Key Takeaways: Defensive Analytics

One-page reference for Chapter 10 concepts

The Defensive Attribution Challenge

Team-Level Defensive Metrics

# EPA Allowed (core metric)
defense_epa = (plays
    .groupby('defteam')
    .agg(
        epa_allowed=('epa', 'mean'),
        success_allowed=('epa', lambda x: (x > 0).mean())
    )
)
# Lower (more negative) is better

EPA Allowed Interpretation

Pass Defense Metrics

# Pressure rate estimation
passes['pressured'] = (
    (passes['sack'] == 1) |
    (passes['qb_hit'] == 1) |
    (passes['qb_scramble'] == 1)
)
pressure_rate = passes['pressured'].mean()

# Coverage (team level)
comp_pct_allowed = non_sacks['complete_pass'].mean()
yards_per_target = non_sacks['yards_gained'].mean()

Run Defense Metrics

# Stuff rate (runs at/behind LOS)
stuff_rate = (rushes['yards_gained'] <= 0).mean()

# YPC allowed
ypc_allowed = rushes['yards_gained'].mean()

# Explosive prevention
explosive_10_rate = (rushes['yards_gained'] >= 10).mean()

Sack Rate vs Pressure Rate

Situational Defense

# Third down
third_conv_allowed = (third_downs
    .groupby('defteam')
    ['third_down_converted']
    .mean()
)

# Red zone
rz_td_rate = (red_zone
    .groupby('defteam')
    ['touchdown']
    .mean()
)

Turnover Analysis

Key insight: Turnovers are largely random

• Year-to-year INT rate correlation: ~0.25
• Fumble recovery is nearly random (~50% rate)
• Don't overweight turnover metrics

turnover_rate = (
    plays['interception'] | plays['fumble_lost']
).mean()

Opponent Adjustment

# Simple adjustment
off_quality = plays.groupby('posteam')['epa'].mean()
expected_epa = plays.groupby('defteam').agg(
    expected=('opp_offense_quality', 'mean')
)
adjusted_epa = raw_epa - expected_epa

Data Source Limitations

Individual Evaluation Challenge

What we CAN'T measure from PBP: - Who was supposed to cover whom - Pass rush wins vs losses - Positioning and angles - Assignment correctness

Requires: PFF grades, SIS charting, or tracking data

Pass/Run Defense Split

pass_epa = passes.groupby('defteam')['epa'].mean()
run_epa = rushes.groupby('defteam')['epa'].mean()

# Positive = better pass defense
balance = run_epa - pass_epa

Common Pitfalls

Deep Pass Defense

deep_passes = passes[passes['air_yards'] >= 15]
deep_comp_pct = deep_passes.groupby('defteam')['complete_pass'].mean()
# High-value area to defend

YAC Allowed

# Reflects tackling/pursuit
yac_allowed = (completions
    .groupby('defteam')
    ['yards_after_catch']
    .mean()
)

Evaluation Framework

1. Overall
   - EPA per play allowed
   - Success rate allowed

2. Pass Defense
   - Pass EPA allowed
   - Sack/pressure rate
   - Comp % allowed

3. Run Defense
   - Run EPA allowed
   - Stuff rate
   - YPC allowed

4. Situational
   - 3rd down conversion
   - Red zone TD rate

5. Context
   - Opponent adjustment
   - Game script filter

Key Limitations

• Individual attribution nearly impossible
• Turnovers are high variance
• Scheme effects not captured
• QB/WR quality affects results
• Sample sizes per player small

Preview: Chapter 11

Next: Special Teams Analytics - evaluating kicking, punting, and returns using EPA frameworks.