Football/ Ayush Pawar

03 / Opposition Complete

Do some players actually thrive against better teams?

How much of a player's output survives when the opponent gets stronger?

  • 2015/16 — 2024/25Seasons
  • 5Leagues
  • 525KPlayer-matches

Finding

Player attacking output falls substantially as opponent strength increases — but we found no reliable evidence that some individuals consistently resist that effect.

02Why it matters

Raw statistics make player comparison deceptively simple. A player who piles up numbers against weak sides isn't necessarily better than one who produces a little less against strong ones.

And if some players really rise to big occasions, that would be worth knowing before a cup final — or a transfer.

Can we separate player level from opponent difficulty?

03What we did

  1. Player output
  2. Team / season context
  3. Venue
  4. Opponent strength
  5. Adjusted player output

We're comparing a player's output against different opponents rather than simply comparing raw totals.

Technical detailHow the estimate is built

is a sequential market rating from pre-match odds: each match uses only earlier matches. Checked against an xG rating, Elo, opening odds and points per game; 0 changed ratings in 155,282 checks.

Within-player with , plus venue, month and kickoff hour; errors clustered by match.

Individual resistance: calibrated standard errors, between-player spread and ; .

04What we found

A stronger opponent hurts everyone.

falls 13.8% for every +1 increase in .

Change in output per +1 SD of opponent strength

Same player, team and season, venue accounted for. Lines are 95% intervals.

Every attacking metric falls by 10–16% for each step up in opposition; xG by 13.8%.

Study 03 · sequential market rating · 2015/16–2024/25

Same player, team and season, after accounting for venue: −15.1%, shots −10.5%, key passes −11.4%, goals −15.8%.

How big is that?

Against the strongest fifth of opponents, xG is 33% lower than against the weakest fifth (xA −35%).

xG by opponent strength, relative to the weakest fifth

Opponents in five equal groups by pre-match rating. Weakest fifth = 100.

Against the strongest fifth, the same player produces 33% less xG than against the weakest.

Study 03 · same player, team and season

The same in every position.

The effect is the same across positions and home/away, and unchanged behind closed doors — unlike home advantage.

xG per +1 SD of opponent strength, by position

Lines are 95% intervals.

Defenders, midfielders and forwards lose about the same share against stronger opponents.

Study 03 · robustness by position

Leagues differ a little.

Steepest in Serie A (xG −16%, xA −18% per SD). In La Liga and Ligue 1, shot and key-pass volume barely drops against strong sides (about −7% to −8%).

xG per +1 SD of opponent strength, by league

Lines are 95% intervals.

Steepest in Serie A (−16.3%), shallowest in La Liga (−12.0%).

Study 03 · per-league models

But are some players different?

Maybe the average effect is strong, but certain players are unusually resistant to elite opposition.

0of 23,022player × metric estimates show reliable resistance

We found no reliable individual player × metric estimates showing consistent resistance to stronger opposition. Season-to-season correlation of a player's “resistance”: 0.012.

The practical rule.

To predict a player's output against elite opponents, use his overall opponent-adjusted level, not his record against elite teams: xG r 0.84 vs 0.75.

Which predicts output against elite opponents better?

Correlation with output in later matches against elite teams.

For every metric, a player's overall adjusted level predicts his output against elite teams better than his past record against them.

Study 03 · out-of-sample check

Adjusting a season changes little — except where it matters.

Season adjustments are typically ±2% because everyone plays everyone twice; up to ±10% for players used unevenly, e.g. Callum Wilson 2023/24 (−9.9%). Bruno Fernandes's elite chance creation survives: 0.467 → 0.460 xA per 90 in 2022/23, 99.6th percentile.

Adjustment matters most for partial seasons, single matches and cross-league comparisons.

Bruno Fernandes, xA per 90: raw and opponent-adjusted

Each season, before and after adjusting for the opponents he faced.

Adjustment barely moves a full season — Bruno's 2022/23 goes from 0.467 to 0.460, still the 99.6th percentile.

Study 03 · season adjustments

05What surprised us

  1. What the raw data suggested

    In the raw data, about 29% of players look like “flat-track bullies” and about 26% like “big-game players”.

  2. What happened after we checked

    That is exactly what noise produces. With proper intervals, one player in about 4,800 shows a reliable pattern (for shots) — what chance alone would produce — and after correcting for multiple tests, none of the 23,022 estimates survives.

  3. Lesson

    Raw “big-game” records reflect context and noise rather than a stable player trait.

06What it means

  • Opposition is one of the biggest contexts in the numbers. Adjust for it — especially for partial seasons, single matches and cross-league comparisons.
  • To judge how a player will do against elite teams, look at his overall adjusted level.

07Limitations

  • Opponent strength is the betting market's pre-match view of a team; it is validated against four alternatives, but it is still a rating, not a fact.
  • Output only — the data cannot see defending, pressing or possession.
  • Cup matches and fixture congestion are not in the data.
  • Observational data: effects are associations within carefully matched comparisons.

SourcesUnderstat · football-data.co.uk · Open-Meteo ERA5 · Wikipedia/Wikidata · OpenStreetMap
Analysis period2015/16 — 2024/25 · frozen dataset · Methodology

08The next question

With context removed, can we finally compare players — across leagues?

04 / Similarity