This is a review of the NBA research conducted by Alexander Franks, Andrew Miller, Luke Bornn, and Kirk Goldsberry applying multinomial logistic regression.
Basketball is a game based on both offensive and defensive skill. However, to date most basketball statistics deal with the offensive side of the game. This limits the ability to evaluate teams and players to strictly one dimension. In an attempt to fill this void, five new defensive statistics are created: volume score, disruption score, defensive shot charts, shots against, and counterpoints.
The first step is creating a model to estimate defensive matchups at every moment in a basketball game. This process was completed for all basketball games played during the NBA 2013-2014 season. The results provide information regarding who is defending who at every point of a possession by estimating an average defender position as a function of offender, ball, and hoop locations. A hidden Markov model is then employed to determine the progression of defensive matchup over the course of a possession.
From this model, the volume score and disruption score statistics are determined. The volume score is the measurement of total attempts a defender faces in a game, which is computed with a multinomial logistic regression. Disruption score measures the ability a defender has in reducing the effectiveness of the opponent he is defending, calculated using a logistic regression to predict shots made and shots missed. From there a defensive shot chart is created to visualize the volume and disruption scores.
The issue with these statistics is that they are static, while the defenders are not. The defense is a continually flowing process with defenders changing which member of the opponent’s team they are guarding. To deal with the dynamic nature of the game, counterpoints are assigned. Counterpoints are a weighted average of points scored per 100 possessions against a defender. Three methods are created for counterpoints, or points against. The original matchup method assigns counterpoints to the defender who was guarding the shooter at the beginning of the possession. The pre-shot matchup method assigns counterpoints to the defender who was guarding the shooter when the shot was taken. The fractional method assigns counterpoints proportionally with each defender receiving points based on what fraction of the possession they guarded the shooter.
All counterpoints look at the number of shot attempts made and the average number of points scored against a defender. These measures provide coaches and analysts with a way to rank players based on their defensive skills. This would be useful when looking at making trades and looking for players to fulfill specific defensive needs.
Taken together, these defensive statistics provide methods to quantify the defensive value of individual players. High ranked defensive players can be analyzed to determine what makes them effective. This information can then be used to create other strong defensive players as well to increase the offensive’s ability to counteract those skills.
Combining both offensive and defensive statistics provides a well-rounded evaluation of each player and their contribution to the team and the game.
Analytics Used: Volume Score, Disruption Score, Defensive Shot Charts, Shots Against, Counterpoints, Hidden Markov Model, Multinomial Logistic Regression, Logistic Regression
Find out how Sports Analytics Expert Victor Holman can give your team the competitive advantage.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Discover the Groundbreaking Sports Analytics Application and Framework coaches and sports analysts are talking about!
Learn all about sports analytics in Victor Holman’s Sports Analytics Blog.