Expected Goals Against

Every shot faced, added up

Expected goals against mirrors xGF: take the xG value of every shot a team has faced this season and sum it. A defence conceding chances worth 1.1 xG a match sits at roughly 42 across a 38-match season, regardless of how many of those shots actually went in.

Own goals are excluded, and the reason is mechanical rather than a matter of judgement. The xG model scores a shot from its distance, angle, body part, assist type and the defensive pressure around it. An own goal has none of those inputs. There is no shooter, no angle chosen, no assist that created it, nothing for a shot-quality model to evaluate. It is not a shot, so it gets no expected goals value, and it stays out of xGA while still counting in the real goals conceded column. That single exclusion is the single biggest reason the two figures drift apart over a season.

Reading the gap to goals conceded

A strong xGA sits next to a poor goals-against record. A defence restricting the opposition to 0.9 xG a match while conceding at a 1.4 a match rate is not the structural problem the scoreline suggests. That gap points at shot stopping or plain misfortune, not at a back line that keeps getting exposed. The structural read and the results read disagree here, and the structural one is the more durable of the two.

A defence looks fine on the scoreboard but carries a high xGA. Few goals conceded so far, but the underlying chances allowed keep climbing. That is a warning sign worth taking seriously. The goalkeeper or the finishing luck of opponents has been carrying a defence that is not actually restricting good chances, and that cannot hold indefinitely.

Two mid-table defences, one number that tells them apart. Both concede around 1.3 goals a match. One has an xGA near 1.0, the other near 1.6. The first is a well-organised defence riding some bad luck. The second is getting outplayed and riding some good luck. Real goals conceded cannot distinguish them. xGA does.

xGA settles whether the defence is structurally sound

Set aside the goalkeeper's form and the run of finishing the team has faced. What is left is one question: how good are the chances this defence lets the opposition have, repeatedly, over a full season. A back line can look shaky on the scoreboard for a month while its underlying shot prevention stays strong. That is the read xGA gives, and it is the one worth trusting over goals conceded alone.

Where this gets misread

• "xGA measures the goalkeeper." It measures the defence's shot prevention: how often, and from where, the opposition gets to shoot. The goalkeeper's contribution is what explains the gap between xGA and actual goals conceded, not what xGA itself is counting.
• "Own goals should count towards defensive quality." They are not shots created by the opponent's build-up play in the way the model measures, so they carry no xG value by design. That is a deliberate scope decision, not an oversight in the data.
• "Low xGA always means clean sheets follow." It means fewer, lower-quality chances conceded on average. It is not a guarantee for any single match, where one well-taken shot can still beat a defence that is doing everything right.