Ever wondered exactly what xG is? Been seeing xT more and more but unsure what it is? Or do you want to learn a little more about a few concepts? Then bookmark this page!
We’ll cover all the main metrics and some of the more specific, less-used ones. This will be a living glossary, so as you ask for more metrics to get added, or as we introduce some new topics in a GYSR or Stat Pack, we’ll add them below.
Quick Reference:
xG: The probability that a specific shot becomes a goal.
xGA: Expected goals against, or the xG of the opponent.
xA: The xG value of shots a player assists.
xT: The change in goal probability in the next few actions by moving the ball from A to B, positive or negative.
xG Chain: The total xG of all possessions a player was involved in.
xG Buildup: xG Chain minus the xG and xA of a player.
xPoints: The number of points a team might be expected to have, based on the xG and xGA of their games.
Possession: The share of game time the team had the ball.
PPDA: The average number of passes allowed per defensive action, only considering the front 60% of the pitch.
High Recovery: Ball wins in the final third or within 40m of the goal.
Line-Breaking Pass: A pass played from in front of the opposing forward, midfield, or defensive line, and completed behind it.
Impect Metrics
Bypassed Opponents (Packing): The number of opponents left farther from their goal than the ball after an action.
Bypassed Defenders: Similar to bypassed opponents, but only considering the 5 farthest back opponents.
Added Teammates: The number of teammates ahead of the ball when making a recovery.
Added Opponents: The number of opponents closer to your goal than the ball when losing the ball/when the opponent gains possession.
Removed Teammates: The number of teammates farther from the goal than the ball when losing the ball/when the opponent gains possession (i.e. farther upfield).
Removed Opponents: The number of opponents farther from the goal than the ball when making a recovery.
Non-Shot-Based xG (Packing xG): Essentially the xG that would be assigned to a shot, if a shot was taken at a specific moment. Values possible shooting positions instead of only assigning values when shots are taken.
Concepts
Transitions: The phase when possession switches teams and neither team is structured. Attacking Transitions are after recovering the ball, Defensive Transitions are after losing the ball.
Second Balls: The chaos phase, typically after long balls or possession losses, where neither team really has possession of the ball. Also the term for a rebounded ball to be picked up in this phase.
General Stats
xG (expected goals)
xG is the probability that a specific shot results in a goal. It ranges from 0.00 to 0.99, with a value of 0.20 indicating about 20% chance that that specific shot results in a goal. Or rather, if this shot was taken 100 times, we should expect 20 goals. It is effectively a measure of shot quality.
At its core, xG looks at the location of the shot, the position of the goalkeeper, the type of shot (foot, header, etc), the type of assist, if any (low cross, high cross, after a dribble, etc), and possibly more info such as positions of players between the shooter and goal, the weather, and more.
A major problem with xG is that every company will use different inputs. Some might not include the location of the keeper, some might include the weather... and depending on the data provider used as the raw data, the exact location of the shot might differ too! This leads to every provider having a different xG value for the same shot. This doesn’t mean xG is bad, it’s just a problem about xG models being commodities instead of standardized, public information.
xG doesn’t care about who is shooting. Players like Son Heung-min and Lionel Messi are players who consistently finish above their xG. While they theoretically would finish 0.20 xG chances at a higher rate that 0.20, xG is meant as the average. An average player taking that same shot 100 times would score 20. Plenty would score more, plenty would score less.
xG can be used as a sum, to see how many goals a team or player “should have scored”, but it is still only one metric, and every metric has flaws. Even a great metric cannot be used as a single truth-giving number. xG was never meant to be used by itself as frequently as it is. Don’t put too much weight to it, and make sure to look at other info to get a better picture.
xA (expected assists)
xA is effectively the xG of a shot that a player has directly assisted. Recently, some providers have changed this metric to Expected Assisted Goals (xAG). When player A passes to player B, and player B takes a shot worth 0.22 xG, player A is credited 0.22 xA.
We have to be careful not to put too much importance on this metric, as it doesn’t exactly capture a player’s playmaking quality. Take the 0.22 xA example. Was that 0.22 xG shot due to the player receiving a brilliant pass from player A, or did player A play an extremely simple pass to player B, who then dribbled by 2 opponents to get into a great position?
xT (expected threat)
While xG and xA use shots as the valuing action, xT uses ball movement. xT tries to value passes and carries based on the start and end location of the ball. Players get credit for moving the ball to more dangerous areas.
xT is the change in the probability of scoring by moving the ball from A to B. For example, the chance of scoring is low when the ball is just over halfway on the flank. But the chances of scoring are much higher just in front of the box. So, a pass from the flank into this area in front of the box should be rewarded. But passing from this area backwards to a #6 in the center circle would result in negative xT, since there’s usually less chances of scoring shortly after the ball is in the center circle than when it’s just in front of the box.
To create the probabilities, the pitch is split up into a big grid. The chances of goals coming in the next X seconds or actions (usually 5 actions) when the ball is in a specific cell of that grid is that cell’s value. We subtract the starting cell’s value from the ending cell’s value to get the xT of that action. For example, passing from a cell with a value of 0.02 to one with 0.07, we get 0.07 – 0.02 = 0.05. This pass is assigned 0.05 xT. Note that only completed passes receive xT values.
But we have to be very careful in how we interpret xT, and what it measures. Take a player with 1.2 xT in a game. This doesn’t mean they “should have” 1 assist. Even though xT works on goals to create probabilities, it’s not “predicting” goals. This player might have created about 1 goal’s-worth of threat for their team. An example would be a fullback consistently playing good passes up the flank to a winger running into space. We can value that fullback’s contribution to their team’s attacking play.
If xG and xA look at final actions or the final phase of attacks, xT is one step back and valuing buildup and progression. Taking it in tandem with xG gives a much more holistic view of player and team performances.
For more info, see this great article by The Athletic.
xG Chain
xG only looks at single shots in isolation, not what happens before. But virtually every shot has something to lead up to it. “Chaining” actions in a possession together can give us more insight into a player’s value to a team’s possession as a whole. xG tries to address the fact that football is a low-scoring game, but xG buildup tries to address the issue of only valuing shots themselves (or only valuing shooters and ignoring players who don’t shoot).
xG Chain looks at unbroken/uninterrupted possessions for a team. Pretend there are 8 actions leading up to a shot worth 0.17 xG. Every player who made an action in this unbroken possession chain get assigned that 0.17 xG as their xG Chain for that possession. This metric helps value players’ contribution to their teams’ quality attacks by crediting everyone involved in the creation of xG, with that xG.
It might not be perfect, of course, because not all actions are created equal. And there will be loads of high-quality actions that don’t receive any values here if the possession doesn’t end in a shot. But it’s a solid metric to use when looking into players’ attacking contribution.
xG Buildup
xG Buildup is a player’s xG Chain excluding their xG and xA. Thus, it tries to find the player’s value in buildup and not be skewed higher towards players that tend to shoot. xG is a big part of xG Chain, and xG Buildup addresses that.
Of course, the same issue with xG Chain exists in xG Buildup: possessions still have to end in a shot, and many high-quality actions and possessions don’t result in shots.
xPoints (expected points)
xPoints is built off xG. The models can differ, just like xG models, so comparing across multiple providers’ xPoints might be beneficial. But they are all essentially trying to figure out how many points a team should have based on the underlying data, or xG, from their games.
Obviously, game performances and who “should have” won or lost is much more complex than a formula using xG. But I actually don’t mind xG-based xPoints models simply because, well... you can’t score or concede, i.e. win or lose, unless you take or face shots. So, the fact that xG only occurs when shots happen isn’t really a limitation for this metric.
Most xPoints models simulate games 100s or 1,000s of times based on xG. Remember, xG is the probability of a shot becoming a goal, so models will “replay” the game’s shots many times to see, on average, how many points each team got.
Possession
Possession is simply the percentage of the game that a team was in possession of the ball. Since the late 2000s, possession has been seen by managers and teams as one of the most important metrics to record high numbers in. The idea was, of course, you score when you have possession, and you can’t concede when you have possession. Thus, having possession is always good.
For much of the last 20 years, teams with more possession tended to win matches. But usually, the best teams tended to have more of the ball, so it’s a correlation and definitely not a causation. Possession isn’t what made them win, although of course that helped.
The trick is what teams do with possession. If they only pass it sideways between defenders, they might get 75% possession, but chances are they won’t create anything. Having possession to limit opponent time on the ball but still having some bite in important moments is how high possession teams win games. Conversely, teams with low possession can still be dominant by stifling opponents and then breaking quickly and effectively.
PPDA (Passes Allowed Per Defensive Action)
PPDA is an estimate of a team’s pressing intensity. Lower values indicate a higher level of pressure here, as having more defensive actions per opponent pass, or allowing fewer passes per defensive action, usually comes as a result of defensive pressure.
It is a very tricky metric, though, because many pressing actions on the pitch aren’t tagged by most providers. And PPDA only really looks at tackles, interceptions, duels, and fouls. For example, closing down an opponent is a pressing action but, without an actual duel or interception, it’s not captured in PPDA.
Teams can also have arbitrarily low PPDA numbers when playing against opponents who boot it long frequently. The opponent is arbitrarily lowering your PPDA value by playing few passes before sending it into duels. Conversely, high pressure levels against teams who are wonderfully skilled at playing through pressure would hide the pressing team’s level of pressure with an arbitrarily high PPDA number. Since, despite pressing intensely, they just can’t get into a duel because of the opponent’s skill on the ball.
Regardless, PPDA compared across a league, and maybe not just a single game, can be a pretty decent estimate of what teams are pressing relatively intensely.
It’s important to note that PPDA typically only considers the front 60% of the pitch, or the middle and final third.
High recovery/Final third recoveries
This one is pretty simple, but a very important metric. Hgih recoveries are possession wins high up the pitch. The two main definitions are either a) in the final third (Opta, for example) or b) within 40 meters of the goal (StatsBomb, for example).
Obviously, winning the ball close to the goal is extremely valuable, so coaching staffs typically put emphasis on this metric, especially in teams trying to dominate the ball or smother the opponent. It can be used together with PPDA to get a more holistic view of a team’s pressing and effectiveness of their press.
Line-breaking pass
A line-breaking pass is a pass that breaks either the opponent’s forward, midfield, or defensive line. These are dangerous passes because they progress the ball up through the opponent and create opportunities (either to progress play by breaking forward and midfield lines, or to create shooting chances by breaking the defensive line).
Impect-Specific Metrics
We use Impect for a lot of the data here at UTAS, and they have several metrics unique to them and not commonly used elsewhere. We’ll explain some of them here.
Bypassed opponents/Packing
“Packing” is Impect’s bread and butter. When an action (pass, carry, etc) progresses the ball past an opponent, they are beaten/bypassed/taken out of the game. This is Packing. The idea is that bypassing opponents takes them out of the game, unable to defend, by being further away from the goal than the ball. To get back into the game, they have to track back or be brought back into the game by a “reverse” action, such as a back pass.
It’s a very simple idea, and is an effective way to measure ball progression. While xT values progression on a static pitch regardless of where opponents were located, Packing is by nature dynamic as it looks at the number of opponents bypassed. A pass within the defensive third is extremely low xT, but if that pass was right after an opponent corner, it might bypass 8 opponents, leaving just 2 recovery defenders and the goalkeeper left to defend. Packing attempts to add context to every action.
A basic example: if there are 9 opponents closer to the goal than the ball when you play a pass, and then 6 when the pass is received, you’ve bypassed 3 players. 9 – 6 = 3. Against a flat 4-4-2, a CB passing between the opposing STs into the DM would bypass 2 players.
Impect treats all players equally for Bypassed Opponents. A number of 3 means 3 opponents were taken out of the game. This could be 2 STs and a CAM, or 3 CBs. But, overall, there is a strong correlation between bypassing opponents and winning games. Impect’s tests have shown that 59% of teams recording a higher number of bypassed opponents than their opponents win that game. So naturally, having players who can progress the ball (or limit opponent progression!) is really important. This concept is what much of Impect’s data is built upon.
Passes and carries naturally get valued by number of bypassed opponents, but we can also look at receiving. Impect has a “bypassed opponents receiving” metric too, that credits the receiver with the same number of bypassed opponents as the passer. This can help analysts and scouts see a player’s ability to find pockets of space, or when comparing DMs, perhaps if one DM drops deep to receive between the CBs or stays a little higher up.
Bypassed defenders/Breaking defense
To address a major blind spot in Bypassed Opponents, Bypassed Defenders only looks at the 5 opponents closest to their goal (including the goalkeeper). Usually these are the keeper, 2-3 CBs, and either both FBs or a FB and a DM. But regardless of who these 5 players are, they are the opponent’s last line of defense at the time of an action.
The correlation of having a better number in this metric than the opponent and winning games is 78%. Nearly four-fifths of the time a team has more bypassed defenders than their opponent, they win the game. But this makes sense: bypassed defenders is, effectively, the quality of a pass or carry breaking the defensive line. Getting in behind the defensive line usually creates a chance and at the very least forces opponents to run backwards; something none of them want to do.
Many managers have recognized that attacking the space behind the back line is one of the most important aspects of the game, and probably the most important thing in open play. Players that can consistently pass, carry, or receive behind the back line cause loads of chaos and either get chances themselves, or create more space for teammates to attack (for example, a cutback cross).
Added teammates/opponents; Removed teammates/opponents
While bypassed opponents and bypassed defenders look at actions that move the ball, Impect also has metrics built on the same Packing concept, that look at events like recoveries, duels, ball losses, and more. All of the events where possession changes teams is valued with adding or removing teammates and opponents.
Take an interception. Impect looks at the location of the interception and how many 1) teammates are ahead of the ball and 2) opponents are behind the ball. If there are 4 teammates in front of the ball, the action is credited with 4 added teammates. And if there are 3 opponents behind the ball, 3 removed opponents.
Ball losses can also remove opponents and add teammates, using that same logic. For example, incomplete passes can be valued like this. If, at the point where the opponent gains possession, there are 2 of your teammates farther from your own goal than the ball, that’s 2 removed teammates. And if there are 7 opponents closer to your goal than the ball now, that’s 7 added opponents.
Non-shot based xG/Packing xG
Impect tries to address a limitation of xG, that a shot must occur, with their “Packing xG” or non-shot-based xG. A shot is not required here, as the metric quantifies all possible shooting positions. Essentially, Packing xG is the xG that would be assigned if a shot happened in that moment. It can take into account where the player is, where the opponents and goalkeeper are, and other info included in standard xG.
Packing xG is a measure of possible shooting positions and might be a better metric to use than xG to look at a team’s performance. It rewards getting into good positions, positions that make the opponent work, so the idea is somewhat of a hybrid between xG and xT.
General Concepts:
Transitions
Transitions are the moments just after winning or losing possession, where teams can be a bit disorganized. Attacking transitions are after winning the ball, and defensive transitions are after losing the ball. The end of a transitional phase is the start of more sustained possession (or defensive) phase, when teams are structured.
Across most leagues, the majority of xG and goals come during transitional phases, so how teams perform in transitional moments is vital. Every team has different intentions during transitions, but the key is how well they implement their intentions. Some teams counter attack, some teams recycle the ball; some teams counter press after losing it, others drop back (but the closest player to the ball should normally close down the opponent to give their teammates time to get into their defensive shape).
Second balls
The second ball phase (important to Neil Harris’ U’s) is the chaos phase where no team has possession. While there’s not always a second ball phase before a transitional phase, transitions happen after second balls.
We can think of second balls as the “mixer”. When goal kicks are taken long into midfield, or a long pass into a duel. Second balls themselves are those rebounds, and players fight to pick it up. It’s important to focus on winning second balls for several reasons, but obviously the biggest reason is that you win possession. You can keep possession if your team is the one starting the second ball, or win it and maybe have a dangerous transition if the opponent starts it.
Second ball moments can suck players in, leaving opponents disorganized; which is why it’s important for coaches to train their players how to position themselves both to 1) be best set up for picking up the second ball and 2) also be conscious not to be too far out of position, in case the opponent wins the ball.
Expansion:
Every metric, every statistic, has limitations. As we recognize these limitations (and the best people will talk about them openly), we create new metrics that are developments on older metrics to address the limitation(s). Two important ones would be post-shot xG (an extension of xG) and field tilt (an extension of possession).
Post-shot xG (psxG)
xG provides goal probability based on when the shot is taken, but post-shot xG looks at what happens after the shot is taken. It considers the placement of the shot relative to the goalkeeper (further from them = higher chances of scoring), the spin rate and projected trajectory of the ball, the speed of the shot, and players that may be blocking the keeper’s view.
xG values the shooting position, while psxG values the shot placement. We can compare the xG and psxG of a shot to see how good a player might be at placing shots. Having a higher psxG than xG may indicate good finishing or shot placement, and less psxG than xG may indicate poor finishing or shot placement.
Naturally, since psxG is the probability of scoring given the placement of the shot, psxG is 0 for off-target shots.
It’s important to use psxG when evaluating goalkeepers. The psxG they face is a much better indicator of the number of goals they should be expected to concede than xG. If a keeper faced 2.8 xG in a game but every shot was sent into row Z, did they really save almost 3 goals? But if they faced 2.8 psxG and kept a clean sheet, that would be a crazy feat.
xG and psxG are similar, but measure different things. When you have access to both, it’s usually best to use them together.
Field Tilt
Field Tilt is possession when excluding the middle third. It’s the share of a team’s passes in the final third vs theirs plus their opponents’ passes in the final third. So, if team A had 75 passes in the final third while team B had 25 in their final third, team A’s field tilt is 75/(75+25) = 75%.
Field tilt is a much better metric than possession to see how “dominant” a team was on the ball. And comparing possession and field tilt offers some interesting insights. A team with high possession but low field tilt probably languished in their own half and struggled to break forward.
Expected Assists ←|→ Expected Assisted Goals
The limitation of xA, that it’s not really the quality of the pass but the quality of the shot (and at the mercy of the shooter) is why some providers have started introducing two xA metrics.
xA as discussed above becomes xAG: expected assisted goals. But xA for these providers is the probability that a completed pass becomes a goal assist. A true “expected assists” metric. It doesn’t require a shot to actually be taken, and every completed pass would have an xA number. But only shot assists would have xAG numbers.
This is a relatively newer split, and the majority of xA you will see is still just the xG of shots a player assisted.