Chapter 21: Further Reading - Win Probability Models

Academic Papers

Win Probability Methodology

  1. "A Win Probability Model for Football" - Burke (2010) - Foundation of modern WP models - Feature selection methodology - Calibration approaches

  2. "Expected Points and Expected Points Added" - Burke - EP foundation for WP models - State-value framework - Integration with WP

  3. "Calibration of Probabilistic Predictions" - Gneiting & Raftery - Calibration theory - Proper scoring rules - Evaluation metrics

Machine Learning Applications

  1. "Gradient Boosting for Win Probability" - MIT Sloan - Advanced model architectures - Feature engineering - Comparison to baselines

Books

Sports Analytics

  1. "The Book: Playing the Percentages" - Tango, Lichtman, Dolphin - Win probability in baseball - Transferable concepts - Decision analysis framework

  2. "Mathletics" - Wayne Winston - Multi-sport probability models - Basic methodology - Practical applications

Statistical Methods

  1. "Applied Logistic Regression" - Hosmer & Lemeshow - Logistic regression theory - Calibration assessment - Model diagnostics

  2. "Probabilistic Machine Learning" - Murphy - Calibration methods - Bayesian approaches - Neural network outputs

Online Resources

Football Analytics Sites

  1. Pro Football Reference - Win probability tools - Historical data - https://www.pro-football-reference.com/

  2. ESPN Win Probability - Live WP tracking - Methodology explanations - https://www.espn.com/

  3. Football Outsiders - WP-based analysis - Decision analysis - https://www.footballoutsiders.com/

  4. The Athletic - WP visualizations - Fourth-down analysis - https://theathletic.com/

Academic Resources

  1. NFL Big Data Bowl - WP model submissions - Code examples - https://www.kaggle.com/c/nfl-big-data-bowl

Data Sources

Play-by-Play Data

  1. nflverse (R) - Comprehensive PBP data - WP calculations included - https://nflverse.com/

  2. nfl_data_py (Python) - Python NFL data access - Pre-calculated WP

  3. College Football Data API - College PBP data - https://collegefootballdata.com/

Historical Archives

  1. Sports Reference - Historical game data - Play-level data

Tools and Libraries

Python

  1. scikit-learn - Logistic regression - Calibration utilities - sklearn.calibration

  2. XGBoost - Gradient boosting - Built-in calibration - https://xgboost.readthedocs.io/

  3. PyTorch - Neural network WP models - https://pytorch.org/

  4. matplotlib/seaborn - Calibration curves - WP charts

R

  1. nflfastR - NFL WP calculations - Pre-built models

  2. ggplot2 - WP visualizations

Industry Resources

Team Analytics

  1. NFL Next Gen Stats - Win probability features - Real-time tracking

  2. ESPN Analytics - Fourth-down decisions - Live WP

Media Applications

  1. FiveThirtyEight - Game predictions - WP methodology articles - https://fivethirtyeight.com/

Methodological Deep Dives

Calibration Techniques

  1. Platt Scaling - Original paper on probability calibration - Post-hoc calibration method

  2. Isotonic Regression - Non-parametric calibration - scikit-learn implementation

  3. Temperature Scaling - Neural network calibration - Simple and effective

Feature Engineering

  1. Interaction Terms - Score × Time interactions - Field position effects

  2. Time Decay Features - Modeling urgency - Late-game dynamics

Suggested Learning Path

Week 1-2: Foundations

  • Study logistic regression for probability
  • Implement basic WP with score/time
  • Understand calibration concepts

Week 3-4: Feature Engineering

  • Add game state features
  • Create interaction terms
  • Handle edge cases

Week 5-6: Advanced Models

  • Implement gradient boosting
  • Experiment with neural networks
  • Compare model performance

Week 7-8: Calibration

  • Analyze calibration curves
  • Apply isotonic regression
  • Validate on held-out data

Week 9+: Applications

  • Build WPA calculator
  • Implement decision analysis
  • Create visualizations