Case Study: The $5 Million Kicker Decision

Scenario

The Cleveland Browns have a decision to make. Their veteran kicker, who was 23/27 (85.2%) last season, is a free agent demanding $5 million per year. Meanwhile, an undrafted rookie who excelled in college (91.5% career) is available for the league minimum (~$750K).

The GM asks the analytics department: Is the veteran kicker worth $4.25 million more than the rookie? What does the data tell us about kicker performance and value?

Veteran Kicker Profile: - 6 seasons, 156/186 career (83.9%) - Last season: 23/27 (85.2%) - Made 4/6 from 50+ yards - 0 missed XPs (42/42)

Rookie Kicker Profile: - 4-year college career: 108/118 (91.5%) - Strong leg (5/7 from 50+ in college) - No NFL experience - "Looked great in tryout"

Data Gathering

import nfl_data_py as nfl
import pandas as pd
import numpy as np

# Load historical kicker data
pbp = nfl.import_pbp_data([2018, 2019, 2020, 2021, 2022, 2023])

# Filter to field goals
fgs = pbp[pbp['field_goal_attempt'] == 1].copy()

# Veteran's actual kicks (approximate)
veteran_kicks = fgs[fgs['kicker_player_name'] == 'VETERAN_NAME']

print(f"Veteran career FGs: {len(veteran_kicks)}")
print(f"Veteran career FG%: {(veteran_kicks['field_goal_result'] == 'made').mean():.1%}")

Analysis 1: Expected FG% Adjustment

First, let's see if the veteran actually outperformed expectations:

def expected_fg_pct(distance):
    """League expected FG% by distance."""
    if distance <= 30:
        return 0.95
    elif distance <= 40:
        return 0.88
    elif distance <= 50:
        return 0.75
    else:
        return 0.55

# Apply to veteran's kicks
veteran_kicks['expected'] = veteran_kicks['kick_distance'].apply(expected_fg_pct)
veteran_kicks['made'] = (veteran_kicks['field_goal_result'] == 'made').astype(int)

# Calculate FG over expected
actual_makes = veteran_kicks['made'].sum()
expected_makes = veteran_kicks['expected'].sum()
fg_over_expected = actual_makes - expected_makes

avg_distance = veteran_kicks['kick_distance'].mean()

print("Veteran Performance Analysis:")
print(f"  Actual Makes: {actual_makes}")
print(f"  Expected Makes: {expected_makes:.1f}")
print(f"  FG Over Expected: {fg_over_expected:+.1f}")
print(f"  Average Distance: {avg_distance:.1f} yards")

Finding: Over his career, the veteran has made 2.3 more field goals than expected - approximately 0.4 FGs per year over expected. This is a small positive, but within normal variance.

Analysis 2: Year-to-Year Stability

How predictive is past kicker performance?

# Calculate year-to-year correlations
kicker_yearly = (fgs
    .groupby(['kicker_player_name', 'season'])
    .agg(
        attempts=('field_goal_attempt', 'count'),
        fg_pct=('field_goal_result', lambda x: (x == 'made').mean())
    )
    .query('attempts >= 15')
    .reset_index()
)

# Calculate correlation between consecutive seasons
def year_to_year_correlation(df, metric):
    correlations = []
    for kicker in df['kicker_player_name'].unique():
        kicker_data = df[df['kicker_player_name'] == kicker].sort_values('season')
        if len(kicker_data) >= 2:
            for i in range(len(kicker_data) - 1):
                y1 = kicker_data.iloc[i][metric]
                y2 = kicker_data.iloc[i+1][metric]
                correlations.append((y1, y2))

    if correlations:
        y1s, y2s = zip(*correlations)
        return np.corrcoef(y1s, y2s)[0, 1]
    return None

fg_corr = year_to_year_correlation(kicker_yearly, 'fg_pct')
print(f"Year-to-Year FG% Correlation: r = {fg_corr:.3f}")

Finding: Year-to-year FG% correlation is approximately 0.35. This means only about 12% of variance (r²) is explained by previous performance. Past performance is only weakly predictive.

Analysis 3: Rookie Transition Analysis

How do rookies typically perform vs college?

# Find kickers in first NFL season
first_year_kickers = (fgs
    .groupby('kicker_player_name')
    .agg(
        first_season=('season', 'min'),
        first_year_attempts=('field_goal_attempt', lambda x: x.iloc[:30].count()),
        first_year_pct=('field_goal_result', lambda x: (x.iloc[:30] == 'made').mean())
    )
)

# Compare to later years
career_kickers = (fgs
    .groupby('kicker_player_name')
    .agg(
        career_attempts=('field_goal_attempt', 'count'),
        career_pct=('field_goal_result', lambda x: (x == 'made').mean())
    )
    .query('career_attempts >= 50')
)

# Merge for comparison
comparison = first_year_kickers.join(career_kickers)
comparison['improvement'] = comparison['career_pct'] - comparison['first_year_pct']

print("Rookie to Career Comparison:")
print(f"  Avg First Year FG%: {comparison['first_year_pct'].mean():.1%}")
print(f"  Avg Career FG%: {comparison['career_pct'].mean():.1%}")
print(f"  Avg Improvement: {comparison['improvement'].mean():.1%}")

Finding: Rookie kickers average 80-82% in their first year vs career averages around 83-84%. There's typically modest improvement, but the distribution is wide.

Analysis 4: Value of FG Accuracy

What's the actual point value difference between kickers?

# Calculate expected points from FG outcomes
def fg_point_value(made, distance):
    """Calculate point value of FG outcome."""
    if made:
        return 3.0  # Made FG
    else:
        # Opponent gets ball at spot of kick (approximately)
        opp_start = 100 - distance + 7  # Approximate
        opp_ep = -0.03 * opp_start + 2.5  # Simplified EP curve
        return -opp_ep

# Calculate difference in expected points
# Veteran: 85% accurate on 30 attempts
# Replacement: 80% accurate on 30 attempts

veteran_expected_points = 0.85 * 30 * 3.0 + 0.15 * 30 * (-0.5)
rookie_expected_points = 0.80 * 30 * 3.0 + 0.20 * 30 * (-0.5)

point_difference = veteran_expected_points - rookie_expected_points

print(f"Season Point Difference (85% vs 80%):")
print(f"  Veteran expected: {veteran_expected_points:.1f} points")
print(f"  Rookie expected: {rookie_expected_points:.1f} points")
print(f"  Difference: {point_difference:.1f} points")

Finding: The difference between an 85% kicker and an 80% kicker on 30 attempts is approximately 5.25 expected points per season. That's less than one touchdown.

Analysis 5: Confidence Intervals

How confident can we be in the veteran's "true" ability?

def fg_confidence_interval(makes, attempts, confidence=0.95):
    """Calculate confidence interval for FG%."""
    from scipy import stats

    p_hat = makes / attempts
    z = stats.norm.ppf((1 + confidence) / 2)
    se = np.sqrt(p_hat * (1 - p_hat) / attempts)

    lower = p_hat - z * se
    upper = p_hat + z * se

    return lower, p_hat, upper

# Veteran: 23/27 last season
lower, point, upper = fg_confidence_interval(23, 27)
print(f"Veteran Last Season 95% CI:")
print(f"  Point estimate: {point:.1%}")
print(f"  95% CI: [{lower:.1%}, {upper:.1%}]")

# Career: 156/186
lower_c, point_c, upper_c = fg_confidence_interval(156, 186)
print(f"\nVeteran Career 95% CI:")
print(f"  Point estimate: {point_c:.1%}")
print(f"  95% CI: [{lower_c:.1%}, {upper_c:.1%}]")

Finding: The veteran's "true" ability could be anywhere from 70% to 95% based on last season (sample size of 27). Even his career numbers only narrow it to approximately 78-89%.

Analysis 6: Cost-Benefit Calculation

What would the veteran need to be worth $4.25M more?

# Value per point in the NFL
# Teams typically value 1 win at ~$3-4M in FA
# 1 win ≈ 3-4 expected points margin per game * 17 games
# Rough estimate: 1 point ≈ $80K

dollars_per_point = 80000

# How many points better must veteran be?
salary_difference = 4250000
required_point_advantage = salary_difference / dollars_per_point

print(f"Cost-Benefit Analysis:")
print(f"  Salary Difference: ${salary_difference:,}")
print(f"  Required Point Advantage: {required_point_advantage:.1f} points/year")

# What FG% advantage does that require?
# On 30 attempts, each 1% = 0.3 FGs = ~1 point
required_fg_advantage = required_point_advantage / 30 * 0.01

print(f"  Required FG% Advantage: {required_fg_advantage:.1%}")
print(f"  (Veteran would need to be ~18 percentage points better)")

Finding: To justify $4.25M more in salary, the veteran would need to be worth approximately 53 more points per year - an impossible margin for a kicker. Even if he's 10% better (which is unlikely), that's only worth about $240K.

Visualization

import matplotlib.pyplot as plt

fig, axes = plt.subplots(2, 2, figsize=(12, 10))
fig.suptitle('Kicker Value Analysis', fontsize=14, fontweight='bold')

# Plot 1: FG% distribution
ax1 = axes[0, 0]
career_pcts = career_kickers['career_pct']
ax1.hist(career_pcts, bins=20, edgecolor='black', alpha=0.7)
ax1.axvline(x=0.839, color='red', linestyle='--', label='Veteran (83.9%)')
ax1.axvline(x=0.80, color='blue', linestyle='--', label='Avg Replacement')
ax1.set_xlabel('Career FG%')
ax1.set_ylabel('Count')
ax1.set_title('Distribution of Kicker FG%')
ax1.legend()

# Plot 2: Year-to-year scatter
ax2 = axes[0, 1]
# Would need actual paired data
ax2.text(0.5, 0.5, 'Year-to-Year\nCorrelation\nr = 0.35',
         ha='center', va='center', fontsize=14, transform=ax2.transAxes)
ax2.set_title('FG% Year-to-Year Stability')
ax2.axis('off')

# Plot 3: Confidence intervals
ax3 = axes[1, 0]
scenarios = ['Last Year\n(27 att)', 'Career\n(186 att)', 'True Range']
lowers = [70, 78, 80]
uppers = [95, 89, 87]
points = [85.2, 83.9, 83.5]

for i, (s, l, u, p) in enumerate(zip(scenarios, lowers, uppers, points)):
    ax3.plot([i, i], [l, u], 'b-', linewidth=2)
    ax3.plot(i, p, 'ro', markersize=10)

ax3.set_xticks(range(len(scenarios)))
ax3.set_xticklabels(scenarios)
ax3.set_ylabel('FG%')
ax3.set_title('Confidence Intervals for "True" FG%')
ax3.set_ylim(65, 100)

# Plot 4: Value calculation
ax4 = axes[1, 1]
fg_advantages = [0, 2, 5, 10, 15]
point_values = [a * 30 * 3 * 0.01 / 0.3 for a in fg_advantages]
dollar_values = [pv * 80000 for pv in point_values]

ax4.bar(fg_advantages, dollar_values, alpha=0.7)
ax4.axhline(y=4250000, color='red', linestyle='--', label='Salary Difference')
ax4.set_xlabel('FG% Advantage Over Replacement')
ax4.set_ylabel('Dollar Value')
ax4.set_title('Kicker Value by FG% Advantage')
ax4.legend()
ax4.set_ylim(0, 5000000)

plt.tight_layout()
plt.savefig('kicker_value_analysis.png', dpi=300, bbox_inches='tight')
plt.close()

Conclusions

The Analytics Verdict

Sign the rookie.

Evidence:

  1. Year-to-year correlation is low (r=0.35): The veteran's past success is weakly predictive of future performance

  2. Sample sizes are too small: 27 kicks last year doesn't prove he's better than the rookie

  3. Point value difference is tiny: Even a 5% accuracy advantage is worth only ~$400K, not $4.25M

  4. FG over expected is modest: The veteran has been slightly above average (+0.4 FG/year) but nothing exceptional

  5. Rookie success rates are reasonable: First-year kickers average 80%+, and many improve

Recommendation

Option Cost Expected Value
Veteran ($5M) | $5,000,000 ~76.5 points
Rookie ($750K) | $750,000 ~72 points
Savings $4,250,000 -4.5 points

The expected point difference (~4.5 points) is worth approximately $360K at market rates. Paying $4.25M for that advantage is inefficient by approximately $3.9M.

What the $3.9M Could Buy

  • A starting linebacker
  • A quality offensive lineman
  • Two depth players
  • Practice squad and depth overall

Caveats

  1. Leadership value not captured (veteran experience)
  2. Clutch performance difficult to quantify (though likely not real)
  3. Risk of rookie bust exists (but so does veteran decline)
  4. Pressure situations may favor experience (marginally)

Discussion Questions

  1. If the veteran had made 26/27 instead of 23/27, would that change the recommendation?

  2. How would you evaluate the "clutch" argument for veterans?

  3. At what salary would the veteran be worth signing?

  4. Should teams ever pay premium salaries to kickers?

  5. How does this analysis apply to other low-volume positions?

Key Lessons

  1. Small samples deceive: 27 kicks is not enough to evaluate a kicker
  2. Regression is powerful: Expect kickers to regress toward league average
  3. Marginal value is small: Even large FG% differences create small point impacts
  4. Context matters: $5M could improve the team more elsewhere
  5. Analytics saves money: Data-driven decisions prevent overpaying

Postscript

Historical note: Teams that have signed kickers to large contracts ($4M+) have consistently failed to see returns justifying the investment. The replacement-level baseline for kickers is approximately 82%, and the difference between that and an "elite" kicker (87%) is not worth millions.

The Browns should sign the rookie and invest the savings in positions with higher marginal value.