📊 statistics

Absolute and Relative Performance Indicators Associated with Match Success Across Six European Professional Football Leagues

This study of 2,058 matches across six European leagues demonstrates that while relative performance indicators (team–opponent differences) are more effective than absolute indicators in predicting match outcomes, the most accurate predictive model is achieved by integrating both absolute and relative metrics.

Original authors: Ângelo Brito

Published 2026-08-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ângelo Brito

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are watching a soccer match. For decades, analysts have been obsessed with counting everything a single team does: How many passes did they make? How much time did they hold the ball? How many shots did they take? It's like judging a chef solely by how many ingredients they chopped, without ever tasting the soup or seeing what the other chef was cooking. This is the world of "absolute" performance—looking at a team's stats in a vacuum. But soccer isn't a solo act; it's a constant, chaotic dance between two teams. If Team A passes the ball 500 times, that's impressive, but what if Team B passed it 600 times? Suddenly, Team A's 500 passes look like a struggle, not a triumph. This paper asks a simple but revolutionary question: To truly understand who wins and why, should we just count what a team did, or should we measure how much better they did it compared to the other team? The answer changes everything about how we look at the beautiful game.

This research, led by Ângelo Brito from the University of Lisbon, decided to settle the debate by looking at a massive mountain of data: every single match from the 2023–2024 season across six of Europe's top professional leagues (including the English Premier League, Spanish LaLiga, and Italian Serie A). That's 2,058 matches and over 4,000 team performances to crunch. The author didn't just look for new stats; instead, he tested three different ways of "reading" the game. First, he looked at Absolute Indicators (what a team did on its own). Second, he looked at Relative Indicators (the difference between what a team did and what their opponent did). Third, he tried Combining both.

The results were as clear as a referee's whistle. When the author tried to predict who would win using only the "Absolute" stats (like total passes or shots), the model was okay, but it missed the mark often, getting the outcome right about 79% of the time. However, when he switched to "Relative" stats—essentially asking, "How many more shots did Team A take than Team B?"—the model became a crystal ball. Suddenly, it could predict the winner with 91% accuracy. The "Relative" approach was a massive leap forward. It showed that knowing a team had 15 shots on target meant very little unless you knew their opponent only had 3. The difference between 15 and 3 is what actually drives victory.

The study found that the most powerful predictors of winning weren't just about having a good offense, but about having a better offense than the other guy. Specifically, having more shots on target than your opponent and being more efficient at turning those shots into goals were the golden keys. When the author combined both types of data (Absolute and Relative), the model got even better, reaching 93% accuracy. But here is the kicker: almost all of that improvement came from the "Relative" part. Adding the "Absolute" numbers only gave a tiny, modest boost.

So, what does this mean for the future of soccer analysis? The paper suggests that we need to stop looking at teams in isolation. A team's performance isn't just a number on a scoreboard; it's a relationship. Just like a race isn't about how fast you run, but how much faster you are than the person next to you, soccer success is defined by the gap between you and your opponent. While counting total passes and shots is still useful for understanding a team's style, if you want to know why a team won or lost, you have to look at the difference. The paper concludes that the best way to model soccer success is to focus on how much a team outperformed their rival, proving that in the game of soccer, context is king, and the only stat that truly matters is the one that compares the two sides.

Technical Summary: Absolute and Relative Performance Indicators Associated with Match Success Across Six European Professional Football Leagues

Problem Statement

Football performance analysis has traditionally focused on identifying absolute match-performance indicators (e.g., total shots, ball possession) associated with competitive success. However, a significant methodological gap exists regarding how competitive performance should be analytically represented. Football is an interactive system where the significance of any action depends on the opposing team's behavior; yet, most research relies on isolated team statistics, implicitly assuming these provide an adequate representation of competitive performance. This study addresses the fundamental question of whether competitive success is more accurately characterized by absolute indicators, relative indicators (team–opponent differences), or a combination of both.

Methodology

Study Design and Sample:
This retrospective observational study utilized a comparative and predictive design analyzing data from the 2023–2024 season across six European professional domestic leagues: the English Premier League, Spanish LaLiga, Italian Serie A, German Bundesliga, French Ligue 1, and Portuguese Primeira Liga. The dataset comprised 2,058 official matches, resulting in 4,116 team-match observations (two observations per match).

Variables:
The study examined three analytical representations:

Absolute Indicators: Values recorded by a team independently (e.g., ball possession %, total shots, shots on target, pass accuracy, offensive efficiency).
Relative Indicators: Calculated as the difference between the focal team's value and the opponent's value within the same match (e.g., Relative Shots on Target = Team Shots on Target − Opponent Shots on Target).
Combined Indicators: An integration of both absolute and relative variables.

Statistical Analysis:

Descriptive and Inferential: One-way ANOVAs were used to compare indicators across wins, draws, and losses, with effect sizes quantified using partial eta squared ( $\eta_p^2$ ).
Predictive Modeling: Multivariable logistic regression models were estimated to predict victory (vs. non-victory). Three models were compared: one using only absolute indicators, one using only relative indicators, and one using a combined set.
Performance Metrics: Models were evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, sensitivity, specificity, and Brier score.
Final Model Selection: A backward stepwise selection procedure was applied to the combined model to identify independent predictors of victory. Standard errors were clustered at the match level to account for the non-independence of the two team observations derived from a single fixture.

Key Results

1. Discriminatory Capacity (Descriptive Analysis):
Relative indicators consistently demonstrated superior discriminatory power compared to absolute indicators.

Effect Sizes: The largest effect sizes were observed for offensive superiority relative to the opponent. Relative offensive efficiency ( $\eta_p^2 = 0.398$ ), relative shots on target ( $\eta_p^2 = 0.261$ ), and relative shot accuracy ( $\eta_p^2 = 0.159$ ) showed substantially larger effects than their absolute counterparts.
Absolute Indicators: While winning teams generally outperformed losing teams in absolute terms (e.g., possession, total shots), the effect sizes were modest (e.g., ball possession $\eta_p^2 = 0.028$ ), indicating that absolute production alone explains a limited proportion of outcome variance.

2. Predictive Performance (Logistic Regression):

Absolute Model: Achieved an AUC of 0.87, accuracy of 78.8%, and a Brier score of 0.143. Sensitivity was notably lower (65.7%) than specificity (86.5%).
Relative Model: Significantly outperformed the absolute model with an AUC of 0.96, accuracy of 90.9%, and a Brier score of 0.071. Sensitivity increased to 86.3%.
Combined Model: Produced the highest overall performance (AUC = 0.97, accuracy = 93.3%, Brier score = 0.058). However, the improvement over the relative-only model was modest (AUC increase of 0.01; accuracy increase of 2.4 percentage points), suggesting that the primary gain in predictive power comes from the relative representation.

3. Independent Predictors of Victory:
In the final multivariable model (combined approach), three indicators remained independently associated with victory:

Relative Shots on Target: Strongest association (Odds Ratio [OR] = 31.59).
Relative Offensive Efficiency: Strong association (OR = 25.88).
Absolute Offensive Efficiency: Retained complementary explanatory value (OR = 3.29).

Key Contributions

Methodological Shift: The study demonstrates that the analytical representation of performance (absolute vs. relative) is a fundamental determinant of explanatory and predictive capacity in football research. It argues that competitive superiority is better captured by team–opponent differences than by isolated production statistics.
Resolution of Inconsistencies: The findings offer a methodological explanation for the historical inconsistencies in literature regarding the importance of variables like ball possession. The study suggests that variability in reported importance may stem from whether performance is analyzed in isolation or relative to the opponent.
Model Optimization: The research establishes that while integrating absolute indicators provides a modest complementary improvement, the vast majority of the explanatory gain in modeling match success is derived from contextualizing team performance through opponent-referenced differences.

Significance and Claims

The paper claims that the primary contribution is not the identification of new performance indicators, but the demonstration that how performance is represented fundamentally influences the information extracted about competitive success.

Relational Nature of Performance: The results support the conceptual view that football is an interactive system where the meaning of a statistic is contingent on the opponent's behavior. Therefore, representing performance through team–opponent differences provides a more informative analytical framework than analyzing isolated team statistics.
Complementary Roles: The study posits that absolute and relative indicators are not mutually exclusive but describe different dimensions. Absolute indicators characterize how a team plays (production), while relative indicators quantify the competitive advantage established. For the specific objective of explaining or modeling match success, the relative framework is superior, though the integration of both yields the most robust models.
Practical Implication: For applied performance analysis, the paper suggests that while absolute statistics remain valuable for describing playing styles, understanding match outcomes requires interpreting these statistics alongside opponent-referenced values to capture the competitive context.

The author maintains a modest stance, noting that these findings are based on a single season's census data and that the strong association of "offensive efficiency" with victory is partly due to its mathematical proximity to the outcome (goals scored). They call for future research to validate these relational principles across multiple seasons, different contexts, and with more granular tracking data.