The Enclosure: Methods and Data
Dataset
4,147 finished Premier League fixtures across 11 seasons:
| Season | Fixtures | Notes |
|---|---|---|
| 2015/16 | 380 | |
| 2016/17 | 380 | |
| 2017/18 | 380 | |
| 2018/19 | 380 | |
| 2019/20 | 379 | 1 match missing |
| 2020/21 | 380 | Covid: restricted/empty crowds |
| 2021/22 | 377 | 3 matches missing |
| 2022/23 | 380 | |
| 2023/24 | 380 | |
| 2024/25 | 380 | |
| 2025/26 | 271 | Season in progress (GW27) |
Total: 4,147 fixtures. All per-match averages are calculated as per team per match (i.e. total across both teams divided by 2× match count).
Data sources
WhoScored / Opta event statistics
-
All match-level statistics sourced from Opta event data via WhoScored.
Stored locally in
src/data/pl-meta/whoscored-events.json. -
Per-team per-match fields used:
take_ons_total,take_ons_successful,key_passes,big_chances_created,through_balls,clearances,interceptions,passes_total,passes_successful,long_balls_total,long_balls_successful,aerials_total,aerials_won,crosses_total,corners_taken -
Shot breakdown fields:
shots.total,shots.regular_play,shots.regular_play_goals,shots.fast_break,shots.fast_break_goals,shots.set_piece,shots.set_piece_goals,shots.from_corner,shots.from_corner_goals,shots.direct_freekick_goals,shots.big_chances,shots.big_chances_goals -
Available but not used in this piece:
tackles_total,tackles_successful,fouls,yellow_cards,red_cards,formation,attendance
mclachbot final third access data
- Final third entries per team per season sourced from mclachbot (2021/22 to 2025/26 only)
- Used for the convergence section: coefficient of variation, City's ratio to league average, top-5 vs bottom-5 gap
- Not available for seasons before 2021/22
Definitions
- Take-on success rate = successful take-ons / total take-on attempts × 100
- Pass accuracy = successful passes / total passes × 100
- Open play goals =
regular_play_goals(excludes fast break, set piece, corner, free kick, and penalty goals) - Set piece goals =
set_piece_goals+from_corner_goals+direct_freekick_goals(excludes penalties) - Fast break Big Chance rate = proportion of fast break shots flagged as Big Chances (Opta qualifier 214) compared to the same proportion for regular play shots
Key findings
Take-on success rate (n=4,147, all 11 seasons)
| Season | Matches | Attempts/tm/m | Success rate |
|---|---|---|---|
| 2015/16 | 380 | 16.49 | 53.7% |
| 2016/17 | 380 | 16.64 | 54.1% |
| 2017/18 | 380 | 15.23 | 56.2% |
| 2018/19 | 380 | 16.16 | 50.9% |
| 2019/20 | 379 | 15.28 | 51.2% |
| 2020/21 | 380 | 16.29 | 52.2% |
| 2021/22 | 377 | 17.47 | 51.8% |
| 2022/23 | 380 | 19.51 | 43.1% |
| 2023/24 | 380 | 19.91 | 42.9% |
| 2024/25 | 380 | 19.18 | 40.8% |
| 2025/26 | 271 | 18.04 | 37.9% |
Seven seasons of stability (50.9%–56.2%), then a single-season 8.7pp collapse in 2022/23. All 17 teams overlapping both 2021/22 and 2022/23 saw their success rate decline. No recovery in the four seasons since.
Clearances per team per match (n=4,147)
| Season | Clearances/tm/m | Interceptions/tm/m |
|---|---|---|
| 2015/16 | 28.1 | 17.0 |
| 2016/17 | 25.3 | 12.6 |
| 2017/18 | 22.0 | 11.2 |
| 2018/19 | 20.3 | 11.0 |
| 2019/20 | 19.5 | 10.6 |
| 2020/21 | 19.1 | 9.8 |
| 2021/22 | 19.7 | 9.8 |
| 2022/23 | 20.5 | 8.3 |
| 2023/24 | 21.8 | 8.8 |
| 2024/25 | 23.9 | 7.9 |
| 2025/26 | 28.3 | 8.3 |
Clearances halved from 28.1 to 19.1 during the possession era (2015/16– 2020/21), then returned to 2015 levels by 2025/26. Interceptions fell continuously from 17.0 to 7.9, suggesting the mechanism is active pressure suppressing risky passes rather than static positioning.
Passing (n=4,147)
| Season | Pass accuracy | Long balls/tm/m | Crosses/tm/m | Corners/match |
|---|---|---|---|---|
| 2015/16 | 76.3% | 80.0 | 20.6 | 10.8 |
| 2016/17 | 76.8% | 80.9 | 19.8 | 10.4 |
| 2017/18 | 76.9% | 78.3 | 18.7 | 10.3 |
| 2018/19 | 77.9% | 73.1 | 17.5 | 10.3 |
| 2019/20 | 78.1% | 70.9 | 19.0 | 10.7 |
| 2020/21 | 79.6% | 66.8 | 18.1 | 10.1 |
| 2021/22 | 78.9% | 66.5 | 17.8 | 10.4 |
| 2022/23 | 79.5% | 63.5 | 17.2 | 10.1 |
| 2023/24 | 81.0% | 61.5 | 17.3 | 10.8 |
| 2024/25 | 81.7% | 57.4 | 17.6 | 10.2 |
| 2025/26 | 80.2% | 63.1 | 18.2 | 9.8 |
Pass accuracy rose from 76.3% to 81.7% over 10 seasons. Long balls per team per match fell by ~28% (80.0 to 57.4). Crosses and corners remained broadly stable at 17–21 and 10–11 respectively. The adaptation to enclosure was conservative passing, not route one.
Creativity (n=4,147)
| Season | Key passes/tm/m | Big chances/tm/m | Through balls/tm/m | Combined |
|---|---|---|---|---|
| 2015/16 | 8.41 | 1.24 | 1.19 | 10.84 |
| 2016/17 | 8.75 | 1.30 | 1.47 | 11.52 |
| 2017/18 | 8.03 | 1.29 | 1.33 | 10.65 |
| 2018/19 | 8.61 | 1.43 | 1.28 | 11.32 |
| 2019/20 | 8.58 | 1.35 | 1.33 | 11.26 |
| 2020/21 | 8.48 | 1.49 | 1.52 | 11.49 |
| 2021/22 | 8.96 | 1.38 | 1.57 | 11.91 |
| 2022/23 | 8.87 | 1.54 | 1.61 | 12.02 |
| 2023/24 | 9.82 | 1.83 | 1.75 | 13.40 |
| 2024/25 | 9.23 | 1.91 | 1.70 | 12.84 |
| 2025/26 | 8.34 | 1.59 | 1.71 | 11.64 |
2023/24 was the peak: key passes 9.82, big chances 1.83, through balls 1.75. By 2025/26 all three had fallen. The combined creative output dropped from 13.40 to 11.64, below the 2021/22 starting level.
Shot sources and goals (n=4,147)
| Season | RP shot % | FB shot % | RP goals/m | SP goals/m | SP goal % |
|---|---|---|---|---|---|
| 2015/16 | 71.9% | 1.2% | 1.79 | 0.61 | 22.6% |
| 2016/17 | 69.9% | 1.5% | 1.79 | 0.70 | 25.0% |
| 2017/18 | 70.7% | 2.1% | 1.80 | 0.48 | 17.9% |
| 2018/19 | 68.3% | 3.4% | 1.77 | 0.61 | 21.6% |
| 2019/20 | 67.4% | 4.3% | 1.74 | 0.58 | 21.3% |
| 2020/21 | 69.0% | 3.2% | 1.75 | 0.45 | 16.8% |
| 2021/22 | 68.0% | 3.7% | 1.84 | 0.61 | 21.7% |
| 2022/23 | 67.3% | 4.0% | 1.82 | 0.58 | 20.4% |
| 2023/24 | 67.5% | 5.0% | 2.16 | 0.65 | 19.8% |
| 2024/25 | 64.7% | 6.9% | 1.85 | 0.61 | 20.8% |
| 2025/26 | 61.3% | 7.0% | 1.56 | 0.75 | 27.2% |
Open play's share of all shots fell from 68.0% (2021/22) to 61.3% (2025/26). Open play goals per match fell from 1.84 to 1.56. Set piece goals (set piece + corner + free kick, excluding penalties) rose from 0.61 to 0.75. The set piece share of all goals went from 22% to 27%.
Fast break shots rose from 1.2% of all shots in 2015/16 to 7.0% in 2025/26, but the fast break Big Chance rate collapsed from 3.05× the regular play rate to 1.31×. Even the distance advantage vanished: fast break shots averaged 3.6 Opta grid units closer to goal than regular play in 2015/16; by 2025/26 the gap was negligible.
Visualisation methodology
Charts
- The Shrinking Common (pitch viz): take-on success rate mapped to a shrinking green rectangle within a football pitch outline. Scale: 52% → ~95% of pitch area, 38% → ~55%. Season stepper with auto-play on scroll.
- The Divergence (bar chart): five horizontal bars (2021/22–2025/26) showing take-on success rate as green fill. Eight deterministically placed cow icons per bar (seeded PRNG, seed=42).
- The Clearance Arc (brick column chart): all 11 seasons. Each column is a stack of bricks, one per whole clearance. Three key columns labelled: first season, possession era trough, and current season.
- The Creativity Arc (stream chart): symmetric Catmull-Rom spline showing combined key passes + big chances + through balls for the last 5 seasons. Arrow pattern overlay.
- The Narrowing (strip plot): per-team take-on success rate for 5 seasons. Each dot is one club. Man City highlighted. Range whisker (min–max) shown per row.
Stepper charts
The pitch viz and divergence bars use 5 seasons (2021/22–2025/26). The clearance chart uses all 11 seasons. This is intentional: the take-on collapse is a post-2022 story best told with tight focus, while clearances require the full arc to show the possession era trough and reversal.
Pre-computed vs hardcoded data
Season aggregates (take-ons, key passes, big chances, through balls,
clearances, interceptions, pass accuracy, long balls, aerials) are
computed at build time from the source JSON and passed to the client via
a <script type="application/json"> element. The
client script includes hardcoded fallback data for reliability.
Shot source breakdowns (regular play, fast break, set piece goals), crosses/corners, final third access, and fast break Big Chance rates are referenced in prose as fixed statistics, not computed dynamically. These will need manual updates when the data refreshes.
Data not used
The following fields are available in the dataset but intentionally omitted from this piece:
- Tackles (total and successful): declined from 19.4 to 16.9 per team per match over the period. Directionally interesting but muddies the clearance/interception narrative.
- Aerials (total and won): fell from 31.5 to 26.0 per team per match (2015/16–2024/25), then spiked to 32.2 in 2025/26. Potentially connected to the set piece rise but not explored.
- Fouls: broadly flat at 20–23 per team per match. No narrative value.
- Cards (yellow and red): available but not relevant to the enclosure thesis.
- Formation: available per match. Could support a future analysis of tactical shape convergence.
- Attendance: available per match. Potentially interesting for the Covid season comparison but not used.
Caveats
- Opta definitional change: the 2022/23 take-on collapse was validated across all 20 teams and cross-referenced against carry data from five European leagues. No Opta definitional change was identified, though the possibility cannot be entirely excluded.
- 2025/26 incomplete: 271 of 380 fixtures (GW27). All 2025/26 statistics are provisional. Per-match averages should be stable barring unusual fixture clustering.
- 2019/20 and 2021/22 missing matches: 1 and 3 matches missing respectively from the WhoScored dataset. Negligible impact on season averages.
- 2020/21 Covid season: restricted/empty crowds across most fixtures. This affects crowd-sensitive metrics but the core statistical trends (take-on rate, clearances) show no anomaly in that season.
- Final third access data limited: mclachbot data only available from 2021/22. The convergence claims cannot be verified against earlier seasons.
- Fast break Big Chance rate: computed from
shots.big_chances(total) andshots.fast_break,shots.regular_play(per play type). The Big Chance flag is not broken down by play type in our data — the prose figures were derived from a separate analysis of the underlying Opta event stream. - No xG model: this piece uses Opta's binary Big Chance flag rather than a continuous xG model. Big Chance thresholds may vary slightly across seasons.
- Hardcoded prose statistics: shot source breakdowns, fast break analysis, final third access, and clearances-per-cross are hardcoded in the article text and will not auto-update when the underlying data changes.