If running in Google Colab, uncomment the cell below. Skip if running locally.

# !rm -rf temp_repo src
# !git clone https://github.com/ringhilterra/analytics_cup_research.git temp_repo
# !mv temp_repo/src .
# !rm -rf temp_repo
# !pip install mplsoccer -q

The Active Support Index (ASI)

Quantifying Off-Ball Movement When It Matters Most

Author: Ryan Inghilterra

Core Question: When a player is pressured, how actively are teammates moving to provide passing options?

Metrics Framework:

Level Metric Formula Interpretation
Per Event Active Support Ratio \(\frac{\text{Active Supporters}}{\text{Nearby Teammates}}\) Support quality in a single pressure moment
Per Player Player ASI \(\frac{\text{Active Support Count}}{\text{Support Opportunities}}\) How often a player moves to support under pressure
Per Team Team ASI \(1 - \text{Static Rate}\) Team’s off-ball movement culture

Definitions: - Active Supporter: Teammate within 35m AND moving >2 m/s - Static Rate: Proportion of events with zero active supporters

Higher values = better off-ball support.

Validation highlights: - ASI correlates with season-level physical output (r = 0.74, p < 0.001) - ASI aligns with positional demands (midfielders 59% vs defenders 45%, p < 0.001) - Team ASI differentiates playing styles (98.3% Perth Glory to 91.5% Macarthur FC) - 58% of players show declining support in H2 (fatigue signal)

# Core imports
import pandas as pd
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')

# Display settings
pd.set_option('display.max_columns', 20)
pd.set_option('display.width', 200)
plt.style.use('dark_background')

# src imports - core
from src import ASIDataLoader, ASICalculator
from src import fetch_match_data, process_all_matches, get_top_players_all_matches, get_team_stats_all_matches, plot_multi_match_team_comparison
from src.asi_data_loader import ASIDataLoader, print_match_summary
from src.asi_core import ASICalculator, ASIConfig
from src.asi_visualizations import ASIVisualizer
from src.nb_helper import get_detail_results_summary, team_stat_asi_summary, get_event_by_id, plot_pressure_from_event_id

# src imports - validation & analysis
from src import calculate_position_stats, plot_position_validation, test_position_significance, print_significance_result
from src import calculate_time_based_asi, get_time_bin_stats, plot_time_trend, compare_halves, print_half_comparison
from src import calculate_player_fatigue, plot_fatigue_comparison, print_fatigue_summary
from src import load_physical_aggregates, merge_asi_with_physical, calculate_physical_correlation, plot_asi_physical_correlation, print_physical_validation_summary

Single Match Analysis

MATCH_ID = 1886347 # Auckland (Home) vs Newcastle (Away) - final score: (2, 0)
# MATCH_ID = 1899585 ## <- can easily change match_id

Data Loading

# First fetch/download data for single match - tracking and event data from SkillCorner GitHub
fetch_match_data(MATCH_ID, verbose=True, skip_if_exists=True)  # Downloads and processes match data to ./data/1886347/
Match 1886347 data already exists, skipping fetch.
PosixPath('data/1886347')
# Load match
loader = ASIDataLoader(data_dir="./data")
match_data = loader.load_match(MATCH_ID)

# Print summary
print_match_summary(match_data)
============================================================
MATCH 1886347
Auckland FC 2 - 0 Newcastle United Jets FC
------------------------------------------------------------
Tracking rows:    956,076
Event rows:       5,079
Pressure events:  773
Pitch:            104m x 68m
Validated:        Yes
============================================================
# show processed tracking data with velocity (speed)
display_cols = ['frame', 'team_id', 'short_name','number', 'x', 'y', 'vx', 'vy', 'speed', 'match_time_min', 'ball_x', 'ball_y', 'player_role.acronym']
match_data.tracking_df.head(3)[display_cols]
frame team_id short_name number x y vx vy speed match_time_min ball_x ball_y player_role.acronym
0 10 4177 L. Verstraete 6 11.70 6.73 -1.242857 -0.407143 1.307845 0.0 0.32 0.38 DM
1 10 4177 F. Gallegos 28 10.16 -2.12 -0.292857 -1.042857 1.083197 0.0 0.32 0.38 AM
2 10 4177 N. Pijnaker 4 16.78 -3.67 0.257143 -0.289286 0.387051 0.0 0.32 0.38 LCB 
match_data.events_df.iloc[:, :8].tail(3) # peak at dynamic_events
event_id frame_start frame_end event_type event_subtype player_id team_id player_name
5076 8_998 58907 58932 player_possession NaN 285188 4177 A. Paulsen
5077 7_2542 58907 58932 passing_option NaN 43829 4177 N. Moreno
5078 7_2543 58907 58932 passing_option NaN 163972 4177 M. Mata

Calculating ASI Per Pressure Event

Configurable Thresholds: | Parameter | Value | Rationale | |———–|——-|———–| | Proximity | 35m | Maximum realistic passing range under pressure | | Velocity | 2 m/s | Threshold separating walking (~1.4 m/s) from jogging/running |

# can adjust these values
PROX_THRESH = 35
VELOCITY_THRESH = 2

config = ASIConfig(proximity_threshold_m=PROX_THRESH, velocity_threshold_ms=VELOCITY_THRESH)
print(config.proximity_threshold_m)
print(config.velocity_threshold_ms)
35
2
# Calculate ASI metrics for each pressure event
calculator = ASICalculator(match_data, config=config)
results_df = calculator.process_all_pressure_events()
print(results_df.shape)
(773, 17)

print("\nPressure Event Results (first 8)")
display_cols = ['frame_start', 'period', 'pressed_player_name', 'num_teammates_nearby', 'num_active_supporters', 'active_support_ratio']
results_df[display_cols].head(8)

Pressure Event Results (first 8)
frame_start period pressed_player_name num_teammates_nearby num_active_supporters active_support_ratio
0 56 1 A. Šušnjar 8 4 0.5000
1 74 1 M. Natta 8 6 0.7500
2 235 1 F. De Vries 4 2 0.5000
3 253 1 D. Ingham 9 8 0.8889
4 314 1 L. Gillion 5 5 1.0000
5 345 1 F. Gallegos 6 3 0.5000
6 499 1 A. Šušnjar 7 0 0.0000
7 549 1 F. Gallegos 9 6 0.6667

Example (Row 1): A. Šušnjar under pressure - 8 teammates within 35m → 4 moving >2 m/s → Active Support Ratio = 0.50 - Half his nearby teammates were actively creating options; half were static.

# Get detailed result for a single event
detailed_results = calculator.get_detailed_results()
# Show one event in detail
event = detailed_results[1]
get_detail_results_summary(event)

Detailed Analysis - Single Pressure Event
============================================================
Event ID: 9_1
Frame: 74
Type: recovery_press

Pressed Player: M. Natta
Position: (-16.9, 20.6)

Support Metrics:
  Teammates Nearby (<35m): 8
  Active Supporters (>2m/s): 6
  Active Support Ratio: 0.75

Teammate Details:
  C. Timmins      | Dist:  14.8m | Speed: 3.8 m/s | ACTIVE
  D. Ingham       | Dist:  56.4m | Speed: 1.6 m/s | FAR
  R. Scott        | Dist:  30.2m | Speed: 1.6 m/s | STATIC
  A. Šušnjar      | Dist:  15.9m | Speed: 3.1 m/s | ACTIVE
  P. Cancar       | Dist:  29.2m | Speed: 1.9 m/s | STATIC

Pressure Moment Visualization

Visualize pressure events with active/static support color coding. Use plot_pressure_from_event_id(event_id, ...) to generate a visualization for any pressure event.

High Active Support Example

# Initialize visualizer
visualizer = ASIVisualizer(match_data)
plot_pressure_from_event_id("9_3", detailed_results, match_data, loader)  # frame 253, Player = D.Ingham

D. Ingham (#14, Newcastle) pressed while facing his own goal. Active Support Ratio = 0.89 — 8 of 9 nearby teammates actively moving. Only the goalkeeper was static (already open). Note #4 sprinting back at 3.7 m/s to offer a passing lane.

Low Active Support Example

plot_pressure_from_event_id("9_790", detailed_results, match_data, loader)  # frame 47507, Player = L.Gillion

L. Gillion (#14, Auckland) pressed near the 18-yard box. Active Support Ratio = 0.11 — only 1 of 9 nearby teammates moving (#17 at 2.2 m/s). Most teammates were static “ball-watching” rather than creating options. This is exactly the scenario coaches want to identify and correct.

Player ASI Analysis

Calculate and rank players by their Active Support Index.

# Calculate player ASI scores
player_scores = calculator.calculate_player_asi_scores(detailed_results)
print(f"Player ASI Scores: Total players analyzed: {len(player_scores)}")
# Display top 10 with minimum 20 opportunities
top_players = player_scores[player_scores['opportunities'] >= 20].head(10)
print(f"\nTop 10 Players (min 20 opportunities):")

exclude_cols = ['player_id', 'player_number', 'match_name', 'matches_count']
top_players.drop(columns=exclude_cols).head(10)
Player ASI Scores: Total players analyzed: 29

Top 10 Players (min 20 opportunities):
player_name team_name player_role_acronym active_support_count opportunities asi_score
0 C. Timmins Newcastle United Jets FC LDM 214 278 0.7698
1 D. Wilmering Newcastle United Jets FC AM 39 54 0.7222
2 F. Gallegos Auckland FC AM 231 330 0.7000
3 M. Scarcella Newcastle United Jets FC AM 27 39 0.6923
4 L. Gillion Auckland FC LW 181 269 0.6729
5 L. Verstraete Auckland FC DM 227 345 0.6580
6 J. Vidic Newcastle United Jets FC LDM 25 38 0.6579
7 N. Moreno Auckland FC RW 46 70 0.6571
8 L. Bayliss Newcastle United Jets FC AM 136 210 0.6476
9 L. Rogerson Auckland FC RW 120 193 0.6218

Player ASI measures how often a player actively moves to support teammates under pressure — the proportion of support opportunities where they were moving >2 m/s.

# Player leaderboard visualization
fig = visualizer.plot_player_leaderboard(player_scores, top_n=None, min_opportunities=20, show=True)

Midfielders cluster at the top — their role demands constant movement. Defenders and goalkeepers rank lower, which is expected given their positional responsibilities. Standout: C. Timmins (LDM) leads with 77% ASI, providing active support in 214 of 278 opportunities — the most reliable off-ball mover in this match.

Team ASI Comparison

Calculate and compare Active Support Index between teams for the match.

# Calculate team-level ASI
team_stats = calculator.calculate_team_asi_scores(results_df)
# provide summary
team_stat_asi_summary(team_stats)
Team ASI Comparison
============================================================

Auckland FC:
  Total pressure events (when pressed): 402
  Team ASI (1 - static rate):           96.0%
  Static Rate (0 active supporters):   4.0%
  Avg Active Supporters:                3.92
  Avg Teammates Nearby:                 7.49

Newcastle United Jets FC:
  Total pressure events (when pressed): 371
  Team ASI (1 - static rate):           94.6%
  Static Rate (0 active supporters):   5.4%
  Avg Active Supporters:                4.13
  Avg Teammates Nearby:                 7.73
# Team comparison visualization
fig = visualizer.plot_team_comparison(team_stats, show=True)

Metric What it Measures Interpretation
Team ASI 1 - static_rate — Proportion of pressure events where at least one teammate was actively moving Higher = Better. A team with 95% ASI means only 5% of pressure moments had zero active support.
Avg Active Supporters Mean number of teammates moving >2 m/s within 35m when ball carrier is pressed Higher = Better. More teammates in motion = more passing options and defensive support.
Static Rate Proportion of pressure events where no nearby teammate was moving >2 m/s Lower = Better. High static rate indicates teammates “ball-watching” instead of creating options.

Quick Read: Higher Team ASI + Lower Static Rate = better off-ball support culture. Avg Active Supporters shows intensity — teams averaging 3+ are creating strong movement.

Run on all 10 matches

# Process all 10 matches (will take a couple of minutes)
results_all = process_all_matches(verbose=False)
# Top 10 players across all matches
top_players_all = get_top_players_all_matches(results_all['all_player_scores'], min_opportunities=50, top_n=300)
exclude_cols = ['player_id', 'player_number', 'match_name']
top_players_all.drop(columns=exclude_cols).head(10)
player_name team_name player_role_acronym active_support_count opportunities matches_count asi_score
0 N. Pennington Perth Glory Football Club LM 211 261 1 0.8084
1 H. Steele Central Coast Mariners Football Club LM 184 228 1 0.8070
2 C. Timmins Newcastle United Jets FC LDM 214 278 1 0.7698
3 L. Verstraete Auckland FC DM 688 936 4 0.7350
4 D. Wilmering Newcastle United Jets FC AM 39 54 1 0.7222
5 P. Makrillos Macarthur FC CF 78 108 1 0.7222
6 J. Lauton Western United RM 120 167 2 0.7186
7 Z. Schreiber Melbourne City FC DM 178 249 1 0.7149
8 T. Gomulka Perth Glory Football Club RM 183 256 1 0.7148
9 A. Thurgate Western United LM 372 524 2 0.7099 
fig = visualizer.plot_player_leaderboard(top_players_all, top_n=25, min_opportunities=20)

Wide midfielders (LM, RM) and central midfielders (DM, AM) dominate the top 25 — positions requiring constant off-ball movement. Only 2 players exceed 80% ASI: N. Pennington (80.8%) and H. Steele (80.7%), both left midfielders, marking them as elite off-ball supporters. The 75th percentile threshold (~70%) separates good from exceptional active supporters.

# Team comparison across all matches
team_summary = get_team_stats_all_matches(results_all['all_team_stats'])
team_summary
team_name total_pressure_events total_static_events avg_active_supporters avg_teammates_nearby matches_played overall_static_rate overall_team_asi
0 Perth Glory Football Club 287 5 4.30 7.66 1 0.0174 0.9826
1 Auckland FC 1368 37 4.18 7.71 4 0.0270 0.9730
2 Melbourne City FC 828 36 3.62 7.22 2 0.0435 0.9565
3 Western United 647 29 4.40 7.60 2 0.0448 0.9552
4 Brisbane Roar FC 453 21 3.56 7.53 1 0.0464 0.9536
5 Wellington Phoenix FC 598 28 3.72 7.77 2 0.0468 0.9532
6 Adelaide United Football Club 392 19 3.57 7.49 1 0.0485 0.9515
7 Newcastle United Jets FC 371 20 4.13 7.73 1 0.0539 0.9461
8 Central Coast Mariners Football Club 266 15 3.63 7.39 1 0.0564 0.9436
9 Melbourne Victory Football Club 674 39 4.00 7.91 2 0.0579 0.9421
10 Sydney Football Club 862 50 3.68 7.74 2 0.0580 0.9420
11 Macarthur FC 317 27 3.41 7.39 1 0.0852 0.9148 
fig = plot_multi_match_team_comparison(team_summary)

Perth Glory (98.3%) and Auckland (97.3%) lead in Team ASI, demonstrating strong off-ball support cultures. Macarthur FC (91.5%) shows the most room for improvement. The 7-point spread suggests ASI can differentiate team playing styles.

Validation: ASI by Position

To validate that ASI captures meaningful off-ball behavior, we examine whether positions with higher movement demands show correspondingly higher ASI scores.

# Calculate ASI stats by position category
category_stats = calculate_position_stats(top_players_all)
category_stats
mean_asi std_asi num_players total_opportunities
position_category
Defensive/Central Mid 0.627 0.083 15 5683
Attacking Mid/Winger 0.586 0.099 62 15705
Forward 0.515 0.101 35 9023
Defender 0.452 0.079 61 20367
Goalkeeper 0.113 0.059 12 2334 
# Visualize ASI by position category
fig = plot_position_validation(category_stats)

# Statistical significance test
result = test_position_significance(top_players_all)
print_significance_result(result)
Midfielders vs Defenders:
  Midfielders mean ASI: 59.4% (n=77)
  Defenders mean ASI: 45.2% (n=61)
  Mann-Whitney U: 4092, p = 4.00e-14 Significant (p < 0.001)

Validation Result: Position categories rank as expected — attacking midfielders and wingers show significantly higher ASI than defenders (p < 0.001). This confirms ASI captures positional movement demands rather than random variation.

Validation: Season Physical Output

Does ASI from tracking data reflect real physical effort? We validate by comparing player ASI scores (from 10 tracking matches) against season-level physical aggregates (175 matches, A-League 2024/25).

# Merge ASI scores with physical aggregates
merged_physical = merge_asi_with_physical(top_players_all, min_opportunities=50)
print(f"Players with ASI and physical data: {len(merged_physical)}")

# Calculate and display correlation
corr_stats = calculate_physical_correlation(merged_physical)
print_physical_validation_summary(corr_stats)
Players with ASI and physical data: 167
Physical Aggregates Validation
==================================================
Players analyzed: 167
Pearson r: 0.739
p-value: 4.47e-30

Quartile Comparison (M/min during possession):
  Low ASI (Q1):  128.8 m/min
  High ASI (Q4): 153.2 m/min
  Difference:    +24.4 m/min (+19%)
# Visualize ASI vs physical output correlation
fig = plot_asi_physical_correlation(merged_physical)

Result: ASI correlates strongly with meters per minute during possession (r = 0.74, p < 0.001). Players in the top ASI quartile cover 24 m/min more than the bottom quartile — a 19% increase in work rate. This external validation confirms ASI captures genuine physical effort: players who actively support teammates under pressure are the same players who cover the most ground when their team has possession.

Time-Based ASI Analysis

Does active support decline as the match progresses? We analyze ASI trends over 15-minute intervals to identify potential fatigue patterns.

# Calculate time-based ASI across all 10 matches
all_results_time_df = calculate_time_based_asi(loader, config)
print(f"Total pressure events with time data: {len(all_results_time_df)}")
Total pressure events with time data: 7063
# Calculate ASI by time bin
time_asi = get_time_bin_stats(all_results_time_df)
time_asi
avg_support_ratio avg_active_supporters num_events
time_bin
0-15 0.552 4.055 1283
15-30 0.534 3.784 990
30-45 0.527 3.770 1128
45-60 0.542 3.956 1302
60-75 0.548 3.984 1013
75-90 0.502 3.683 992
90+ 0.513 3.730 355 
# Visualize ASI trend over match time
fig = plot_time_trend(time_asi)

# Compare First Half vs Second Half
half_result = compare_halves(all_results_time_df)
print_half_comparison(half_result)
half_result['half_stats']

First Half vs Second Half ASI:
  H1 mean: 53.7%, H2 mean: 53.1%
  Mann-Whitney U: 6298558, p = 0.459 (not significant)
avg_ratio std_ratio avg_active num_events
half
First Half 0.537 0.303 3.877 3565
Second Half 0.531 0.303 3.872 3498

Observation: Active support ratio shows a slight downward trend within each half, with the lowest values appearing in the final 15 minutes (50.2%). However, the overall first half vs second half difference is not statistically significant (p = 0.46). This suggests fatigue-related decline in off-ball support may exist but requires larger sample sizes to confirm. The pattern warrants further investigation with more match data.

Player Fatigue Analysis

Which players maintain their off-ball support throughout the match vs those who fade? We compare each player’s ASI in the first half vs second half.

# Calculate player fatigue metrics across all 10 matches
qualified = calculate_player_fatigue(loader, config, min_opps=15)
print(f"Players with 15+ opportunities in both halves: {len(qualified)}")
Players with 15+ opportunities in both halves: 151
# Show top maintainers (lowest fatigue drop)
# Note: Negative fatigue_drop = H2 improvement (player increased ASI after halftime)
qualified[['player_name', 'h1_asi', 'h2_asi', 'fatigue_drop', 'total_opps']].round(3).head(10)
player_name h1_asi h2_asi fatigue_drop total_opps
76 J. Yull 0.457 0.699 -0.241 202.0
58 D. Pierias 0.231 0.465 -0.234 209.0
144 L. Bayliss 0.604 0.784 -0.181 210.0
142 E. Adams 0.419 0.595 -0.176 215.0
129 J. Randall 0.606 0.777 -0.171 202.0
68 H. Van Der Saag 0.403 0.554 -0.150 185.0
153 M. Ruhs 0.498 0.643 -0.145 313.0
204 F. Talladira 0.403 0.538 -0.135 233.0
193 A. Bugarija 0.667 0.800 -0.133 179.0
44 J. Brimmer 0.525 0.653 -0.128 661.0 
# Visualize: Top 5 Maintainers vs Top 5 Faders
fig = plot_fatigue_comparison(qualified, top_n=5)
print_fatigue_summary(qualified)


Fatigue Analysis Summary (151 players):
  Players who improved H1→H2: 63
  Players who declined H1→H2: 87
  Avg fatigue drop: 1.0%

Key Finding: More players show declining off-ball support in the second half (58%) than improving (42%), with an average 1% drop. However, individual variation is substantial — some players significantly increase their movement after halftime (possibly “warming into” the match), while others show clear fatigue patterns. This player-level analysis could help coaches identify which players need earlier substitution or targeted conditioning work.

Impact & Use Cases

Stakeholder Application
Scouts Identify “highly-active supporter” players who consistently work off-ball to support teammates under pressure
Coaches Diagnose static tendencies; design training to improve support patterns
Analysts Compare team playing styles; evaluate support quality by zone
Players Objective feedback on off-ball contribution

Future Improvements

  • Context-aware “active support”: Account for whether a player is already open (opponent proximity). A stationary player in space may be better positioned than one running into traffic.
  • Advanced support classification: Incorporate velocity direction vectors and pass opening angle calculations to distinguish between movement toward useful space versus movement away from the play.