Performance Analysis Metrics — What the Numbers Actually Tell Us About Football

Introduction

Football performance analysis has undergone a revolution in the past decade. Where coaches once relied on subjective post-match assessment, clubs now receive thousands of data points per match — GPS physical outputs, event-log technical data, tracking data for every player’s position every 0.1 seconds, and a growing library of derived metrics attempting to quantify tactical and technical quality. The challenge is not generating data — it is interpreting it correctly. Understanding what performance metrics actually measure, what they predict, and where their limitations lie is essential for anyone working with football data.

The Science

Physical GPS metrics: GPS systems measuring positional data at 10–25 Hz (samples per second) generate per-match outputs including:

• Total distance (TD): Total metres covered across all movement speeds. Range 10,000–12,000m in elite matches
• High-speed running (HSR): Distance covered >19–21 km/h (threshold varies by system). Range 800–1,400m per match; highly position-dependent
• Sprint distance: Distance covered at >25 km/h. Range 200–500m per match
• Accelerations/decelerations: Count of changes exceeding a threshold (typically ≥2 m/s²); emerging evidence for metabolic cost significance
• Player Load (PL): Accelerometer-derived composite measure of mechanical stress in all planes

Expected goals (xG): Probabilistic model of goal probability from a given shot, based on historical data on shot conversion rates for a given location, angle, shot type, and preceding play context. xG is a better predictor of future match outcomes than actual goals scored — teams that outperform their xG (score more goals than shot quality predicts) tend to regress toward xG over subsequent matches. xG identifies quality of chance creation beyond the noise of shot conversion variation.

Pressing metrics (PPDA): Passes Allowed Per Defensive Action — the ratio of opposition passes to pressing team’s defensive actions in a defined zone. Low PPDA = high-intensity pressing. Originally developed by StatsBomb; now widely used in tactical analysis of pressing intensity.

Spatial and possession control: Pitch control models estimate the probability that each player controls each pitch cell at any moment — providing a continuous spatial dominance measure. VAEP (Valuing Actions by Estimating Probabilities) assigns value to every on-ball action based on its effect on goal probability.

Limitations of physical metrics: GPS distance and speed metrics are systematically influenced by GPS signal quality, weather, stadium coverage, and satellite configuration — different GPS systems produce non-comparable absolute values. Within-player comparisons over time (using the same system) are valid; cross-system comparisons are not.

What Research Says

The relationship between high-intensity running distance and match outcome in elite football is complex and context-dependent: some studies link greater HSR to winning, while others (including Collet 2013) find that higher-ranked teams often cover less total distance because they control possession — so physical output is only a partial, context-dependent predictor of results.

Collet (2013) examined possession metrics in Journal of Sports Sciences, finding that possession percentage alone poorly predicted match outcomes (correlation ~0.2 with goals scored), while quality of possession (passes leading to dangerous zones, xG from possession sequences) was a substantially stronger predictor — establishing that quality metrics outperform volume metrics in football analysis.

Reviews of the validity and reliability of GPS-based physical performance metrics have established that total distance and HSR measures have acceptable reliability (CV <5%) when using the same GPS system, while acceleration counts have higher variability and require larger sample sizes for reliable interpretation.

Did You Know? The data revolution’s most significant practical impact on tactical football analysis came through Expected Threat (xT) and Expected Possession Value (EPV) models — which assign a value to moving the ball to different pitch zones based on historical probability of eventually scoring. These models allow analysts to evaluate passes, dribbles, and carries in terms of their value creation — not just their success/failure binary. A short pass to a less dangerous zone can increase xT; a long pass into space can increase it more. These models have fundamentally changed how clubs evaluate player contributions that don’t appear in traditional statistics.

Applied to Football

Implementing effective performance analysis:

1. Benchmark physical metrics within-player, not across players. Each player’s HSR and sprint output is compared to their own seasonal baseline, not to population norms. A wide forward’s 1,200m HSR and a centre-back’s 700m may both represent 100% of individual capacity.
2. Use xG to separate performance from luck. A team that loses 2–1 but has xG 2.1 vs 0.8 performed well despite the scoreline. Season-level xG tracking separates quality teams from fortunate ones.
3. Contextualise PPDA with territorial context. PPDA alone doesn’t distinguish a high-line intensive press from a conservative deep block. Combine with average defensive line height for tactical context.
4. Track acceleration load, not just HSR. Accelerations and decelerations represent metabolic and mechanical load that GPS distance misses. High-sprint-distance players may have lower acceleration counts — different load profiles with different injury and fatigue implications.
5. Multi-match sample before drawing conclusions. Physical performance varies substantially match-to-match (CV ~15–20% for individual HSR). Single-match conclusions are unreliable — use 5+ match rolling averages.

Key Takeaways

• GPS metrics must be benchmarked within-player and within-system — cross-player and cross-system comparisons are unreliable
• xG predicts future match outcomes better than actual goals scored — quality of chance creation outperforms finishing variance
• PPDA quantifies pressing intensity; contextualise with territorial data for tactical interpretation
• Acceleration/deceleration counts represent mechanical and metabolic load invisible to speed-based GPS metrics
• Single-match physical metrics have 15–20% individual variability — require multi-match averaging for reliable interpretation

References

• Collet, C. (2013). The possession game? A comparative analysis of ball retention and team success in European and international football, 2007–2010. Journal of Sports Sciences, 31(2), 123–136.

Written by Hüseyin Akbulut, MSc Sport Scientist · Founder of Sporeus

Hüseyin Akbulut is the founder of Sporeus and author of THRESHOLD (EŞİK), a 540-page Turkish-language book on endurance science.

• 540pp THRESHOLD Book
• MSc Sport Sciences
• Marmara University

Full profile →