Chapter 23: Case Study 1 - The 2017 NBA Draft: Quantifying the Tatum vs. Fultz Decision

Introduction

The 2017 NBA Draft presented one of the most consequential draft decisions in recent memory. The Philadelphia 76ers, holding the #1 pick, selected Markelle Fultz from Washington. The Boston Celtics, after trading down from #1 to #3, selected Jayson Tatum from Duke. Within two years, this sequence was widely viewed as a historic blunder by Philadelphia and a masterful move by Boston.

This case study examines how draft models would have evaluated these prospects pre-draft, what signals (if any) indicated Tatum was the superior prospect, and what lessons we can draw for draft modeling methodology.

Part 1: Pre-Draft Statistical Profiles

Markelle Fultz - Washington (2016-17)

Basic Statistics: - 23.2 PPG, 5.7 RPG, 5.9 APG - 47.6% FG, 41.3% 3PT, 64.9% FT - Minutes: 35.7 per game (25 games) - Age at draft: 19.1 years

Per-100-Possession Rates: - Points: 32.8 - Rebounds: 8.1 - Assists: 8.4

Conference: Pac-12 (adjustment factor: 1.00)

Jayson Tatum - Duke (2016-17)

Basic Statistics: - 16.8 PPG, 7.3 RPG, 2.1 APG - 45.2% FG, 34.2% 3PT, 84.9% FT - Minutes: 33.1 per game (33 games) - Age at draft: 19.1 years

Per-100-Possession Rates: - Points: 24.2 - Rebounds: 10.5 - Assists: 3.0

Conference: ACC (adjustment factor: 1.05)

Physical Measurements

Part 2: Translation and Projection Analysis

Statistical Translation

Fultz NBA Projection (Year 1):

Points: 32.8 × 0.72 × 1.00 = 23.6 pts/100 poss
Rebounds: 8.1 × 0.85 × 1.00 = 6.9 reb/100 poss
Assists: 8.4 × 0.68 × 1.00 = 5.7 ast/100 poss

Tatum NBA Projection (Year 1):

Points: 24.2 × 0.72 × 1.05 = 18.3 pts/100 poss
Rebounds: 10.5 × 0.85 × 1.05 = 9.4 reb/100 poss
Assists: 3.0 × 0.68 × 1.05 = 2.1 ast/100 poss

Red Flag Analysis

Fultz Red Flags: 1. Free throw percentage: 64.9% - Well below the 70% threshold - Risk multiplier: 1.30 - This was THE critical data point that models should have flagged

2. Small sample size: 25 games - Washington went 9-22 - Limited high-quality competition

3. Usage rate: Very high (35%+) - Production came with dominant ball usage - Unclear how skills translate to complementary role

Tatum Red Flags: 1. Lower scoring volume - Shared touches with other lottery picks (Giles, Kennard) - Usage-adjusted scoring was competitive

2. Assist rate - Limited playmaking creation

Hidden Strengths

Tatum Hidden Strengths: 1. Free throw percentage: 84.9% - Elite FT% strongly correlates with NBA three-point shooting - Projected shooting improvement likely

2. Size/length profile - 6'7" with 6'11" wingspan - Ideal wing dimensions

3. Conference-adjusted scoring - ACC production against superior competition - Tournament success (Elite Eight)

Part 3: Model-Based Projections

Career Win Shares Projection

Using the draft model methodology from Chapter 23:

Fultz Projection:

Base projection from statistics: 58 career WS
Age adjustment: × 1.10 (freshman)
FT% penalty: × 0.85
Conference adjustment: × 1.00
Final projection: 54 WS
90% CI: [22, 86]

Tatum Projection:

Base projection from statistics: 45 career WS
Age adjustment: × 1.10 (freshman)
FT% bonus: × 1.05
Conference adjustment: × 1.05
Size bonus: × 1.05
Final projection: 55 WS
90% CI: [25, 85]

Key Observation

Even a properly calibrated model would have had Fultz and Tatum as essentially equivalent in expected value, with overlapping confidence intervals. The difference was in the risk profile, not the central projection.

Risk-Adjusted Projections

Bust Probability: - Fultz: ~32% (elevated by FT%, small sample) - Tatum: ~22% (standard for top-3 pick)

All-Star Probability: - Fultz: ~35% - Tatum: ~30%

Adjusted Value (accounting for risk): - Fultz: 54 × (1 - 0.32 × 0.5) = 45.4 risk-adjusted WS - Tatum: 55 × (1 - 0.22 × 0.5) = 49.0 risk-adjusted WS

When properly accounting for the FT% red flag, Tatum had higher risk-adjusted value.

Part 4: What Models Should Have Shown

The Free Throw Signal

Markelle Fultz's 64.9% free throw shooting should have been the dominant signal in any draft analysis. Historical data strongly shows:

Guards with sub-65% FT in college: - 45% bust rate (vs. 25% baseline) - Average NBA career WS: 18 (vs. 35 for guards with 80%+ FT) - Three-point shooting translation: Very poor

This single data point should have raised serious concerns. The question is why it didn't dominate the pre-draft discourse.

Possible Explanations for the Miss

1. Sample size concerns worked both ways - 25 games gave limited FT attempts - Evaluators may have dismissed it as noise - But small samples should increase uncertainty, not dismiss warnings

2. Video scouting contradiction - Fultz's shooting looked smooth on film - Led to cognitive dissonance between eye test and statistics

3. Recency bias - Strong late-season performances - Shooting slump early in season attributed to "adjusting"

4. Consensus momentum - Once Fultz was established as #1, confirmation bias dominated - Mock drafts create feedback loops

What a Properly Weighted Model Would Have Shown

A properly calibrated model would have shown Tatum as the superior risk-adjusted pick, even if Fultz had slightly higher upside.

Part 5: Actual Outcomes

Career Results (Through 2024)

Markelle Fultz: - Diagnosed with thoracic outlet syndrome affecting shooting motion - Career (through 2023-24): ~8 Win Shares total - Never developed into projected star - Currently: Rotation player (Orlando Magic)

Jayson Tatum: - 5× All-Star - 3× All-NBA First Team - Career Win Shares (through 2023-24): ~65 and climbing - Projected career total: 120+ WS

Model Evaluation

Fultz's outcome was catastrophically bad, falling well outside typical model bounds. Tatum's outcome has exceeded projections, though within the confidence interval.

Part 6: Lessons for Draft Modeling

Lesson 1: Free Throw Percentage Is Not Optional

The Fultz case reinforces that sub-70% free throw shooting for perimeter players should be a near-disqualifying red flag. Models should heavily weight this variable and perhaps apply non-linear penalties for extremely low values.

Recommendation: Apply a multiplicative bust probability penalty of 1.5x for guards shooting below 65% FT.

Lesson 2: Confidence Intervals Matter More Than Point Estimates

Both players had similar point projections (~55 WS). The difference was in the risk profiles. Decision-makers should: - Focus on distribution shapes, not just means - Consider downside scenarios explicitly - Apply risk penalties for high-variance prospects in non-rebuilding situations

Lesson 3: Small Samples Increase Uncertainty

Fultz's 25-game sample should have widened confidence intervals and reduced confidence in the projection. Instead, evaluators treated his performance as more reliable than it was.

Recommendation: Apply explicit sample size penalties that increase projection uncertainty for limited data.

Lesson 4: Consensus Can Be Wrong

The pre-draft consensus overwhelmingly favored Fultz. This consensus was wrong. Draft models should: - Be independent of mock draft consensus - Weight statistical signals appropriately regardless of scouting consensus - Flag cases where statistical and scouting evaluations diverge

Lesson 5: Injury Risk Is Non-Zero

Fultz's career was derailed by an unusual injury (thoracic outlet syndrome). No model could have specifically predicted this. But: - High-variance outcomes include injury scenarios - Projection confidence intervals should account for injury probability - Health history and red flags deserve weight in evaluation

Part 7: Building a Better 2017 Draft Model

Revised Feature Weighting

Based on the Fultz/Tatum case, suggested feature weight adjustments:

Non-Linear FT% Treatment

def ft_adjustment(ft_pct):
    if ft_pct >= 0.80:
        return 1.05  # Bonus for elite FT shooting
    elif ft_pct >= 0.75:
        return 1.00  # Neutral
    elif ft_pct >= 0.70:
        return 0.95  # Minor concern
    elif ft_pct >= 0.65:
        return 0.85  # Significant concern
    else:
        return 0.70  # Major red flag

Retrospective Model Output

With revised weighting, the 2017 model would have shown:

Exercises

Exercise 1

Recalculate Fultz's projection using the revised FT% adjustment function above. How does this change his risk-adjusted value relative to Tatum?

Exercise 2

Identify another historical draft where a shooting red flag was ignored. What were the consequences?

Exercise 3

Design a validation study to determine the optimal weight for free throw percentage in draft models.

Exercise 4

The Boston Celtics traded down from #1 to #3, acquiring additional assets. Using pick value curves, evaluate whether this trade was justified even if Fultz had panned out.

Conclusion

The 2017 draft class will be studied for years as an example of consensus failure. Markelle Fultz's 64.9% free throw shooting was a clear statistical red flag that proper draft models should have flagged. While no model could have predicted his specific injury, appropriate risk adjustment would have favored Jayson Tatum.

The key lesson is not that Tatum was obviously better than Fultz pre-draft—reasonable analysts could disagree. The lesson is that Fultz carried meaningfully higher risk that should have been reflected in his valuation. Teams that built better risk-adjusted models would have made better decisions.