SoccerAnalytics
Data Deep-Dives

The 2026-27 Premier League, By the Numbers

A living data ledger for the Premier League season — every number computed from completed matches, refreshed before each build. Zero predictions, zero invented numbers.

By C. B. Zakarian 6 min read

This page is a living snapshot of the 2026-27 Premier League, and right now it is proudly, deliberately empty of results — because the season hasn't kicked off yet. The opening round is scheduled for August 21, 2026 (ESPN's fixture list currently opens with Arsenal hosting Coventry City that evening), and from the first final whistle onward this page will fill with numbers computed from completed matches only: goals per game, the draw rate, home advantage, and the running test of the shots-on-target rule. Until then, what you get is the structure — what will be tracked, how, and from what source. What you will never get here is a prediction dressed up as a number.

A note on sourcing, before anything else. This page runs on the same machinery that carried this site's 2026 World Cup coverage from the opening match to the final: a refresh script (data_layer/refresh_epl.py) re-pulls ESPN's public scoreboard and match-summary feeds before each site build, keeps only matches with state=post — actually finished — and writes them to a snapshot this site serves at /data/epl_2026_27_results.json, so you can check every figure against the raw file yourself. Last refreshed August 16, 2026. The snapshot currently holds 0 completed matches, which is exactly right: nothing has been played, so there is nothing to report. That zero is the whole editorial policy of this site in one character.

The shape of the season

The structural facts, verified against the fixture list the feed returns today: 20 clubs, each playing every other club home and away — 38 rounds, 380 league games — across a window running from August 21, 2026 to a final round currently listed for May 23, 2027. ESPN's feed presently carries 370 of those 380 fixtures with concrete dates; fixture lists fill in through the season as kickoff slots are finalised for broadcast, so that count can lag the full 380 without anything being wrong.

Compared to the World Cup this site just finished covering, that is a different statistical animal — and a better one. The World Cup gave us 104 games in five weeks and made every conclusion fight against small-sample noise. A league season delivers nearly four times as many matches, played by a stable set of teams who meet each other twice, which is the closest thing observational sports data offers to a controlled experiment. Patterns that had to be flagged as “suggestive” on 28 knockout games get a real trial here.

What this page will track

  • Goals per game. The headline rate, computed from completed matches and nothing else, with its running average charted as the sample grows. For context: recent Premier League seasons have run around three goals a game, and 2023-24's widely-reported 1,246 goals (3.28 a game) stands as the modern high-water mark. Those context figures are public reporting, not this site's pull — every number this page states as its own will come from the snapshot linked above.
  • The shots-on-target rule. The signature finding of the World Cup series: in decided games, the team that matched or beat its opponent on shots on target won 70 of 80 (87.5%), including 24 of 28 decided knockouts. That ledger, its exceptions, and its logic get their definitive statement here — and this page is the live out-of-sample test. Every decided Premier League match gets scored against the rule as it completes.
  • Home advantage. A World Cup, played at neutral-ish venues, can't measure it. A league season is the canonical dataset for it. This page will keep the running home-win / draw / away-win split; for what the number has historically been and why it has drifted, see home advantage, quantified.
  • The draw rate. Knockout football suppresses draws by design; league football doesn't. The share of games finishing level is one of the cleanest signatures of how competitive a league's mid-table really is, and it needs no model — just counting.
  • The shape of scorelines. Not just the average but the distribution: how many goalless games, how many blowouts, and whether the extremes are drifting. The World Cup coverage showed a mean can hide a polarised distribution; the same histogram discipline applies here.

Where the season stands

  • The season hasn't kicked off. 0 of the 380 games are in the books, so every number this page will track stands at zero — honestly. The first completed match replaces this launch state with the real running figures; nothing appears here before it happens on a pitch.
  • What changes as matches complete. Once the snapshot holds finished games, this page re-renders with the real numbers — goals per game, the draw and home-win rates, the biggest scorelines, and the shots-on-target ledger — plus the distribution chart, drawn from the same file. The prose and the chart are generated from one snapshot, so they cannot drift apart.
  • What never changes. No projected table, no title odds, no “expected finish”. There are good sites for forecasts; this page is a ledger, and ledgers only record what happened.

The method, stated before the data exists

Writing the methodology down before any matches are played is the point — it means the method can't quietly bend around the results. So, on the record:

  • Source. ESPN's public scoreboard feed for the Premier League (soccer/eng.1), with each completed match enriched from the public match-summary endpoint: possession, shots, shots on target, passes, corners, fouls and cards for both teams. Raw JSON, parsed in code — the same pipeline, endpoint-for-endpoint, as the World Cup body of work.
  • Inclusion rule. A match enters the dataset when its state is post — finished. Postponed, abandoned or in-progress games do not exist as far as this page is concerned until they finish.
  • Fail-soft refresh. Each pull is compared against the previous snapshot and can only grow the match count; a network error or a short feed response restores the last good file rather than silently shrinking the season. The World Cup version of this guard ran for five weeks without regressing once.
  • One source, admitted. Scores are as verifiable as data gets, but the box-score stats come from a single provider, and providers occasionally disagree on possession or shot counts by a point or two. When a number matters, the raw snapshot is one click away.

The honest limits, up front

Early-season league numbers are famously treacherous: three rounds in, someone will be averaging three points a game who won't finish top eight, and some genuinely good side will sit in the relegation places. This page's defence is structural — it reports rates with their sample size attached and refuses to extrapolate. When the season is three games old, it will say “30 matches” loudly and draw no conclusions a 30-match sample can't carry. The numbers will appear here as the matches complete; nothing more, and nothing sooner.

Sources & further reading

More from Data Deep-Dives

Data Deep-Dives

The Shots-on-Target Rule

In decided 2026 World Cup games, the team matching or beating its opponent on shots on target won 70 of 80 — 24 of 28 knockouts. The ledger, and the next test.

7 min read