NBA Dollars & Sense - Part III: Player value by position

This is the third piece in the multi-part series of posts, Dollars & Sense. This model is the second of a few models that will be combined to create the final NBA Dollars & Sense model. Part III takes players, groups them by position, and determines their values relative to other players at their position.

When the game of basketball was first invented, positions were never specified. The positions simply evolved around the differing styles and size of players. The names of the positions have not changed, but the differing styles and sizes of players in each of those positions has changed. Most centers used to be banned from shooting the ball outside of ten feet. Most of the scoring throughout the beginning of the history of basketball was done by larger players. Before 1979, no one could score three points without being fouled while making a shot. All of those things have since changed. Point guards, shooting guards, small forwards, power forwards, and centers now all have different responsibilities on the court, some of which heavily overlap.

However, just because player styles blend between positions, it does not mean that players aren’t judged and compared in terms of position. Depending upon a player’s position, he could be valued at a much higher value than an equivalent player of a different position. The problem with this idea is that it is hard to tell how much better or worse a player is than his position average. It’s even difficult to simply identify an “average” point guard, or center, or any position for that matter. Those are the problems this model tries to solve.

More from Nylon Calculus

By taking the player data referenced in Part II, this model groups each player by the position at which they spent most of their minutes this past season. Once each group has been made, a regression, involving the same advanced statistics as Part II, is run for each position. The resulting values are, like Part II, each player’s value in terms of percentages of the salary cap.

After each player’s estimated value is determined, the mean value of each position group is recorded. Using this mean, the model finds a sort of deviation for each player, respective to their position grouping. This deviation represents the number of deviations that a player’s value is from the average of their position. For example, if 2015-16 Nic Batum’s deviation value is 1.512 and the mean of the small forward grouping is 7.40 percent, then we can assume that Batum’s estimated value is 7.40% + (1.512 * 7.40%) = 18.59% of the salary cap. As if the basketball world needs another acronym for a stat, this does happen to be a nifty one:

(Estimated Salary – Position Salary Mean) / Position Salary Mean = DIMES

That’s right, DIMES or Deviation In Mean Expected Salary. The more (or less) DIMES a player has, the better (or worse) of a player he is relative to his position. One important caveat with this estimation is that any players that project a value below the league minimum are simply listed as minimum value players. That would be why the lower bound of players cleanly cuts off at about -1 DIMES.

Position Averages:

PG Average Cap%: 6.62%

SG Average Cap%: 7.07%

SF Average Cap%: 7.40%

PF Average Cap%: 8.45%

C Average Cap%: 9.15%

Looking at the Tableau Viz and the position averages above, a few interesting players stand out from last season:

LeBron James

Yup, you guessed it. LeBron is still head and shoulders above the average small forward, and almost any player, in the league. Looking at the figure, LeBron is certainly the most dominant, and most valuable player by his position. He leads the league with 3.406 DIMES, equaling an Estimated Cap% Value of 32.6%, not far off from his Cap% for next season of 32.9%.

Giannis Antetokounmpo

Giannis is an odd case. He’s really a small forward moonlighting as a point guard. Given the way the Bucks used him in the second half of last season, along with the proposed plan to have him handle the ball more often during the upcoming season, Giannis is listed as a point guard in the model. This probably hurts his value a little bit, given that his skill set lines up a little bit more with the qualities that the model finds valuable for small forwards as opposed to those valuable to PGs. Also, he had a limited amount of time in the point guard role, so his percentage and efficiency stats may not be at the level of other elite point guards.

Michael Beasley

The model really likes the former Kansas State star. In fact, the model thinks he’s worth just over 11% of the cap. Like Giannis, Beasley may be feeling the effects of a small sample size, except in the positive direction. Over the 20 games he played, he averaged about 13 points, 5 rebounds, and 1 assist per game. However, given that he only played a mere 18 minutes per game, the model does recognize Beasley’s impressive production in such a small span (even if he did average 20 FGA/36 min).

Along with these three, there are a few other oddities from the model. For every LeBron that the model nearly hits on the head, there are always a few other estimated values (Justise Winslow, Allen Crabbe) that seem to stray from perceived value. Take a look, play around, and see what values with which you agree or do not agree.

Home/Nylon Calculus

NBA Dollars & Sense – Part III: Player value by position

By Max Schimanski | Sep 23, 2016