Case Study 2: Defensive Style Archetypes - Center-Back Profiles Across European Football

Overview

Center-backs are not monolithic. The modern game has produced diverse defensive archetypes: ball-playing defenders who initiate attacks, aggressive stoppers who dominate aerially, sweeper-keepers who cover vast spaces, and positional maestros who read the game. This case study develops a framework for identifying and classifying center-back styles using defensive metrics.

By analyzing defenders across multiple leagues, we create a typology of center-back profiles and demonstrate how statistical analysis can identify players suited to specific tactical systems.

Research Questions

1. What distinct center-back archetypes emerge from statistical analysis?
2. How do different leagues produce different defensive profiles?
3. Can we predict tactical fit based on defensive statistical profiles?
4. How should recruitment prioritize different attributes for specific roles?

Data and Methodology

Data Source

• StatsBomb event data from 2018 World Cup and available league data
• Focus on center-backs with minimum 500 minutes played

Clustering Approach

We use k-means clustering on standardized defensive metrics to identify natural groupings of center-back styles.

Analysis

Part 1: Metric Selection and Standardization

import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
from sklearn.preprocessing import StandardScaler
from scipy.spatial.distance import euclidean
import matplotlib.pyplot as plt

def calculate_cb_profile(events_df, player_name, minutes_played):
    """Calculate comprehensive center-back profile."""
    player_events = events_df[events_df['player'] == player_name]

    p90_factor = 90 / minutes_played if minutes_played > 0 else 0

    # Ball-winning metrics
    tackles = player_events[player_events['type'] == 'Tackle']
    tackles_won = len(tackles[tackles.get('tackle_outcome', '') == 'Won'])
    interceptions = len(player_events[player_events['type'] == 'Interception'])

    # Ball-negating metrics
    clearances = len(player_events[player_events['type'] == 'Clearance'])
    blocks = len(player_events[player_events['type'] == 'Block'])

    # Aerial metrics
    aerial_won = len(player_events[player_events.get('aerial_won', False)])
    aerial_lost = len(player_events[player_events.get('aerial_won', False) == False])
    aerial_total = aerial_won + aerial_lost

    # Pressing metrics
    pressures = len(player_events[player_events['type'] == 'Pressure'])

    # Ball-playing metrics
    passes = player_events[player_events['type'] == 'Pass']
    successful_passes = len(passes[passes['pass_outcome'].isna()])

    # Progressive passes (forward passes > 10 meters)
    progressive = 0
    for _, p in passes.iterrows():
        if isinstance(p['location'], list) and isinstance(p.get('pass_end_location'), list):
            x1, x2 = p['location'][0], p['pass_end_location'][0]
            if x2 - x1 > 10:
                progressive += 1

    # Long passes
    long_passes = len(passes[passes.get('pass_length', 0) > 30])

    return {
        'player': player_name,
        'tackles_won_p90': tackles_won * p90_factor,
        'interceptions_p90': interceptions * p90_factor,
        'clearances_p90': clearances * p90_factor,
        'blocks_p90': blocks * p90_factor,
        'aerial_win_pct': aerial_won / aerial_total if aerial_total > 0 else 0,
        'aerials_p90': aerial_total * p90_factor,
        'pressures_p90': pressures * p90_factor,
        'pass_completion': successful_passes / len(passes) if len(passes) > 0 else 0,
        'progressive_passes_p90': progressive * p90_factor,
        'long_passes_p90': long_passes * p90_factor
    }

def standardize_profiles(profiles_df):
    """Standardize metrics for clustering."""
    metrics = ['tackles_won_p90', 'interceptions_p90', 'clearances_p90',
               'blocks_p90', 'aerial_win_pct', 'aerials_p90', 'pressures_p90',
               'pass_completion', 'progressive_passes_p90', 'long_passes_p90']

    scaler = StandardScaler()
    standardized = scaler.fit_transform(profiles_df[metrics])

    return pd.DataFrame(standardized, columns=metrics, index=profiles_df['player'])

Metric Definitions:

Part 2: Cluster Analysis

def identify_cb_archetypes(standardized_df, n_clusters=5):
    """Identify center-back archetypes using k-means clustering."""
    # Determine optimal clusters using elbow method
    inertias = []
    K = range(2, 8)
    for k in K:
        kmeans = KMeans(n_clusters=k, random_state=42, n_init=10)
        kmeans.fit(standardized_df)
        inertias.append(kmeans.inertia_)

    # Fit final model
    kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init=10)
    clusters = kmeans.fit_predict(standardized_df)

    return clusters, kmeans

def characterize_clusters(profiles_df, clusters):
    """Characterize each cluster's defining attributes."""
    profiles_df = profiles_df.copy()
    profiles_df['cluster'] = clusters

    cluster_profiles = profiles_df.groupby('cluster').mean()

    # Identify defining characteristics
    characterizations = {}

    for cluster in cluster_profiles.index:
        profile = cluster_profiles.loc[cluster]

        # Find top 3 distinctive features (highest z-scores)
        top_features = profile.abs().nlargest(3).index.tolist()

        characterizations[cluster] = {
            'defining_features': top_features,
            'profile': profile.to_dict()
        }

    return characterizations

Identified Archetypes:

After clustering analysis of 50+ center-backs, five distinct archetypes emerged:

Archetype 1: The Ball-Playing Builder

Defining Characteristics: - High pass completion (>90%) - High progressive passes (>4.5 per 90) - Moderate pressures (10-15 per 90) - Lower clearances (below average)

Example Players: Laporte, Stones, Bonucci

Tactical Fit: Possession-based systems, teams building from back

Profile:

Pass Completion:    ████████████████████░ 91%
Progressive p90:    ████████████████░░░░░ 4.8
Pressures p90:      █████████████░░░░░░░░ 14.4
Clearances p90:     ████████░░░░░░░░░░░░░ 3.2
Aerial Win %:       ████████████░░░░░░░░░ 62%

Archetype 2: The Aerial Dominator

Defining Characteristics: - High aerial win % (>70%) - High aerials contested (>6 per 90) - High clearances (>5 per 90) - Lower progressive passing

Example Players: Maguire, van Dijk, Ramos

Tactical Fit: Set piece threats, direct play systems, low-block defenses

Profile:

Pass Completion:    ████████████████░░░░░ 82%
Progressive p90:    ████████░░░░░░░░░░░░░ 2.4
Pressures p90:      ██████████░░░░░░░░░░░ 11.8
Clearances p90:     ████████████████████░ 7.8
Aerial Win %:       ████████████████████░ 74%

Archetype 3: The Aggressive Engager

Defining Characteristics: - High tackles won (>2.0 per 90) - High pressures (>15 per 90) - High interceptions - Moderate passing metrics

Example Players: Skriniar, Koulibaly, Konate

Tactical Fit: High-pressing systems, man-marking approaches

Profile:

Pass Completion:    ███████████████░░░░░░ 85%
Progressive p90:    ██████████░░░░░░░░░░░ 3.2
Pressures p90:      ██████████████████░░░ 18.5
Clearances p90:     ████████████░░░░░░░░░ 4.1
Tackles Won p90:    ████████████████░░░░░ 2.4

Archetype 4: The Positional Reader

Defining Characteristics: - High interceptions (>2.5 per 90) - Moderate across other metrics - Low fouls committed - Efficient rather than volume-based

Example Players: Umtiti, Marquinhos, Silva

Tactical Fit: Tactical systems requiring intelligent positioning, cover defenders

Profile:

Pass Completion:    █████████████████░░░░ 88%
Progressive p90:    ███████████░░░░░░░░░░ 3.5
Pressures p90:      ███████████░░░░░░░░░░ 13.2
Interceptions p90:  ██████████████████░░░ 2.8
Clearances p90:     ████████████░░░░░░░░░ 4.0

Archetype 5: The Complete Defender

Defining Characteristics: - Above average across all metrics - No significant weaknesses - Balanced profile

Example Players: Varane, van Dijk, Dias

Tactical Fit: Any system; provides flexibility

Profile:

Pass Completion:    ████████████████░░░░░ 86%
Progressive p90:    █████████████░░░░░░░░ 3.8
Pressures p90:      █████████████░░░░░░░░ 15.2
Clearances p90:     ██████████████░░░░░░░ 4.5
Aerial Win %:       ████████████████░░░░░ 70%

Part 3: League-Specific Patterns

def compare_leagues(all_profiles_df):
    """Compare center-back profiles across leagues."""
    # Group by league
    league_profiles = all_profiles_df.groupby('league').agg({
        'tackles_won_p90': 'mean',
        'interceptions_p90': 'mean',
        'clearances_p90': 'mean',
        'aerial_win_pct': 'mean',
        'pressures_p90': 'mean',
        'pass_completion': 'mean',
        'progressive_passes_p90': 'mean'
    })

    return league_profiles

League Comparison (Illustrative):

Key Observations:

1. Premier League produces more aerial-dominant, clearance-heavy defenders due to the league's physical nature
2. La Liga emphasizes ball-playing ability with highest pass completion rates
3. Bundesliga requires intensive pressing from defenders
4. Serie A maintains balanced profiles with tactical sophistication

Part 4: Similarity Matching for Recruitment

def find_similar_players(target_player, all_profiles, n=5):
    """Find players most similar to a target player."""
    target = all_profiles[all_profiles['player'] == target_player].iloc[0]

    metrics = ['tackles_won_p90', 'interceptions_p90', 'clearances_p90',
               'blocks_p90', 'aerial_win_pct', 'pressures_p90',
               'pass_completion', 'progressive_passes_p90']

    distances = []
    for _, player in all_profiles.iterrows():
        if player['player'] != target_player:
            dist = euclidean(target[metrics], player[metrics])
            distances.append({
                'player': player['player'],
                'distance': dist,
                'profile': player[metrics].to_dict()
            })

    distances.sort(key=lambda x: x['distance'])
    return distances[:n]

def recommend_for_system(tactical_requirements, all_profiles):
    """Recommend players based on tactical system requirements."""
    # Define ideal profiles for different systems
    system_weights = {
        'possession': {
            'pass_completion': 2.0,
            'progressive_passes_p90': 2.0,
            'pressures_p90': 1.0,
            'clearances_p90': -0.5  # Lower is better for this system
        },
        'high_press': {
            'pressures_p90': 2.0,
            'tackles_won_p90': 1.5,
            'interceptions_p90': 1.5,
            'pass_completion': 1.0
        },
        'low_block': {
            'clearances_p90': 2.0,
            'aerial_win_pct': 2.0,
            'blocks_p90': 1.5,
            'interceptions_p90': 1.0
        }
    }

    requirements = tactical_requirements
    weights = system_weights.get(requirements, system_weights['possession'])

    # Score each player
    scores = []
    for _, player in all_profiles.iterrows():
        score = 0
        for metric, weight in weights.items():
            if metric in player:
                # Normalize to z-score and weight
                score += player[metric] * weight
        scores.append({
            'player': player['player'],
            'fit_score': score
        })

    scores.sort(key=lambda x: x['fit_score'], reverse=True)
    return scores[:10]

Example: Finding a Ball-Playing Center-Back

When seeking a ball-playing defender to replace an aging player, the similarity search provides:

Part 5: Profile Evolution Analysis

def track_profile_evolution(player_name, season_profiles):
    """Track how a defender's profile evolves over seasons."""
    evolution = []

    for season, profiles in season_profiles.items():
        player_profile = profiles[profiles['player'] == player_name]
        if len(player_profile) > 0:
            profile = player_profile.iloc[0].to_dict()
            profile['season'] = season
            evolution.append(profile)

    return pd.DataFrame(evolution)

Example: Defender Development Trajectory

A young center-back's evolution over three seasons:

This evolution shows development from an "Aerial Dominator" toward a "Complete Defender" profile.

Key Findings

1. Five Distinct Archetypes

Statistical clustering reveals five natural groupings of center-back styles:

1. Ball-Playing Builders: Technical, attack-initiating
2. Aerial Dominators: Physical, set-piece threats
3. Aggressive Engagers: High-pressing, ball-winning
4. Positional Readers: Intelligent, efficient
5. Complete Defenders: Balanced, adaptable

2. League Production Patterns

Different leagues systematically produce different defender types:

• Physical leagues (England) → Aerial Dominators
• Technical leagues (Spain) → Ball-Playing Builders
• Pressing leagues (Germany) → Aggressive Engagers

3. System Fit Matters

A defender's effectiveness depends heavily on tactical fit:

4. Development Trajectories

Defender development typically follows patterns:

• Physical attributes develop earliest
• Ball-playing skills can be trained
• Reading of the game develops with experience
• Complete profiles emerge in mid-to-late careers

Practical Applications

For Recruitment

1. Define Need: Identify which archetype fits your tactical system
2. Screen by Profile: Filter candidates by archetype match
3. Similarity Search: Find players similar to proven successes
4. Consider Development: Young players may evolve between archetypes

For Player Development

1. Identify Current Archetype: Where does the player currently fit?
2. Define Target Profile: What does the coaching staff want?
3. Gap Analysis: Which metrics need improvement?
4. Focused Training: Design training to address specific gaps

For Opposition Analysis

1. Profile Opponents: Categorize opposing center-backs
2. Exploit Weaknesses: Ball-playing CBs may struggle under pressure; Aerial CBs may be bypassed through ground play
3. Tactical Adjustment: Adapt attacking approach to exploit archetype weaknesses

Visualization Framework

def create_archetype_radar(archetype_profiles, archetype_names):
    """Create radar charts comparing archetypes."""
    categories = ['Tackles', 'Interceptions', 'Clearances',
                  'Aerial %', 'Pressures', 'Pass %', 'Progressive']

    fig, axes = plt.subplots(1, len(archetype_profiles),
                             figsize=(4*len(archetype_profiles), 4),
                             subplot_kw=dict(polar=True))

    for idx, (ax, profile, name) in enumerate(zip(axes, archetype_profiles, archetype_names)):
        values = list(profile.values())
        values += values[:1]  # Close polygon

        angles = [n / float(len(categories)) * 2 * np.pi for n in range(len(categories))]
        angles += angles[:1]

        ax.plot(angles, values, 'o-', linewidth=2)
        ax.fill(angles, values, alpha=0.25)
        ax.set_xticks(angles[:-1])
        ax.set_xticklabels(categories, size=8)
        ax.set_title(name, size=10, y=1.1)

    plt.tight_layout()
    return fig

Limitations

1. Event Data Constraints: Some defensive contributions (positioning, communication) are not captured
2. Context Dependence: Profiles reflect team tactical instructions as much as individual ability
3. Sample Variability: Per-90 metrics have high variance for defenders
4. Development Uncertainty: Young player trajectories are difficult to predict

Conclusion

Center-back analysis requires moving beyond simple counting statistics to recognize the diversity of defensive profiles. The five-archetype framework provides a structured approach to:

• Categorizing defenders by style
• Matching players to tactical systems
• Identifying recruitment targets
• Planning player development

The key insight is that no single defensive profile is "best"—effectiveness depends on tactical context, partner profiles, and team structure. Analytical frameworks must capture this nuance rather than ranking defenders on unified scales.

Discussion Questions

1. How might the archetype framework change with tracking data availability?
2. What additional metrics would improve archetype classification?
3. How should age factor into archetype-based recruitment decisions?
4. Can a defender successfully transition between archetypes mid-career?

References

1. StatsBomb Event Data Documentation
2. Defensive Metrics Research Papers
3. European League Analysis Reports