New technology and statistics will change the way we understand basketball, even if they also create friction between coaches and front-office personnel trying to integrate new concepts into on-court play. The most important innovation in the NBA in recent years is a camera-tracking system, known as SportVU, that records every movement on the floor and spits it back at its front-office keepers as a byzantine series of geometric coordinates. Fifteen NBA teams have purchased the cameras, which cost about $100,000 per year, from STATS LLC; turning those X-Y coordinates into useful data is the main challenge those teams face.1
The 15 teams, once again: New York, Orlando, Boston, Washington, Milwaukee, Toronto, Philadelphia, Cleveland, Dallas, Oklahoma City, Minnesota, Golden State, Houston, San Antonio, and Phoenix.
Some teams are just starting with the cameras, while others that bought them right away are far ahead and asking very interesting questions. Those 15 teams have been very secretive in revealing how they’ve used the data, but one team that has made serious progress — the Toronto Raptors — opened up the black box in a series of meetings this month with Grantland.
The future of the NBA, at least in one place, looks like this:
That’s Jason Kidd hitting a 3-pointer off a Carmelo Anthony pick-and-roll in the first quarter of Toronto’s February 22 home win over the Knicks; the Knicks are in blue, passing the little yellow ball around, and the Toronto players are colored white. It looks simple, but the process of getting there took a bunch of people, including three Toronto front-office employees, more than a half-decade of work. In simple terms: The Raptors’ analytics team wrote insanely complex code that turned all those X-Y coordinates from every second of every recorded game into playable video files. The code can recognize everything — when a pick-and-roll occurred, where it occurred, whether the pick actually hit a defender, and the position of all 10 players on the floor as the play unfolded. The team also factored in the individual skill set of every NBA player, so the program understands that Chris Paul is much more dangerous from midrange than Rajon Rondo, and that Roy Hibbert is taller than Al Horford.2
There is much, much more, including a massive set of statistics on team tendencies I’ll address in a future piece.
That last bit — the ability to recognize individual player skills — is crucial for the juiciest bit of what the Raptors have accomplished: those clear circles that sort of follow the Toronto players around and have the same jersey numbers. Those are ghost players, and they are doing what Toronto’s coaching staff and analytics team believe the players should have done on this play — and on every other Toronto play the cameras have recorded.3 The system has factored in Toronto’s actual scheme and the expected point value of every possession as play evolves.4 The team could use that expected value system to build an “ideal” NBA defense irrespective of the Toronto scheme, but doing so today would be pointless, since part of the team’s job is to sell a sometimes skeptical coaching staff on the value of all these new numbers and computer programs, says Alex Rucker, the Raptors’ director of analytics.
This includes a whopping 140,000 logged plays, about half of which are pick-and-rolls.
In simple terms: The expected value of a possession when Kyle Korver is taking an open corner 3 is much higher than when Josh Smith is taking a contested 20-footer.
“You need that coaching perspective,” Rucker says. “But we are still looking for where the rules are wrong — areas where there are systemic things that are wrong with what we do on the court. But any system needs to comply with what the coaches want, and what the players can do.”
One early finding: The ghost players are consistently more aggressive on help defense than the real Toronto players. Check out DeMar DeRozan’s ghost (no. 10) as Raymond Felton (no. 2) and Tyson Chandler (no. 6) run a pick-and-roll on the right wing during the first stage of this same New York possession:
Ghost DeRozan is in the middle of the paint to deter any pass to Chandler, while Ghost Landry Fields (no. 2) has zoned up between Anthony (no. 7) and Kidd (no. 5) on the weak side. Fields is actually closer to Kidd, in part because the program understands the higher expected value of a corner 3. That’s why Ghost Rudy Gay (no. 22) has moved a bit toward Fields’s man, even as the real versions of both players stick much more closely to their original assignments.
The Knicks eventually reset and swing the ball to Anthony for a pick-and-roll with Chandler. Look again at Ghost DeRozan vs. Real DeRozan as Chandler rolls to the hoop:
Ghost DeRozan has left Kidd to slam into Chandler much earlier, and much higher on the floor, than Real DeRozan actually did. Since Anthony is dribbling toward the right side of the floor, the left side becomes the “weak” side, and it is the duty of the weakside defender in almost every NBA defense to help on the big man (Chandler) rolling to the paint. DeRozan does that here, but he does it too late; doing so earlier would stop Chandler sooner and allow both DeRozan and Jonas Valanciunas (no. 17) to retreat more easily to their original marks.
The ghosts in Toronto’s ideal defense are almost always more aggressive helpers than real players, and that’s true across the league, according to the analytics team and Toronto’s coaching staff. Teams either haven’t realized they should be sending even more help toward the middle and the strong side, and sending that help sooner, or they haven’t fully convinced players to behave in this way.
“There’s a clear deterrent value in having guys in the sight line of the guy with the ball,” Rucker says. “They don’t even have to do anything. If they’re just in there, they’ll make a difference.”
Attaching numbers to things like the placement of a defensive player represents a major advancement in sports statistics.
“Anybody who is going to pooh-pooh this kind of analysis will say things like, ‘You can’t measure defense, because it’s all about the guy who doesn’t help or rotate,'” says Keith Boyarsky, the team’s technical director of analytics. “That it’s about what you can’t measure. But that’s exactly what we’re measuring.”
It can be tough, say coaches, to sell players on drifting so far from their own assignment. “Guys don’t want to be embarrassed, or see themselves on TV giving up a dunk or an open 3,” says Micah Nori, a longtime Raptors assistant who has worked closely with Rucker’s team. “Even though basketball is essentially five guys guarding the ball, it’s hard to get players away from the concept of, ‘This is my guy.'”
Having players a foot or two out of position can be fatal to a defense, says Tom Sterner, a Raptors assistant and something of a tech guru.5 “The players are just so quick in the NBA,” Sterner says. “One or two feet can make a huge difference.”
Sterner, who began his NBA career as a video coordinator for the Magic 23 years ago, has a master’s degree in sports administration and computers, and was once chair of the NBA’s technology committee.
Ultra-aggressive help defense is really hard work. Replay that clip and watch how far DeRozan’s ghost has to move as the Knicks swing the ball. That’s brutal, and it’s not a coincidence that the only team that consistently mirrors the help defense of its ghosts is Miami, Rucker says. The Heat have three of the best wing defenders in the league in Shane Battier, LeBron James, and Dwyane Wade, and the latter two are among the NBA’s most gifted pure athletes. James can mimic DeRozan’s hyperactive ghost in a way no other player can, Rucker says. “LeBron basically messes up the system and the ghosts,” Rucker says. “He does things that are just unsustainable for most players.”
Nori offers another reason for widespread overcautious help defense: Players are worried about committing a defensive three-seconds violation. But Nori says they shouldn’t be. “Have you seen a game where they’ve called more than three?” he asks. “Even if they call three, if I can get away with 15, I should be in the paint. Everything you hear from the league is about flow and pace and not slowing the game down with technical fouls.”
Here’s another example of what the Raptors’ analytics team describes as underhelping — a George Hill–David West pick-and-roll on the left side of the floor from a Pacers-Raptors game this season:
Two things happen here, and both are representative of mistakes defenses make all the time, Rucker and his team say. First: Amir Johnson (no. 15) gets himself way out of position relative to his ghost (also No. 15) by chasing Hill toward the middle of the floor:
Johnson is very good at chasing ball handlers beyond the 3-point line, but he didn’t have to go so far out here, since Hill dribbles away from the basket and is thus making himself a non-threat.
The program logs Johnson for a breakdown, and that breakdown leads to another mistake from Gay (no. 22):
West (no. 21) now has the ball, and with Johnson so far away, the ghost team has recognized a crisis the real players failed to see. The ghost of Valanciunas (no. 17) has abandoned Roy Hibbert (no. 55, on the left block) to run out at West, and Ghost Valanciunas can do that with confidence, because he knows Ghost Gay will leave Paul George (no. 24) in order to crash down on Hibbert in the paint. But neither player moves, and West casually drains a midranger.
Just a reminder: The movement of the ghosts is based mostly on the expected value of a possession, calculated multiple times per second, as that possession moves along; Ghost Gay is the rotator in part because the program understands that leaving George open is less damaging than DeRozan (no. 10) leaving a shooter in the corner. The ghosts are not moving randomly, or even just in the way the team’s coaches would like. They are moving in a way the Raptors believe specific defensive players should move to thwart specific plays executed by specific offensive players with wildly different skill sets.
It is a very impressive piece of work. “Most teams are using spreadsheets or just using our reports,” says Brian Kopp, executive vice-president at STATS. “The Raptors go a step beyond that, which only a few teams are doing, and their visualizations are the best I’ve seen.”6
Kopp hasn’t seen everything teams are doing, since teams are secretive even in dealing with STATS.
That does not mean it has been an easy sell to the team’s coaching staff, though Sterner and Nori are enthusiastic about analytics and have helped craft the ghost defense. Everyone likes the ghost system, but some of the larger analytics-related issues have caused friction between the front office and some of the coaches — even if everyone involved is mostly polite about it. “It’s always going to be a challenge,” says Ed Stefanski, Toronto’s executive vice-president of basketball operations. “A lot of high-level coaches have come out against analytics, but it’s the wave of the future, and you’ve got to jump on.”
Bryan Colangelo, the Raptors’ GM, had already set Toronto on the SportVU path before hiring Stefanski in the fall of 2011, and Stefanski credited Colangelo with pushing the Raptors in the right direction.
The coaches, even the most receptive ones, seem to view analytics and SportVU mostly as a tool to confirm what they already think and know. Some samples:
Dwane Casey, Toronto’s head coach: “It’s a good backup for what your eyes see.” Casey added, “It may also shed light on something else,” a sentiment both Nori and Sterner echoed at points. “But you can’t make all your decisions based on it, and it can’t measure heart, and chemistry, and personality.”
Sterner: “It helps reinforce your gut. Most of the time, your gut is pretty much right.”
Nori: “More than anything, it’s a tool to help confirm what your eyes see.”
The analytics team agrees that most of the new knowledge will be along the margins — that coaches leaguewide get most of the big, systematic things right — but that the analytics will nonetheless offer more in the way of new discoveries that might contradict what we think we know. “A lot of coaches will say how great it is that analytics confirm what they already see,” Boyarsky says. “The fact of the matter is, that’s not really true.”
An example: The analytics team is unanimous, and rather emphatic, that every team should shoot more 3s — including the Raptors and even the Rockets, who are on pace to break the NBA record for most 3-point attempts in a season.
Take this possession from the same Raptors-Pacers game (I’ve removed the ghost players to make it easier to visualize):
The analytics team would have liked Gay (No. 22), a below-average career 3-point shooter really struggling this season, to jack up a contested 3 at this moment — with about six seconds left on the shot clock:
Gay instead whips a pass to DeRozan, an even worse three-point shooter, in the corner, and DeRozan tries an ultra-difficult drive-and-dish to a cutting Gay that George Hill (No. 3) deflects out of bounds with less than a second left on the shot clock. (The analytics team would have also been fine with DeRozan shooting a three there, and with Andrea Bargnani, No. 7, shooting earlier in the possession.)
For Rucker and his team, this is a question that gets at the value of particular shots, the impact of the shot clock, and how coaches teach players. “When you ask coaches what’s better between a 28 percent 3-point shot and a 42 percent midrange shot, they’ll say the 42 percent shot,” Rucker says. “And that’s objectively false. It’s wrong. If LeBron James just jacked a 3 on every single possession, that’d be an exceptionally good offense. That’s a conversation we’ve had with our coaching staff, and let’s just say they don’t support that approach.”
The coaches aren’t even close to being onboard with such a 3-happy philosophy yet. “To have guys who shoot 3s that can’t break that 35 percent break-even point, you have to really evaluate that,” Sterner says.
“You can shoot as many 3s as you’d like,” Casey says, “but if you don’t make them, that philosophy goes out the window. There’s always going to be disagreements. Analytics might give you a number, but you can’t live by that number.”7
Nori points out Toronto is shooting more 3s this season, and credits that in part to the influence of analytics.
Casey is obviously right that DeRozan is a bad 3-point shooter. But the analytics team argues that even sub–35 percent 3-point shooters should jack more 3s,8 and that coaches should probably spend more time turning below-average 3-point shooters into something close to average ones.
Paul Millsap is an example of a player who Rucker, Boyarsky, and Eric Khoury, the team’s analytics engineer, believe should have free rein to attempt 3s.
“Player development and coaching are scarce resources,” Rucker says. “You only have so much practice time. At a very basic level, a guy going from 25 percent to 30 percent from 3-point range is far more meaningful than a guy improving from 35 percent to 40 percent from midrange.”
The divide touches both on player development and on the degree to which the limitations of personnel should guide how a team plays. In very general terms, coaches tend to see personnel as paramount and limiting, while the analytics guys — from an upstairs office, of course — are more comfortable allowing players to stretch themselves toward more mathematically sound decisions. In other words: Paul Millsap is 30-of-109 career from 3-point range? Who cares! Let him shoot, and shift some of his practice time toward a shot that doesn’t conventionally fit his skill set and build.
Valanciunas, of course, has been the chief focus of Toronto’s player development staff this season, and that has been another source of tension between the analytics team and the coaching staff. Valanciunas, like most rookies, misses rotations, overhelps, and commits other sins of positioning on defense. Coaches hate that stuff, and they’ve often nailed Valanciunas to the bench in crunch time in favor of Aaron Gray — a fundamentally sound player who lacks NBA athleticism.
The numbers in large part disagree with that tactic, at least as it relates to Valanciunas’s defense. The Raptors’ defense has been better with Valanciunas on the floor.9 More importantly, the visualization data shows that Valanciunas is active and athletic enough to make up for all his defensive mistakes, Rucker and his team say.
The offense has actually been worse, to a much larger degree, per NBA.com.
“With Jonas — yeah, he’s making mistakes,” Boyarsky says. “But who cares?”
Casey said he hasn’t had deep discussions with the analytics team about Valanciunas, but Sterner has, and he agreed it’s sometimes a thorny issue of valuing culture over results. “You want your defense to be sound,” Sterner says. “Even though the production might be better, you still want [Valanciunas] doing the right thing.
But let’s not exaggerate: This isn’t Moneyball, with people at each other’s throats and folks threatening to quit their jobs. It’s not even close to that, actually, and that’s in part because the SportVU data do something most smart NBA people have been doing for a long time: combine video (the “eye test”) with advanced statistics. Understanding sports has never been about one or the other; it’s about both, and the cameras represent the most advanced actualization of that marriage. The coaches in Toronto helped the analytics team build the ghost system, and the analytics team sends the coaching staff regular e-mails with advanced numbers on upcoming opponents — e-mails the coaches read.
Coaches have also started using “points per possession” in conversations with players, and even tentatively passing along some tidbits from the cameras. The players think it’s all sort of geeky — Kyle Lowry cackled when I asked DeRozan about them — but teams are going to learn all sorts of new things from this data, and teams that exploit that knowledge on the court will gain some advantages. Finding synergy between coaches and stats guys is a big piece of snagging those advantages, and that’s going to be a tough process in some places. But the process has clear value.
“We’re still just in the developmental phase,” Rucker says of translating stats for coaches. “And things are much better than they were four years ago.”
10 Things I Like and Don’t Like
1. The Efficiency and Guile of Monta Ellis
Ellis has been on fire since the All-Star break, and especially since the J.J. Redick trade. He’s picking his shots more carefully, leaning more on his (very good) passing skills around the basket, and going crazy in fourth quarters. He has also gradually added a bit of clever misdirection to his game, occasionally faking as if he’s going to participate in some common Milwaukee action, only to suddenly dart someplace else for the “real” play.
Here’s a nice example:
Ellis starts by pretending as if he and LARRY SANDERS! are going to jog over from the elbows to the left side of the floor in tandem for a quick-hitting pick-and-roll there — a classic Bucks set dating to John Salmons’s Milwaukee heyday, and probably earlier. But as the defense bites, Ellis veers back to the right side for a pick-and-pop with Ersan Ilyasova that completely wrong-foots the Wizards’ defense. Fun stuff.
2. The Watchability of the New Magic
The Magic are bad, but they are surprisingly watchable for an NBA junkie curious about some intriguing and positionally ambiguous young parts. Tobias Harris has done well in extended minutes, combining a power game with a decent 3-point stroke, and Moe Harkless looks like he’s going to be a nice two-way slasher. Pairing that duo with Al Harrington in a funky and smallish front line, with Harrington playing center, makes for some enjoyable (and risky) ball, and replacing Harrington with a traditional center in Nikola Vucevic only ups the “interesting young guy” quotient. Give these guys a watch.
3. The Non-Replaying of Controversial Plays
Late in Miami’s blowout of Toronto on Sunday, Dwyane Wade drew a blocking foul while rising up for what would have been a monster highlight dunk and colliding with John Lucas III near the basket. The crowd went nuts, claiming Wade committed a charge, but large sections of the arena only had so-so views of a play that occurred in a jumble. Was Lucas in the restricted area? Was he moving? We all waited for a replay. None came.
This is a pet peeve: In-arena staff should replay every controversial bang-bang call. Failing to do so is a disservice to fans.
4. “Nonchalant” As a Verb
Nonchalant became my favorite English-language word the moment in my teenage years when I heard a baseball play-by-play guy refer to a second baseman “nonchalanting” the ball over to first base with a soft throw.
So I grinned when Denver’s play-by-play guy, Chris Marlowe, got on Tayshaun Prince for “nonchalanting” a fast-break layup during Denver’s exciting Friday win over Memphis, leaving the ball on the front of the rim. Is this translatable to non-sports usage? Can someone “nonchalant” a job interview, or even a date?
5. Markieff Morris Posting Up
The bigger Morris hasn’t shown as much improvement, on either end, as the Suns likely hoped to see in season no. 2, and his post-up chances have been especially inefficient. Morris is shooting just 31 percent out of post-ups, per Synergy Sports, and giving him the ball there usually results in a tough, fading midrange jumper. Defenses understand that, and so don’t send much help toward Morris on the block, eliminating the possibility of a kick-out pass to an open shooter.
Morris probably still projects as a role-playing stretch power forward, but in a lost season, it’s fine for Phoenix to experiment with him. This particular experiment hasn’t yielded useful results — yet.
6. Luke Ridnour’s Misdirection Plays
I’m a sucker for any set that plays upon a defense’s tendency to spot the early part of a typical NBA action and assume it will unfold in the usual way. Ridnour has some guile in his game, and he fools teams often on plays like this:
The play looks like a classic “flex” action, with Ridnour first setting a screen for Nikola Pekovic under the rim. Ninety percent of the time, he’d follow that up by taking Derrick Williams’s screen near the foul line and popping out for a jumper at the top of the key or a pick-and-roll. But Ridnour knows Golden State is expecting that, so he cuts off the usual pattern and curls around to the baseline for an open jumper. Fun, fun, fun.
7. Jeff Teague’s Ball-Watching
Let me be clear: I’ve really enjoyed watching Teague’s jump this season, especially in the last six weeks, and he’s going to be a very good player for whichever team signs him this summer. But he’s still finding the balance on defense between gambling for steals/watching the ball and marking his own guy. Teague’s steals are an important generator of turnovers and easy points for an Atlanta team that has ranked near the top of the league in forcing turnovers and doesn’t generate easy points via free throws. But smart teams know they can back-door Teague when he focuses too greedily on the ball. Teague will find that balance eventually; he’s a smart and very talented player.
8. The Nuggets, Standing Out of Bounds
A few weeks ago, George Karl laughed when I noted that the Nuggets often generate spacing by having a guy stand out of bounds. “Hey,” Karl said, “sometimes the other team actually guards you out of bounds.”
See if you can spot Andre Miller’s involvement on this play as Miller’s man, Kendall Marshall, watches JaVale McGee coast in for a dunk:
9. The Selective Physicality of J.J. Barea
Barea is a brave guy in some respects, but he’s also a flopper who only breaks out his trademark faux-physicality when it’s convenient for him — usually when it might draw an offensive foul. But there are also times when Barea has a physical assignment — sliding in from the weak side to jam an open roll man in the lane, for instance — and fails to perform it:
10. The Stunted Distribution Skills of Kemba Walker
Walker has upped his scoring efficiency, a very encouraging sign in Charlotte. But his assist rate hasn’t increased at all over his rookie mark, and his passing game, especially out of the pick-and-roll, has sort of stalled out. Walker has a tendency to either pull up right away for an open jumper or drive hard into the lane, where he can get out of control and trap himself in tight spaces. The in-between game — the hesitation dribbles, the smart manipulation of space — still eludes him, as it does many young point guards.
Walker is the victim of a lot of mitigating circumstances here. Charlotte has huge problems spacing the floor, especially when both Ben Gordon and Gerald Henderson are on the bench, and Walker is working with what has to be one of the very worst offensive front lines in NBA history. There isn’t even an average high-volume finishing threat among Bismack Biyombo, Brendan Haywood, Byron Mullens, Jeff Adrien, Tyrus Thomas, Josh McRoberts, and DeSagana Diop. Good luck racking up dimes when you’re dropping the ball off to those guys. Charlotte needs a big in the worst way.
This article has been updated to correct information on a strategic action.