Statistics via Sports: Summer Lab 2024

• Taught as part of the Wharton Sports Analytics Summer Research Lab
• Recommended prerequisites: calculus, probability, R coding; matrix algebra helps

Daily format

• 1 to 2 hour lecture
• Hands-on active learning lab where you will analyze a real-world sports dataset

Regression modeling

• Simple linear regression
  • planned lecture
  • lab, data: MLB pythagorean win percentage, MLB team payroll
• Multivariable linear regression
  • planned lecture
  • note: estimating the coefficients
  • lab, data: NBA team-seasons for the four factors, punts
• Example of the research process
  • planned lecture
  • lab:
    • get into groups and start thinking about a research project
    • plan to read relevant literature and start with a replication of existing analysis
    • finish up the previous labs
• Logistic regression
  • planned lecture
  • note: logistic regression & gradient descent
  • lab, data: field goals, 2023-2024 NCAA men’s basketball game results and team info from Kaggle
• Confounding
  • planned lecture
  • lab, data: MLB half-innings data for park effects
• Models do what they’re told
  • planned lecture
  • lab, data: NFL expected points

Frequentist statistical inference and uncertainty quantification

• Significance and p-values
  • planned lecture
  • lab, data: diving, TTO (time through the order)
• Normal approximation (CLT) and binomial proportion confidence interval
  • planned lecture
  • lab, data: NBA Players 2023-2024
• The bootstrap
  • planned lecture
  • lab

Shrinkage & Bayesian statistics

• Priors & the power of fake data
  • planned lecture
  • lab
• Empirical Bayes
  • planned lecture
  • live lecture
  • lab, data: NBA player-game box scores, field goals
• Shrinkage estimation
  • planned lecture – need to re-write this
  • live lecture
  • lab, data: first putt success percentage training data and held-out test data
• Fully Bayesian models
  • planned lecture
  • live lecture
  • lab, data: NFL game-by-game data for Bayesian power rating and home field advantage model
• Regularization & ridge regression
  • planned lecture
  • lab, data: NBA lineup data; this is in a .rds format that takes significantly less storage than a .csv file, to convert to a .csv file in order to use in a non-R language you’ll need to load it into R and then save it as a .csv

Machine learning

• Bias-variance trade-off
  • planned lecture
  • lab, data: MLB half-innings data for park effects
• Decision trees
  • planned lecture
  • lab: just work on research
• Random forests & Boosting
  • planned lecture
  • lab, data: NFL play-by-play data for win probability modeling

Miscellaneous

• Kelly betting
  • planned lecture
  • live lecture
  • lab
• Clustering (K means hard clustering)
  • planned lecture
  • NBA synergy play type frequencies
    

If I had more time

• Topics included purely as labs that deserve their own lecture
  • Selection bias
  • Parametric inference
• Topics alluded to in class that deserve their own lecture
  • Multicollinearity
• Other topics
  • Tracking data
  • Causal inference
    • Regression (observational studies) is Not causation
    • Randomized controlled trials
  • Miscellaneous
    • Soft clustering (GMMs & EM algorithm)
    • Rare events
    • Multiple hypothesis testing (Bonferroni correction & Benjamini Hochberg)
    • GEV distribution for max race running time