← Blog

The team with the ball misses more. It is not shooting from further out.

25 August 2026 · our own archive: 17,219 matches, 32,380 club-matches, 54 competitions, 116 league-seasons

Same club, same season. On days it holds eight points more of the ball than it usually does, it scores 0.81 goals for every expected goal. On days it holds eight points less: 1.22. That is 0.55 goals a match between the two, t = 32, and it moves in one direction across all five bands. The share of shots taken inside the box moves by 1.8 points. The usual explanation is the one thing that barely changes.

Where this came from

Under our page on League One's second round, someone offered two explanations for why the home sides took 166 shots worth 16.70 expected goals and scored 10. First, that away sides sitting deep get better shots on the counter. Second, that the side chasing the game pads its expected goals with long-range attempts instead of real chances. Both are testable, so we tested them, and then tested the thing they were reaching for.

The comparison is a club against itself

Comparing possession sides to counter-attacking sides compares good teams to bad ones, and good teams finish better, so the answer would be about squad quality rather than about the ball. Instead every club-match is placed against that same club's own average possession in that same league and season (clubs with at least ten matches). A team that normally has 52 per cent and had 63 that day sits in the top band; the same team on 44 sits in the bottom one. Squad quality, division and season are held still by construction.

possession that dayshotsxG per shotgoals per xG
8+ below own average10.00.1171.22
3 to 8 below11.70.1141.12
within 3 of own average12.80.1111.03
3 to 8 above13.90.1070.94
8+ above own average15.30.1000.81

Goals per expected goal, by how much of the ball a club had that day relative to its own norm. 32,380 club-matches. Expected goals in our archive is calibrated: across the whole sample clubs score 1.007 goals per expected goal at home and 1.009 away, so a figure of 1.00 means par.

It is not where they shoot from

If dominant sides padded their expected goals with shots from distance, the share of attempts taken inside the box would fall as possession rises. It falls by 1.8 points, from 65.5 per cent to 63.7. Two other things move a great deal more.

possession that dayin the boxblockedon targetgoals per SoT
8+ below own average65.5%24.8%38.0%37.6%
3 to 8 below65.3%25.9%36.1%35.3%
within 3 of own average64.7%26.5%35.2%32.3%
3 to 8 above64.1%27.5%33.8%29.7%
8+ above own average63.7%29.2%31.3%25.6%

Blocked shots as a share of all shots, and goals per shot on target, by the same bands.

The share of shots that never reach the goal at all climbs from 24.8 to 29.2 per cent, and accuracy falls from 38.0 to 31.3 per cent. The largest move of all is the last one: goals per shot on target drops from 37.6 to 25.6 per cent, a fall of 32 per cent. Expected goals is measured before the ball is struck. It can see that a shot was taken from twelve yards; it cannot see the two defenders between the ball and the goal, or the keeper already set. That is the part a packed penalty area takes away, and it is the part expected goals never charges for.

The obvious objection: teams that are losing have more of the ball

A side that concedes first will chase the game, so possession and a conceded goal arrive together and the effect could be an artefact of the scoreline. Two checks.

First halves only. Possession measured against the club's own first-half average, goals and expected goals from the first half alone. 13,756 matches. Second, matches level at half-time. Everything from the full match, but only where the score was tied at the break. 6,904 matches.

possession that dayall matchesfirst halveslevel at half-time
8+ below own average1.221.221.00
3 to 8 below1.121.100.93
within 3 of own average1.031.030.83
3 to 8 above0.940.950.80
8+ above own average0.810.840.69

Goals per expected goal in each sample. The level-at-half-time column sits lower throughout because tied first halves are low-scoring matches by selection; what matters is the gap between the top and bottom row, not its height.

The gradient survives both. First halves keep 94 per cent of it, and matches level at half-time keep 75 per cent. Chasing the game is part of this. It is not most of it.

Back to the round that started it

League One, matchday twohomeaway
possession56%44%
shots166136
share of shots in the box68.1%60.3%
xG per shot0.1010.094
big chances created1816
shots on target27.1%36.8%
expected goals16.7012.78
goals1015

All twelve matches, League One matchday two, 20 and 22 August 2026.

Both offered explanations fail here in the same direction. The home sides took 68.1 per cent of their shots from inside the box against the away sides' 60.3, and League One's long-run figure for home sides is 63.9 — they shot from better places than home sides normally do, not worse. Their expected goals per shot was 0.101 against the away sides' 0.094, and they created 18 big chances to 16. What actually broke was accuracy: 27.1 per cent of home shots hit the target against a long-run 33.6.

The effect in this article is real and it applies — the home sides did have 56 per cent of the ball. It is also small. At that much possession the expected correction is 0.94, which turns 16.70 expected goals into 15.7 rather than 16.70. They scored 10. So the structural effect accounts for about 15 per cent of the shortfall and variance accounts for the rest: 10 goals or fewer from 15.7 is a one-in-11 round, or one in 18 before the correction. And the away sides, whose finishing was supposedly the story, scored 15 from 12.78 — something that happens 30 per cent of the time.

What to do with it

The numbers in the first table are a correction factor. When a club has just dominated the ball and its expected goals looks damning, multiply by the band it played in before deciding it was robbed; when a club sat deep and looked wasteful, do the same in the other direction. The correction is worth roughly 0.55 goals a match between the extremes, which is not nothing and is also not the whole gap in any single match.

What this is not. This says the side with the ball converts its chances at a lower rate than expected goals implies. It does not say possession is bad — the top band still takes 15.3 shots a match against the bottom band's 10.0 and still scores more per match. The volume wins more than the conversion loses. It says the two are being double-counted when a chance-based number is read as a verdict.
Method, in full. Every match where our archive holds expected goals and possession for both sides: 17,219 matches across 116 league-seasons and 54 competitions. Club-matches are excluded when the club has fewer than ten matches in that league-season, because its own average would not be an average. A handful of source rows carry impossible expected-goals values — one match arrives with 780 — so rows above eight expected goals, or above 0.4 expected goals per shot, are dropped. Without that filter the whole archive appears to underperform expected goals by 28 per cent, which is a broken feed, not a finding.
Take the numbers with you. Every table on this page is in posxg.xlsx — the bands, the two controls and the League One round.
Sources. Match statistics from our own feed, stored in our own database. Every rate here is our own calculation. No third-party table is reproduced.

← Back to the blog