Nov 28, 2014; Atlanta, GA, USA; New Orleans Pelicans forward Anthony Davis (23) in action against the Atlanta Hawks in the fourth quarter at Philips Arena. The Hawks defeated the Pelicans 100-91. Mandatory Credit: Brett Davis-USA TODAY Sports
Freelance Friday is a project that lets us share our platform with the multitude of talented writers and basketball analysts who aren’t part of our regular staff of contributors. As part of that series we’re proud to present this guest post from Kevin Yeung. Kevin is a Toronto-based writer who contributes at the SB Nation network and Vantage Sports. You can follow him on Twitter at @KevinHFY.
I’ll start with a disclaimer: this article is not meant to test the degree of validity of RAPM (regularized adjusted plus/minus). Smarter math minds with a more thorough knowledge of the formula would argue either side of that a hell of a lot better than I could with my basic knowledge of its workings. I’m going to write this operating under one simpler principle: information can be gleaned from RAPM, and its statistical antecedent RPM.
And this is interesting to me because RAPM has been so divisive. Many in the statistical community have taken to stats based off RAPM, while many others among the greater basketball community have taken issue with it in some way. ESPN’s launch of Jeremias Engelmann’s RPM (real plus/minus) last year was a big win for the advancement of adjusted plus-minus models, but at the same time, there are many respected and smart NBA heads that reject them. And because RAPM is both a complex formula and a one-number output to measure overall on-court impact, it’s one of the hardest stats to take for granted.
For those who don’t know, the premise of RAPM follows the most basic rule of algebra: isolate your variable, then solve. APM (adjusted plus/minus), from which RAPM was constructed, does that by accounting for the other players on the floor. Then, RAPM added a ridge regression to tease away extreme examples generated by small samples. Engelmann’s RPM, derived from RAPM, adds further enhancements including “Bayesian priors, aging curves, score of the game and extensive out-of-sample testing,” according to the original ESPN article that introduced the stat. RPM is also what I’ll use for my case-in-point examples later on.
Stats and traditional basketball thinking often go hand in hand in 2014, which is unequivocally a great thing. But the hand-wringing over RAPM is a fascinating thing that elicits memories of old-fashioned ‘stats versus scouts’ conflicts of the past, because for all the work put into it by great analytic minds, so many people still consider it with a high degree of distrust.
You have to look no further than one of the NBA’s most skilled players for a case-in-point: last year’s Anthony Davis. Davis was at the center of a Twitter debate a few months back, related to the lower-than-expected rating that most adjusted plus-minus models assigned him. He had a RPM (ESPN) of 1.43, good for 98th in the league. That was a guy that was ranked 3rd on ESPN’s #NBArank this summer, and 33rd the year before.
It felt a lot like Davis was the poster boy for the ‘Plus-Minus is stupid’ movement. He’s too talented of a player for any stat that purports itself as a holistic representation of their impact to rank him so poorly. And absolutely no one will argue that Davis is less skilled than the immortal Steve Novak, who ranked eleven spots above him with a RPM of 1.75.
Allow me to go on a tangent here, and reference this old gem from Henry Abbott, where he points out Novak’s high defensive Synergy ratings in the 2012-13 season (RIP Synergy). Novak had the same defensive rating as then-teammate Tyson Chandler, and held an “elite” ranking in isolation defense. Which is weird, because Novak is slow, weak, and objectively, a car fire of a defender.
But Abbott’s point is that Synergy isn’t in the business of judging skill, but in the one of measuring efficiency. And as Novak proves, being a car fire of a defender can be pretty efficient. Because players get goaded into shot-jacking whenever they see a guy of Novak’s reputation (this often being a guy mindful of his own limitations and ceding the pull-up jumper) defending them, the percentages play out surprisingly well for defenses.
The point here isn’t that Novak is a good defender (he ranked 221st among 437 players in DRPM last season, so right near the median), but that no stat tracks skill. Skill often plays into efficiency, but stats can’t put a number on the ‘true’ ability of a player. They chart plays and crush them down into a single number of efficiency – which is very different from being a measure of talent.
Going back to Davis, RPM defined his impact last season as slightly positive on both offense and defense. There was no denying all the different things he could do on the court even back then, but the question wasn’t what he could do. It was how much that helped the team. RPM factor in box score information, but impact can’t solely be defined by that information (which is where PER failed) and so RPM goes further. Davis averaged 20.8 points on 51.9% shooting last season to go with 10.0 rebounds, 1.6 assists, 1.3 steals and a league-leading 2.8 blocks. That was good for a PER of 26.5, which ranked 4th in the NBA.
But did what Davis do help his team outscore the opponent? Looking deeper, Davis’ impact on the court might not have been as great as his individual production would’ve indicated. He allowed 48.8% shooting at the rim last year per SportVU – not bad, but closer to average than elite for a big man. For all of Davis’ impressive steal and block highlights, deeper film work shows that his defensive fundamentals were still raw as well. On offense, Davis was both prolific and efficient, but a lot of his offense came from midrange. For all the good that one player scoring well can do, that scoring being generated from midrange likely makes it hard for him to be the centerpiece of an efficient offense, when his man can still easily help off him and limit ancillary scoring. Last season, the Pelicans’ offensive rating was only 1.7 points greater with Davis on (105.4) than off (103.7).
Again, efficiency isn’t synonymous with skill, and that may be the greatest misconception with an all-in-one stat like RAPM. The final list it produces will never come close to resembling #NBArank or anything similar. But it’s illuminating in its own way, another source of empirical data that can tell us things we don’t know. Like maybe that Davis was a bit overrated last year, and specifically that we could’ve pegged him at slightly positive on both ends.
RAPM doesn’t answer questions. As a one-number stat that crunches a lot of information together, it lacks the nuance to explain much. But it’s a stat that can point in the right direction and ask worthwhile questions – often questions we may not have considered otherwise.
Last year’s Minnesota Timberwolves somehow ranked 12th in defensive rating, despite fielding a frontcourt of Kevin Love and Nikola Pekovic that many people saw as a defensive sieve. The two combined for less than one block per game, and there’s plenty of video evidence to show them allowing easy layups at the rim. But RPM gave positive defensive values to both Love and Pekovic. Love ranked 25th out of 83 power forwards with a 1.54 DRPM, while Pekovic ranked 25th of 64 centers with 2.19.
The why is up to us, but objective research can satisfy that. Love ranked 3rd in defensive rebounding percentage (29.5), and both bigs were a significant part of a defensive effort that led the league in limiting opponents’ free throw rate (only .177 free throw attempts per field goal attempt)[1. There’s also a height prior that jacks up players’ RAPM ratings for size.]. Then-coach Rick Adelman once decried the Minnesota bigs’ lack of fouling, most likely because of all the easy shots they ceded without a fight inside, but limiting free throw chances has its benefits too.
It’s findings like these that can rearrange our way of basketball thinking. Not everything follows conventional knowledge, and RAPM is one of the few stats that track by removing all the noise instead of directly tallying totals. If a stat like True Shooting Percentage tells us something about a player by what they do, RAPM tells us something about a player by subtracting what everyone else does. It’s a very different path to an answer.
Not every answer has been explainable. Corey Brewer held a positive ORPM rating, in fact one of the highest for a shooting guard (ranked 9th with 2.03). All the finishes off Love outlet passes in mind, Brewer is still only roughly league-average in True Shooting Percentage and a terribly inconsistent three-point shooter. Maybe he had a different way of keeping his man’s attention glued to him as a perimeter role player, but I haven’t seen anything to back that up.
Like with anything else, RPM isn’t perfect. Being a stat that doesn’t conclusively back itself up, it has to be taken with a grain of salt. Like other similarly holistic stats, such as offensive rating, it still can’t track the way players are used on the court – it only tracks their impact. Role and manner of usage matter, but can’t be boiled down to a number to include into a formula.
Nick Collison, Vince Carter, Chris Andersen, Patty Mills and Danny Green all rank in the top-50 of RPM, which is a pretty good sign of that they left a good on-court impact last season in the way they were deployed (i.e, Collison as a screener for Kevin Durant and Russell Westbrook, Carter as a spot-up specialist supporting Rick Carlisle’s endless menu of pick-and-rolls). But if you started a lineup of those five guys and expected them to dominate, the smart money says it’d fail spectacularly. The skill level for those five to stand on their own just isn’t there, and defenses would be able game-plan accordingly.
In this Seth Partnow/Nylon Calculus piece exploring the nuance that escapes math, Seth writes, “The underpinnings of the MoreyBall “layups and threes” strategy are unassailable: higher expected points per shot is better than lower. But what happens when the other team properly adjusts?” A similar principle applies here. RPM can paint a picture, but it won’t supply the context. The nuance of basketball strategy can’t be quantified by numbers, and there lies one of RAPM’s failings.
But I also suspect that’s one reason why RAPM has been so misunderstood. Nobody’s saying that Nick Collison and Patty Mills are top-50 NBA players unless they live and die by their RPM rating. But in their current role, Collison, Mills and the rest are providing significant on-court value to their teams.
And that’s where the value of these adjusted plus-minus models can be seen. Throw skill out the window, at least until you get to the ‘how.’ By cutting out as much noise as possible, RAPM leaves one variable, the player, and one number indicative of their impact. Applying basketball nuance from there is a very different process to an answer than starting from a player’s production and trying to fit it into a world of noise. It’s a lot easier to solve a puzzle given to you with a single piece missing than it is to solve one you’re given a single piece of.