Predicting Team USA’s Performance and the Pursuit of Lineup Fit

With the 2016 Olympic Men’s Basketball Tournament underway, and Team USA off to the expected 2-0 start, ESPN has created a tool to construct a Dream Team starting lineup.

“Who should we start” is always a fun game, especially heading into a tournament where Team USA is likely to be relatively unchallenged. This is the closest we’re likely to see to a genuine fantasy basketball superteam (with all due respect to the Durant-added Warriors) take the court together, so finding the best possibly grouping is an interesting exercise.

Unfortunately, this lineup rater makes some assumptions that render the outputs somewhat meaningless.  First is requiring the standard G-G-F-F-C lineup, which is a an outdated model by a few decades and one that’s not relevant in a league where positions are more fluid than strict like baseball. Second, and more worrisome, the “score” for a given lineup is simply the sum of the individual RPM’s. You cannot rate a team simply by adding up every player’s metric score — this is true for every well-known metric, from PER to RAPM — because there are real, significant issues with fit, most of which are magnified on Dream Teams.

Here’s the basic issue with translating NBA success to Dream Team success: there’s only one ball, and a huge swath of the players chosen derive a significant portion of their value from either shooting or handling the ball. Thus, super teams created on a “sort by metric ‘X’ basis” are often less imposing than they appear. That is a natural, sensible consequence when you jam so many scorers onto a team when, say, Kobe Bryant is the fourth leading scorer.

The Corrections

The basic methodology here is using a player’s BPM in the NBA in the adjacent seasons — e.g. for a 2012 player, use his BPM in seasons 2012 and 2013 — then adjust it and calculate the team’s predicted rating much as Nathan Walker has done quite successfully in past NBA season win predictions for usage rates, but with an additional adjustment for assist rate.

The second step is linking a team’s BPM and their performance in the Olympics via the team rating SRS, which is based on point differential. For a greater breadth of data, I included the World Championships too using my own team rating numbers available here, (which, I believe, are unique in the public domain, as basketball-reference does not include the World Championships.) With a limited data set, one should be wary with conclusions, but the assist and usage adjustments did appear to add significant information. However, the results weren’t perfect, and I noticed a trend: teams with fewer three-point shooters seemed to under-perform, and vice versa.

In fact, the regression models I ran weighed “spacing” (i.e. three-pointers attempted per minute in the NBA, adjusted for league averages) to an absurd degree. This is probably an artifact of a certain period in Team USA history. Three of the bottom four teams in three-pointers attempted were the extremely disappointing 2002, 2004, and 2006 teams. This should not shock people who follow Team USA religiously, or similar tournaments, because three-point shooting, due in part to the shorter line, is extremely important.

With some trial and error and Bayesian analysis, I constructed a more stable model for predicting Team USA performance, but I must caution people from taking these numbers as gospel: it’s still an experimental model built on limited data.

Predicted SRS = 9.7 +1.01*BPM -0.76*BPM adjustment +25.3*Team avg adj. 3PTA per MP

(The last variable there looks maddening, but it’s only looking at the average three-pointers taken per minute by five players adjusted for league averages.)

For a sense of scale, some BPM adjustments are as high as eight or nine points before the coefficient is applied. The 2008 (Redeem Dream) Team had the highest adjustment, but their raw ratings were so high it ultimately didn’t deter them. The 2004 squad (the Bronze Team), however, had the third largest BPM adjustment, but had one of the lower raw BPM scores and the second lowest three-point rate. They were the only US team with an adjusted point differential lower than several other nations in a tournament, so in retrospect the disappointing performance was unsurprising given the players selected.

The Results

You can see the results for all these teams plus the projected 2016 using their numbers from last season and an aging curve since every other team uses stats from the following season. For minutes, I just grabbed their averages from the exhibition games, which are just a rough guide of who will play — but it’s good enough for now. Team USA squads spread their minutes evenly in the real games too. Since the method I used was dealing with limited data, the ratings are grouped together pretty closely, but that might say more about the chaotic results from a short tournament than anything else.

Table: Team USA ratings

Keep in mind that the age curve I used for 2016 may actually be underrating the set of young stars — they’ll be better than most think, even without a few MVPs.

This year’s team appears to be one of the strongest, save for the two famous Dream Teams from the 90’s and the two redemption teams during LeBron James’ heyday (and yes, the 1992 team is still on top.) While the 2016 roster lacks staggering firepower, they score well on fit — their BPM adjustment doesn’t sink them and they have a lot of outside shooting, which historically has been a boon to these tournament teams. In fact, this goes back to the 90’s, when a 1994 team, devoid of the star-power of the first Dream Team, destroyed its competition — they were loaded with shooters, from Reggie Miller and Mark Price to Dan Majerle and Joe Dumars.

The recipe for success for constructing a Team USA roster isn’t just loading up on highly rated players, as many of their skills become redundant offensively. Instead the team should find a balanced roster where the stars derive their value from multiple avenues — a one-dimensional scorer is worth a lot less on a stacked team — and stock the court with outside shooting. Even the first Dream Team is a good example because many of their stars were great defenders too and not just scorers.

It’s tough to explain how significant these effects are since these rosters don’t exist in the NBA, unless you’re the Warriors. Their new Mega-Death lineup consists entirely of Team USA members with Stephen Curry (2010, 2014), Klay Thompson (2014, 2016), Andre Iguodala (2010, 2012), Kevin Durant (2010, 2012, 2016), and Draymond Green (2016). (Of course, their old Death lineup consisted entirely of Team USA members as well, for those paying attention, with Harrison Barnes. They essentially swapped out one Olympian for another, so what’s the big deal?).

But we all remember the basketball experiments of LeBron James and friends in Miami and Cleveland. You can force a team into greatness with raw power, but there are diminishing returns and it won’t be as seamless as one would think. The Miami-LeBron team, for example, was a bit of a disappointment at first until the team realized they needed to focus on shooting and defense — even LeBron worked on his outside shot. And those effects are tough to show without the living basketball experiment that is Team USA, throwing superstars together in a five-on-five game with only one ball.

The “ideal” Team USA lineup, by the way, according to ESPN is Chris Paul, Stephen Curry, Kawhi Leonard, LeBron James, and Draymond Green. Their collective RPM is 43.9, but that massive number overstates something at a fundamental basketball level. The usage/assist adjustment for that lineup a whopping 11.6, and while there are still issues with applying an adjustment like that — there are other factors to adjust, like rebounding, and some players thrive in a reduced shooting role while others sink — the greater sin is pretending it doesn’t exist.