Building a super team

May 30, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30, right) is congratulated by Oklahoma City Thunder forward Kevin Durant (35) after game seven of the Western conference finals of the NBA Playoffs at Oracle Arena. The Warriors defeated the Thunder 96-88. Mandatory Credit: Kyle Terada-USA TODAY Sports
May 30, 2016; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30, right) is congratulated by Oklahoma City Thunder forward Kevin Durant (35) after game seven of the Western conference finals of the NBA Playoffs at Oracle Arena. The Warriors defeated the Thunder 96-88. Mandatory Credit: Kyle Terada-USA TODAY Sports /
facebooktwitterreddit

Klay Thompson recently told reporters, “Yeah there’s only one ball. And we’re gonna put it through the hoop.”

Here we are, in the dystopian future, where it still feels surreal that Durant has joined the Golden State Warriors. Three of the most sublime scorers in the NBA, all in the same lineup, joined by the league’s most versatile players, a lineup where Andre Iguodala is nominally the “worst” player.

Naturally this got people thinking: is there anyone who can compete? And this led to a great thought experiment game over at the Ringer. Using their defined cost system for players, is there any possible way to build a super team that can compete with the Warriors while staying under $15? And this led me down a slightly tangential path: can I build a team modeled after the Warriors megadeath lineup under those same cost constraints?

More from Nylon Calculus

Using Basketball Reference, with an assist from Draft Express, I compiled a list of the five players from the Warriors lineup and 115 other players in the NBA (subjectively the top 115 players in the league), with each of those players representing a multidimensional data point. Those dimensions broke out to 15 different statistical features for each player (with all volume stats being per 36 minutes): 2PT attempts, 3PT attempts, free throws, offensive rebounds, defensive rebounds, assists, steals, blocks, points, 2PT%, 3PT%, eFG%, FT%, defensive box plus-minus, usage percentage. Of course, you can use any statistics you want, but I went with those 16 to provide a good balance of playing style and effectiveness. Of course, the stats are going to naturally bias more towards offense as well, but I tried to get in some defense too.

The Warriors players were my “original” points and the other 115 players were the “comp” points. Now with the data all in one place, the first test was going to be fairly simple. Let’s find another 5-man lineup out of the “comp” players which is statistically the closest to the Warriors lineup (the shortest total “distance” away, so to speak). No cost controls or any tweaks. The result of this first experiment was: James Harden, J.J. Reddick, Patrick Beverly, LeBron James, Al Horford.

On first glance, that seems alright, but something felt off. The megadeath lineup is predicated on shooting, excellent individual (and team) defenders, and great passing from unconventional spots. Just as well, their ability to switch everything on defense is naturally rooted in their size and quickness advantage. Everyone needs to generally be able to defend everyone. I could still do better.

So I went in and weighted the statistics that really defined the Warriors lineup: assists, points, 3PT%, steals, blocks, and defensive box plus-minus. If a “comp” player was worse than an “original” player in one of those six categories, that distance/error would be weighted heavier and thus penalized. If they were better, the error would be weighted down and thus rewarded. I then also added in a 16th statistical feature for comparison – standing reach, to try and approximate the size and length of the players as well.

Playing the Game

Now the Ringer game was ready to be played, for real. After running the program, the results were compiled into radar charts, with the most similar players as picked by the lowest weighted distance. The Warriors player is represented in a filled in yellow shade, while the comp players are represented as red outlines. Basically, when looking at the radar charts, how well does the outline approximate the shaded player?

With no cost controls, the resulting lineup was: James Harden, Kawhi Leonard, Patrick Beverley, LeBron James, and Paul Millsap.  

I’d take this team to battle. I would put them up against the Warriors and let them run. You have a great small ball forward and the Atlanta Hawks’ best player in Millsap, the Defensive Player of the Year in Kawhi, a scoring machine in James Harden, a great individual defender in Beverly, and oh yeah – LeBron James, the best player of this generation. Everyone on this team has range, they can run up and down the court, and they can all (with the exception of Harden) defend well. In fact, Millsap had a DBPM of 4.2 (!) last season, which was one of the highest scores in that category among all players and even higher than Draymond Green.

Unfortunately, by the Ringer’s cost system, this team also costs $22, which is $7 over our allowed limit. So back to the drawing board. The other quirk: Beverley is way smaller than the player to whom he was most closely compared, Andre Iguodala, whose standing reach is 7.5” longer than Beverley. There’s no chance of Beverley then being able to guard some of the players Iguodala excels against. So I did two things at this junction: to further account for size, I added in a condition that all the result “comp” players’ standing reach should be within 3” of the Warriors player to whom they’re being compared, and then I cobbled together a version of gradient descent to optimize the team under the cost constraint. I wanted the lineup which was the closest statistically to the Warriors lineup while still staying within budget.

This then gave us our final answer to the Ringer game, coming in right at $15: James Harden, J.J. Reddick, Trevor Ariza, Paul George, and Paul Millsap. I wouldn’t say that this team is as good as the non cost-controlled version, but they’d most definitely still compete against the Dubs.

Trevor Ariza and J.J. Redick were impressively effective approximations of Iguodala and Klay, respectively. Those two comparisons had the two lowest errors (were the closest to their respective players) in the resultant 5-man unit. We also swapped Paul George in for LeBron, which was initially surprising, but made sense on further review. The statistical difference between LeBron and George compared against Kevin Durant wasn’t as large as it was for another player versus Harden or Millsap when comparing against Curry and Green, respectively.  And taking a look at the radar chart, George actually still does a fairly good approximation of Durant. Paul George is basically a cost-effective LeBron James.

James Harden stayed put throughout all iterations, which is understandable. Even though his defense is lacking, Harden might be the only player not already on the Warriors who can approach Curry’s level of scoring mastery, while being a great facilitator to boot. I thought I would see Chris Paul sneak in, but his relatively low shooting volume likely did him in. Regardless, we should pour one out for anyone trying to compare to Steph; Harden was by far Steph’s closest comp, and his approximation also still had by far the most error of anyone else within the result lineup.

The Implications

While this was just a thought experiment game on the surface, a more sophisticated version of this model could be crucial to actualizing team building philosophies. We always hear that it’s a copycat league, but now by playing the Ringer game, we just put that to practice by trying to find a team that could be like the Warriors. If a small market team wanted to build itself in the image of the Spurs, a similar model could be helpful for identifying players who would fit into that image. In free agency, who should a team target that would mesh with their ideology and style? Where might there be market inefficiencies (for example, Nic Batum being only rated at $1 in this game’s cost structure)? Once we got to the $1 range, we basically are forced to make a tradeoff between DBPM and 3P%, but some players who might still fit that 3-and-D model effectively were Korver, Marvin Williams, and Nic Batum. Credit to Rich Cho for doing so well in accumulating versatile, two-way wings without breaking the bank.

There’s nothing quite like the impending sense of inevitability in sports, and one thing appears clear: there’s no team in the NBA as presently constructed which at least on paper looks to be able to dispel the Warriors aura as the actual embodiment of that inevitability. But that doesn’t mean there aren’t going to be challengers. Who’s ready and going to stop them from putting the ball through the hoop, as Klay Thompson would be inclined to say?

Next: Introducing Dredge: A play-by-play metric