which reward the player who has already allowed the situation to develop.
The complication is attribution, and it is severe. If a pass goes through a gap between two midfielders, whose fault is the gap? The one who stepped out, the one who did not slide across, or the coach who asked for a shape neither of them could hold? A count assigns the bypass to both bodies and says nothing about responsibility.
Positional defending makes this worse. In a zonal structure a defender is often bypassed on purpose, because he has been told to hold a line and let the ball go past rather than chase it. He is doing his job correctly and accumulating a bad number.
The usable version is comparative rather than absolute. A defender bypassed far more often than his positional peers, in the same system, over a long enough sample, is telling you something. The same defender compared against a peer in a different system is telling you about the two systems.
Manual coding, tracking data, and why two versions disagree
The original implementation of the idea was human. Analysts watched matches and coded the actions, which is expensive, slow and surprisingly accurate on the judgement calls a machine finds hard, such as whether a bypassed player had any realistic chance of intervening.
The modern approach uses positional tracking, which gives the coordinates of every player several times a second and makes the geometry computable. It is faster, cheaper per match and completely consistent, and it is worse at the judgement calls, because it has to encode "could he have intervened" as a rule about distance and velocity rather than as an opinion.
The two therefore disagree in predictable places. Machines over-count bypassed players who are geometrically in front of the ball and practically irrelevant. Humans under-count in fast transitions where they cannot see everybody at once. Neither error is random, which means the difference between two versions of the same match is not noise you can average away.
There is a third source of disagreement that has nothing to do with method. Packing is a commercial product rather than an open standard, so there is no published specification anyone can check an implementation against. Figures that circulate on social media are frequently reconstructions built from event data by people who chose their own corridor rule. Some are careful. None are the same as one another.
Packing against expected threat and expected goals
Three families of metric now compete to describe the same part of the game, and they answer three distinct questions.
Expected goals values the shot. It asks how often an attempt from this position, of this type, in this situation, is scored. It is the most mature of the three and it says nothing about how the attempt was constructed. The mechanics are set out in what expected goals actually measures.
Expected threat and possession-value models value the state of the ball. They assign every location and situation a probability of leading to a goal within the next few actions, and credit each pass or carry with the change it produced. This is the most general framework and it captures the sideways pass that improves the angle, which packing scores as zero.
Packing values the structural damage. It is the only one of the three that asks about the defenders rather than about the ball, and that is both its distinctive contribution and its limitation. It knows that four players were beaten. It does not know whether beating them made a goal more likely.
The honest summary is that a possession-value model will out-predict packing on almost any goal-related question, and packing will describe what happened to a defence in a way the value model cannot articulate. Coaches tend to prefer the second kind of information, because it maps onto instructions they can give. Recruitment departments tend to prefer the first, because it maps onto money.
How to use a packing figure when you are shown one
Five questions before you allow it to change your opinion of a player.
Broad or narrow? Opponents bypassed and defenders bypassed are not interchangeable, and the second is the one that relates to danger. A graphic that says only "packing" is usually showing the broad figure because it is the bigger number.
Per what? A total per ninety minutes ranks by opportunity. A rate per hundred passes or per possession ranks by intent. The two lists are not similar.
Passer or receiver? Since both are credited, a large figure could belong to the player who found the pass or to the player who made himself findable. Both are valuable, and they are different skills that need different players next to them.
What was the completion? A bypass that ends with the opponent recovering the ball has damaged nothing. Long-ball totals in particular need checking against retention.
What happened in the next ten seconds? The metric stops at the moment of the pass. Structural damage that is immediately reversed by a backward pass, or wasted by a poor first touch, is not damage. Set the packing figure against final-third entries and shots and it becomes a useful description. Read alone, it is a count of an event whose consequences it never observed.
There is one habit worth adopting whenever a bypass figure appears in a scouting document. Ask to see the same player's number split by the phase it was accumulated in, separating settled possession from transition. Beating four opponents against an organised block is a skill. Beating four opponents in a broken game where nobody is in position is largely a description of the chaos rather than of the player, and the two are routinely added together into a single figure that reads as though they were the same achievement.
Used with those caveats, the idea remains one of the more genuinely original contributions to football measurement. It asked a question nobody else was asking, which is what the defence looks like after the pass rather than where the ball ended up, and every possession-value framework built since has had to account for the answer.
More of our work on how football is measured is collected in the football section, and the full run of explainers sits on the blog.
Common questions
What is packing in football statistics?
Packing counts how many opposition players an action takes out of the play, by leaving them behind the ball. A pass or a dribble that leaves four opponents between the ball and their own goal line rather than in front of it scores four. It was built to distinguish passes that break a defensive structure from passes that merely complete.
How is packing calculated?
The ball's position before and after the action is compared, and every opponent who was between the ball and the goal being attacked before the action but is behind the ball after it is counted as bypassed. Some implementations restrict this to a corridor around the direct line to goal, and some weight the count by the bypassed player's defensive role.
What is the difference between packing and IMPECT?
Packing counts all bypassed opponents equally. The related figure usually described as defenders bypassed, or impect, counts only the players occupying defensive positions closest to the goal, on the reasoning that beating a striker high up the pitch and beating a centre-back on the edge of the box are not the same achievement. The second figure is the more selective and the more predictive of danger.
Can packing be gamed?
Easily. A long ball hit over an entire back line bypasses everybody, so a side that plays direct can accumulate a large packing total without creating anything. Backwards passes followed by long forward ones do the same. Any reading of a packing number needs the accompanying receiving position and what happened next.
Is packing better than expected goals?
They answer different questions. Expected goals values the quality of a shot. Packing values the structural damage of a pass or carry, most of which happens before any shot exists. Neither replaces the other, and packing on its own says nothing about whether the bypassed opponents mattered.
Where can I find packing data?
It is a commercial product rather than an open one, so it appears in club analysis and occasionally in broadcast and press coverage rather than on free statistics sites. Figures circulating without a named source should be treated with caution, since the counting rules are not standard across every implementation.
Filed under Football·football · analytics · data · scouting · passing · defending