ABSTRACT

Until recently, the field of sports analytics was constrained by the data collected by officials and fans. In basketball, for example, play-by-play data has existed for decades, capturing who is on the court, as well as outcomes for each play. Even with a coarsening of the action

of Methods and

to the level of possession or play outcomes, significant analysis of the game has been possible. Examples include basic statistical summaries such as field goal percentage and usage rate, as well as more advanced metrics such as regression-adjusted plus-minus variants and win shares (Kubatko et al., 2007). See Oliver (2004) and Shea and Baker (2013) for book-length treatments of the types of analyses possible with traditional, hand-collected basketball data. Despite the relative success of employing play-by-play and box score data to understand basketball, such data inherently lacks information on actions leading to the outcome of the play and fails to capture contributions of the other nine players on the court for each play-by-play event.