One of the harder things about being a statistically savvy NBA fan is that there is no one be-all end-all metric like baseball’s WAR that basketball fans can point to. There are a variety of stats, of which I have written a fair amount about, but they all seem to have different ways of trying to calculate a player’s overall impact on the court. But this can be fun, as we can use these different metrics as ammo for arguing for or against certain players.
So, when deciding to create a multiple linear regression model to project player’s overall impact, I had to make a decision on what metrics to use. In this case, I decided to use a blend of two popular stats: Real Plus-Minus and VORP. Neither of these metrics is perfect (I prefer the former to the latter), and they both use different methods to their madness. To summarize, RPM uses advanced computing techniques to adjust a player’s plus minus for competition and quality of lineup for the opposing team and their own. VORP uses box-score statistics and some on-off blends to arrive at a value for a players performance when on and off the court. Similar results, but different ways to get there. RPM tends to pickup on more of the intangibles of a player, VORP players box-score stats. VORP thinks Russell Westbrook might be the best ever, RPM thinks he was barely a top-10 player. VORP thinks Ricky Rubio is barely a starter, RPM thinks he is a border line all-star.
While there are more examples of discrepancies between the statistics, by blending these two statistics together, I am attempting to grab both player intangible skills and their statistics, to try to weight both sides of the metrics evenly (when combining the stats, I normalized the two values so scale differences would not effect value).
When I created the model, I used a similar method. To build the model, I used a blend of per-36 minute box score stats and real plus minus. Because it’s a blend, players who are great in one stat, but not so much in the other (such as Joe Ingles, Andre Iguodala, and Pau Gasol, who are favored by RPM), are expected to take a larger step back that what probably is going to happen. But these cases are merely the exceptions to the rule. Another thing to note is that the model expects some of the best NBA players to regress to the NBA mean a bit, so some of the top players we see taking a step back, but still being overall great NBA players. A couple players expected to take a solid step up next season are Joel Embiid (if healthy), James Harden (RPM had him a lower than other metrics last season), Giannis Antetokounmpo. Kristaps Porzingis, and Andre Drummond, to name a few. Also, if you don’t know who Nikola Jokic is yet, please read up.
One more thing to note. These values are normalized, so the projection you see is a z-score for a distribution with mean = 0 and standard deviation = 1.
Well, here are the projections for the 2018 NBA season:
Excel file: s_pm