Nylon Calculus: Deconstructing preseason NBA player rankings

Certain quotes and statements are staples at the turn of the NBA calendar every season. I’m in the best shape of my life. I’ve put on 10 pounds of muscle. I’m working on my 3-point range. (Well, I haven’t personally done any of those things, but you get the drift.)

The onset of this season was no different. Interspersed among all the optimism is usually a few mentions of motivation from the media, sometimes (often times?) from some sort of preseason ranking. Take it from Magic Johnson himself, in reference to the recently released Sports Illustrated Top 100 for the 2018-19 season:

"“Kuz is a different guy. Man, I mean he got mad because he wasn’t in the top hundred so I’m glad they didn’t put him in the top hundred so he can keep working hard. It’s that type of drive that we have with these players.”"

Preseason player rankings are a staple of any major media publication. Sports Illustrated, ESPN, Bleacher Report — everyone gets in on the act. Given the steady stream of player rankings pouring in, around this time last year, I launched this tweet into the universe:

Everyone tries to be as objective and as meticulous as possible (okay, maybe not everyone, strictly speaking) when it comes to doing the research on these rankings, but they’re inherently designed to be controversial, and that’s okay! Beauty is in the eyes of the beholder, as it were. Different evaluators value and incorporate various facets of the game differently, as do fans. But like I found last year, at least we should give some credit to SI for not being wildly oscillating year to year in their rankings –- that’s really one of the few things you can ask for with these types of exercises.

But this is Nylon Calculus after all, and I wanted to take these player rankings one step further (okay, several steps), using the SI rankings as my base. Deconstructing the Top 100 list actually reveals a lot about both the actual talent landscape of the NBA as well as the perception of such.

Breaking down the 2018 SI Top 100

This year’s edition of the Top 100 features a whopping 34 players who are 24 years old or younger, as well 32 players between the ages of 27 and 29, the ostensible prime of their careers. This says two things: not only are there a lot of really good players in their primes, but the NBA is also in great hands with the wealth of young talent coming up the ranks.

And that young talent is not simply bunched together at the bottom of the list either. Guys like Ben Simmons, Jayson Tatum, and Devin Booker are all in the top half of the rankings, with the reigning Rookie (?) of the Year the highest of them all, coming in at 26th. I’ve written at length praising Simmons’ abilities as a versatile defender, but Ben Golliver writes this regarding the prodigious Aussie:

"“He has a questionable motor? Simmons came out of the gate as a clear plus defender – ranking 35th league wide in Real Plus-Minus – while also carrying Philadelphia to eight straight wins without Joel Embiid to close the season.”"

Ben Simmons is a tank happening to play point guard, a dynamic playmaker who nevertheless fits none of the traditional notions of what the position should look like. Our minds look at Simmons and are wired to instinctively call him a power forward, but if his head coach Brett Brown even calls him a point guard, then why shouldn’t the rest of us? He’s one of the clearest windows yet into the position-less NBA.

But Simmons is just the appetizer. All three of the sub-24-year-olds in the Top 10 of the ranks are unicorns: Joel Embiid, Giannis Antetokounmpo, and Anthony Davis. Embiid, the 7-foot Defensive Player of the Year runner-up who can handle the ball and shoot threes. Giannis, a physical marvel who can kinda sorta literally cover the length of the court in a single bound. Davis, fresh off a Top 3 MVP voting finish who blots out opposing shots on one end of the floor and faces up over helpless defenders on the other. All three are poised to dominate the NBA for the foreseeable future, and all three display skillsets that defy traditional categorizations.

Davis and Giannis were already Sports Illustrated’s 5^th and 6^th ranked players while Embiid slid into 9^th in only his second actual season in the league, looking to upend the traditional LeBron James-dominated hierarchy.

And what a run for James it has been over the last 15 years. He’s the only player in the Top 10 to grab All-Defense, All-NBA, and All-Star honors while also winning both an MVP and a championship in the same season. Oh, and he did it for two consecutive seasons. While players like Embiid are only in the infancy of their careers, James showed little signs of slowing down in his thirties. Last year, he posted the highest Box Plus-Minus and VORP (Value Over Replacement Player, derived from BPM) since his 2012-13 season in Miami, which was arguably the best season of his career.

Still, aging is a natural process for most of us. As the NBA has refreshed its stockpile of talent the last few years, we’ve seen mainstays like Manu Ginobili, Tim Duncan, and Kobe Bryant all wave farewell. Whether it’s injuries, diminishing athleticism, or something else altogether, it is understandably difficult to stay productive into your late thirties. Only three players on the SI Top 100 this year were 34 years or older, and none of them were ranked higher than 87^th.

While on the clear back nine of their careers, Dirk Nowitzki, Pau Gasol, and Andre Iguodala all have enjoyed incredible amounts of success in the NBA. Dirk’s cumulative VORP ranks 18th all-time in the shot clock era, with Pau not too far behind at 25th and Iguodala, the least individually heralded of the trio, coming in at 43rd. (LeBron James, of course, is first, and with some distance.) While VORP is by no means an end-all-be-all statistic, it does speak to the sustained stretch of productivity and success that all three have been able to string together over the last couple decades, and highlights the intelligence and refined skill level that defines their games.

On the other hand, I also have a painting of Dirk hanging on my wall (shouts to the masterful Ryan Simpson), and would probably be waxing poetic about him even if he weren’t 18th all-time in career VORP.

Enter the swarm

Now, simply breaking down the actual SI rankings is one thing. But remember, at the top of this piece, I did mention that we would go multiple steps further. Most major and popular rankings are all done in some sort of aggregate fashion, whether by a group of voters, a small panel consensus, or some combination of multiple methods. Essentially, multiple viewpoints are coalescing and reaching some equilibrium understanding of how NBA players stack up relatively to each other. So if real human beings could do it, could we train a hundred “particles” to come to a group understanding in as close to the same way as humans?

The driver of this type of analysis is a technique called Particle Swarm Optimization. Swarm optimization is inspired by nature. A single ant does not by itself have the capability to construct a structure to navigate risky spaces. But a colony of ants, all communicating with themselves and optimizing their interactions with the group (see where this is heading?) can come together as something greater than the sum of their parts and achieve a task that would have been impossible for just one ant.

Similarly, PSO is an optimization technique that utilizes that enhanced group intelligence by initializing lots of “particles” and allowing them to each understand the movements and findings of the rest of the group in order to converge onto the best possible solution. Manny Perry of Corsica had a really good explainer on Swarm as he used it in hockey analysis (and which was partly the inspiration for this piece) a few years ago that helped me get started.

Envision a three-dimensional basketball problem (say, with points, rebounds, and assists). You’re trying to converge onto what weighted combo of those variables will get you the most accurate measures of any player’s overall ability. In practice, PSO looks a lot like this:

Like any optimization technique, Swarm works to minimize some custom defined loss or error function over a number of iterations.

We’re going to now expand the previous example to cover about 30-40 different variables (encompassing rate statistics, advanced metrics, and biographic info) and use Swarm to see if we can get a weighting of those variables (essentially, coefficients) that produces an overall ranking which is closest to the ranking from the Sports Illustrated list.

Overall, the swarm does a pretty decent job! (You can find the whole list of rankings here — don’t ask me why it’s in text format, there are no good explanations.) The final rankings produced by PSO had an average error of under 4 spots (including all the players who were unranked and honorable mentions) and were within 10 spots of the SI rankings if we exclude the players who were mutually unranked on both lists. For a technique that isn’t necessarily the most sophisticated or high-end model, that’s not bad at all.

Some of the blind spots are obvious. Players like DeMarcus Cousins who suffered debilitating injuries will be appropriately penalized by the real world rankings, but not by the algorithm. Additionally, the real rankings are done with an eye on the future, whereas Swarm is using data only from the previous season (I tried checking if multiple seasons of recent data improved performance — it, unfortunately, did not). Swarm also naturally disagreed heavily on players who derived most of their value from defense (like Rudy Gobert) or whose impact is not easily captured by box statistics (like Draymond Green, whose stat lines barely register at times, but who is an indisputable lynchpin for the Warriors).

Like Manny states in his own explainer, using PSO here is essentially like teaching to the test. We are training these particles to learn how the experts rank players, not necessarily what the objective rankings of those players should be, a concept that has many intriguing offshoots. Our swarm learned that players who do more and who do more things well tend to be valued highly. The top five players predicted by PSO are James Harden, LeBron James, Anthony Davis, Giannis, and Russell Westbrook (in that order), all stat sheet stuffers. Incidentally, that does mean that the swarm nailed the top 3 in MVP voting from last season (hmmm…).

It’s even more interesting to then take what the swarm has learned about how experts ranked players from last season and apply it to other prior seasons. For example, how about the 2010-11 season, when Derrick Rose won MVP? What does PSO tell us about that season? The predicted top 10 were, in order: LeBron James, Dwight Howard, Dwyane Wade, Derrick Rose, Kevin Durant, Pau Gasol, LaMarcus Aldridge, Chris Paul, Kobe Bryant, and Russell Westbrook. Some familiar faces in there, to be certain (full results here).

A few things immediately stick out. Our swarm seems to agree that Rose’s MVP was a weak decision at best, likely the product of LeBron fatigue, and should have gone to either James or Howard that year. And let’s not overlook the spice that is Kobe getting ranked 9^th and well behind his teammate Pau Gasol. Hit us up, NBA Desktop. In some corners, this is borderline malpractice, but from other perspectives, it continues along the fraught path of re-examining Kobe’s career in relation to Shaq and Gasol.

But applying our PSO coefficients to the 2010-11 season also reveals something fundamental about how the NBA has changed since then, specifically about how the value of different positions has evolved.

With the presence of roving trees such as Tim Duncan, the landscape back then was dominated far more by big men. Even though there are a large number of incredibly skilled bigs in the league today, a natural byproduct of a pace-and-space league is that guards and wings – the Splash Brothers of the world – will continue to rise in value. Players like Anthony Davis and Joel Embiid have demonstrated with great intent that the center position is most definitely not dead, but that doesn’t mean that team construction principles haven’t evolved past the traditional understandings of the center position.

Sometimes that evolution manifests itself in Nikola Jokic running point for the Denver Nuggets. Other times, it’s manifested as Erik Spoelstra starting a lineup with the 6-foot-7 Justise Winslow at center. Now that should put a smile on Magic Johnson’s face, Kyle Kuzma offseason workout videos aside.