top of page


Improving Sports Analytics

Salil Singh, AJ Adejare


Identifying the role a given player plays in a team’s success. A number of expected value and win probability-based statistics across sports are often used crudely to attribute success responsibility to players. However, a mechanistic or interpretable understanding remains elusive. Much of what we have is highly circumstantial – on lineups, personnel, opponents, strategies/schemes. Can we focus on a single sport or survey the landscape to identify the sports where most success has been achieved on such problems to port them to specific sports (e.g.: Chess’ elo has found repurposed usage in other sports, particularly for predictive and gambling purposes; baseball sabermetrics)?
NFL: Focussing on the offensive side would be a good starting point. Can we identify the roles position groupings – quarterback, offensive line (interior and exterior), receiving corps, backs – are playing? Raw counting statistics, as well as play-by-play scheme-aware (Next Gen Stats) data, are publicly available. The NFL data bowl would be a good target destination for such work.
NBA: Similar challenges, if more tractable, exist for the NBA.
Soccer: xG and new age statistics. The salience of singular plays is hard to capture with aggregate statistical approaches.

There are a number of different perspectives and focusses this can take. The hope would be to do something that can ideally be tested against existing metrics and the betting/prediction markets.

Screen Shot 2022-06-03 at 11.31.35 AM.png
github URL
bottom of page