How Advanced are Advanced Stats? A Comparison of Advanced Player Statistics for NBA Competition

Abstract

Like many sports, modern basketball bears little resemblance to its original form. At its conception in 1897, there was no three-point line, dribbling was illegal, and the number of players on the court could range from three to forty. Over the past 125 years, the game has evolved dramatically, and the days of shooting into peach baskets are long gone. Today’s modern game is fueled by advancements in technology and analytical thinking. Drawing inspiration from the sabermetrics movement in baseball, basketball teams and players are frantically searching for any edge they can obtain through statistical analysis. Subsequently, player evaluation methodologies have changed dramatically. Traditional stats such as points, rebounds, assists and turnovers barely scrape the surface in measuring player performance. Advanced stats such as true shooting percentage, usage rate, and wins above replacement have been added to a vast pool of metrics. This thesis will attempt to compare the descriptive power of four of these advanced player efficiency metrics by analyzing their ability to predict historical outcomes. Utilizing multiple different regressions of past player performance and game outcomes, predictions will be made on the point total, point spread, and moneyline of individual games. Presumably, the metric which most accurately evaluates player efficiency will have the highest accuracy of prediction. Next, the thesis will turn towards the predictive power of these metrics by analyzing their ability to predict future outcomes. Using the same types of regressions, we will analyze whether any of these statistics can be used to assemble a profitable betting strategy. If there are any inefficiencies in the gambling market, it is possible these advanced statistics may help to reveal them. Such findings could be quite useful to both sportsbooks and bettors alike.