Football Statistics Limitations: Sample Size, Role, Game State and Definitions

A higher expected goals total does not mean a team deserved to win. It measures shot quality, not the defensive structure that forced those shots wide. A single-match xG delta of 1.2 to 0.8 is a measurement under specific, hidden conditions. It is not a verdict.

Opta and CIES Football Observatory provide technically precise data. The trouble starts when an analyst draws a conclusion the metric cannot support. An xG tally tells you where shots were taken. It does not tell you why a low block prevented better ones. Possession percentage describes who had the ball. It does not describe who controlled the match. Every number you see is a product of definitions chosen by a data provider. A different definition would yield a different story.

A single match is noise. A team records 0.3 xG and wins 1-0 against a 2.1 xG opponent because a deflected cross loops in. That delta tells you nothing about attacking quality. It is one data point in a sequence that needs 15 to 20 matches to stabilise.

Model variation compounds small samples. Two expected goals models can disagree by 0.3 xG on the same shot. One factors in defender positioning; the other weights only shot location and body part. Over 38 league matches these differences regress toward a stable mean. Over one cup tie they do not. The sample size problem is most acute in knockout competitions, where entire seasons are judged on two legs.

A defender who wins every aerial duel in a single fixture has not proved elite ability. He probably faced a forward who never challenges in the air. A goalkeeper's save percentage across three matches can swing from 90 per cent to 55 per cent on a handful of deflections. Wait for a full season of playing time before trusting any performance metric as a signal rather than a fluke.

A centre-back in a high press records fewer interceptions than one in a low block. He is not worse. His job is to step into midfield and force the pass, not to sit and pick it off. PPDA measures how quickly a team closes down. A defender who triggers the press gets no credit if a teammate makes the tackle. The number captures the outcome, not the role that created it.

Possession percentage is the clearest example. A team with 35 per cent possession can win by design if its defensive structure is coherent. The metric counts completed passes, not what those passes achieved. A holding midfielder who plays sideways and backward all match posts a 92 per cent pass completion rate. The playmaker who attempts the line-breaking ball posts 72 per cent. The number punishes risk and rewards safety.

Tracking data can isolate off-the-ball movement. Most public analytics still rely on on-ball events. A striker who drags two centre-backs wide for 90 minutes creates space others exploit. He records zero shots, zero key passes, and a low match rating from event data. You cannot evaluate a player without defining what his role asked him to do. No box score tells you that.

A team 2-0 down at half time faces a changed opponent. The leading side drops into a compact shape, concedes territory, and stops pressing high. The trailing side dominates possession and racks up shots against a set defence. Every metric collected after the first goal is a product of that game state, not of the teams' underlying quality.

Possession percentage in the second half of a match a side is winning measures the opponent's urgency. It does not measure the leader's passing ability. Expected goals accumulated while chasing a deficit come against a low block that is happy to concede low-quality shots from range. The number looks identical to a dominant attacking performance. The conditions that produced it are the opposite.

A pressing intensity figure like PPDA collapses when a team stops pressing after taking a lead. Compare a side's PPDA when the match is level to its PPDA when ahead by two goals. You will find a gap of 3 to 5 passes per defensive action. Analysing a full-match average without splitting by game state is like averaging a sprinter's speed across a marathon.

Metric	What It Measures	What It Cannot Prove
Expected goals (xG)	Shot quality based on location, body part, assist type and defensive pressure	That the higher-xG team created better chances; a low block forces shots the defence wants to concede
Possession percentage	Share of total passes completed in a match	Control of the match; a team can hold the ball in harmless areas while the opponent creates the only dangerous attacks
PPDA (passes per defensive action)	How many passes an opponent completes before a defensive intervention	Pressing effectiveness in isolation; a low-block team concedes a high PPDA by design and may be defending perfectly
Distance covered (km per player)	Total running distance from tracking data	Tactical discipline; extreme values often indicate disorganisation, not superior effort, and a well-structured side may cover less ground

Opta and Statsbomb measure the same match and produce distinct xG totals. The gap is not error. It is definition. One provider includes defender positioning in its model; the other weights shot location and body part differently. A header from eight yards out might be 0.15 xG in one system and 0.25 in the other. The difference is the width of a goal-scoring chance. It compounds across a season.

The CIES Football Observatory calculates market value using an algorithm that factors in contract length, age, performance metrics and league economics. Transfermarkt uses a crowd-sourced model. The two estimates for the same player can diverge by 40 per cent. Neither is wrong. They answer distinct questions with distinct inputs. Quoting one as a fact is a category error.

Possession: Completed Passes Or Time On The Ball?

Opta calculates possession as the percentage of total passes completed by a team. A side that plays long balls and presses high may have 42 per cent possession by this measure but control the territory that matters. An alternative method uses time the ball is in a team's control, derived from tracking data. The two methods can produce a 5 to 8 percentage point gap in the same match. When a broadcast graphic flashes a possession figure, it does not tell you which definition it used.

Pressures And PPDA: What Counts As A Defensive Action?

PPDA divides the number of opponent passes by the number of defensive actions: tackles, interceptions and fouls. Providers disagree on whether a challenge that forces a miscontrol counts. One provider's PPDA of 8.2 is another's 6.1. The ranking of pressing teams flips depending on the definition. Before building a scouting report on a rival's pressing intensity, check which definition the data uses. Without that check, you are working from fiction.

Tracking data captures the position of every player and the ball 25 times per second. It can measure off-the-ball runs, defensive shape and passing-lane occupation that event data misses entirely. The limitation is not the data. It is that most public analysis still uses event data. Tracking data is expensive, siloed, and requires a separate analytical skillset.

A defensive structure that concedes zero shots from the central box is invisible in xG until a shot is taken. Tracking data can show that the structure prevented any pass from entering the danger zone. The problem is access. Clubs pay six figures for this data. The public analyst working with free event data sees half the match and draws conclusions as if it is the whole.

The most dangerous error is the one the analyst never sees. A team in a weak domestic league compiles gaudy stats. A striker records 0.9 xG per 90 minutes in the Eredivisie and 0.3 in the Premier League after a transfer. The data was accurate in both leagues. The context was not transferable. The metric measured performance against a specific defensive standard, not an absolute quality.

Model variation interacts with league style. An xG model built on Premier League shot data may overrate shots from range in Serie A, where pressure is applied differently. The model is not broken. It is calibrated to a distinct population. Applying it without adjustment is like using a thermometer calibrated for Celsius and reading the number as Fahrenheit.

Academy productivity figures from the CIES Football Observatory count homegrown players appearing in a top-division match within three years of leaving the academy. A club that loans its graduates to a lower division gets a zero. A club that gives a teenager one substitute appearance gets a one. The definition is consistent and the number is correct. It does not measure what a fan thinks it measures.