[rquoter]BOSTON — Think about how cool (and overwhelming) it would be to able to answer almost any question you could imagine about the NBA. If you could find out, for instance, Kevin Durant’s shooting percentage from the left baseline on shots he takes after dribbling the ball with his left hand at least twice. We’re getting there, as was made clear during a presentation by Sandy Weil today at the MIT Sloan Sports Analytics Conference. Weil charted a whole bunch of stuff based on data generated from high-tech cameras, with three-dimensional capabilities, a company called STATS, LLC has installed at a handful of NBA arenas this season. The cameras capture an image 25 times per second and record every event on the court and the location of all the key elements involved — the 10 players, the ball, the referees, etc. The potential is enormous. A few key bullet points before I get into the meat of the findings: • Weil’s data was based on cameras installed at three arenas, but Brian Kopp, a vice-president at STATS, told me the cameras are currently in place at five arenas — San Antonio, Houston, Dallas, Oklahoma City and Golden State. The Warriors stand out there, since the other three clubs are known for their early and enthusiastic embrace of advanced stats and general geekiness. The Warriors have been known mostly for ugly intra-team disputes over the last few years. The fact that Golden State has signed up for the STATS cameras is a very good sign for Warriors fans that the team — and the new ownership group — really gets this. • The Celtics, another team that has embraced advanced metrics, is next in line to get the cameras, Kopp told me. • A quick aside: Somewhere between 15 and 17 NBA teams have representatives here at Sloan, depending on whom you talk to. That’s a few more teams than were here last season. But still: With the amount of publicity this conference gets now — Malcolm Gladwell, Jeff Van Gundy, Bill Simmons and Mark Cuban are among many headliners — it’s surprising to me that nearly half the NBA is not here. Some of that might be due to the economy, since the last three years have not been a time of hugely expanded hiring in any industry. But still. • Back to Weil’s work. He’s the guy whose work has (according to some) debunked the idea that there is such a thing as a “hot” shooter. He joked today that his initial analysis of all this new STATS data – nearly 7,000 charted shot attempts — hasn’t produced anything that would buck conventional wisdom. But I’m not sure that’s true, or that it even matters. Take this nugget: Kopp took me aside after the presentation and showed me data on the field-goal percentage of players who shoot after receiving passes from specific teammates. San Antonio players, for instance, shoot 60 percent after receiving passes from Tim Duncan. That’s a very high number, Kopp said, and it’s not entirely unexpected, given how much attention Duncan still draws in the post and on pick-and-rolls. But it’s a higher number than the comparable figure for most point guards. You know whose passes so far this season have also led to a 60 percent accuracy rate for teammates? Monta Ellis, according to what Kopp showed me. The same Monta Ellis most NBA die-hards critique as an inefficient gunner who doesn’t help his teammates all that much. Interesting, no? • The bulk of Weil’s work focused on how field-goal percentage changes when shooters are guarded closely. Not surprisingly, field-goal percentage drops the closer the primary defender gets to the shooter, and it drops even more the closer a second defender gets to the shooter. Overall, tight defense drops shooting percentage by about 12 percentage points. That’s huge. A tightly contested layup produces about the same amount of points per shot attempt as a wide-open 19-footer, Weil said. • The other finding that has major implications for your favorite team: Catch-and-shoot attempts are much more efficient than other types of shots when you control for distance and the presence of a defender. A player’s shooting percentage jumps significantly when the last thing he does before the a shot is the act of catching a pass — and not the act of dribbling. But if you catch a pass and hold the ball for about 2.25 seconds, whatever advantage you gained from catching the pass disappears. This makes sense, since holding the ball gives your defender a chance to catch up to you and prepare to defend your next move. Think about the argument this data is making against the sort of isolation plays that have made someone like Carmelo Anthony into a star. It doesn’t mean isolations have no place in basketball — not when there’s a clock that limits the time each team has before it must shoot the ball. But this does raise the possibility that this kind of data, as it becomes more accessible, is going to make isolations look even worse than we already know they are. • One cool thing: Tip-in tries are very low percentage shots. In Weil’s data set, only 22 percent of tip attempts went in. This would encompass all kind of tap-backs — the controlled, two-handed tap, and the desperate whack at the ball in mid-air. He suggests that it would be better to grab the rebound (if you can) and reset the offense. • Overall, shooting percentage on hook shots is about the same as on jumpers. I find this interesting, for some reason. • Defenders, it turns out, are very smart in the way they negotiate space. If an offensive player with the ball about 17 feet away retreats by a foot or two, the defender will follow, but not very aggressively. But if that offensive player steps from two-point range to three-point range, Weil’s data shows defenders pursue as closely as they can. • Another bit of conventional wisdom Weil’s work confirms: Teams shoot a better percentage on possessions that start with a defensive rebound or a forced turnover. But once you dig even deeper, you find that the higher shooting percentage on these possessions only really exists if the team shoots quickly — early in the shot clock, before a defense can get set. This would indicate that teams that try to force lots of turnovers, such as the Grizzlies and Celtics, might be making a smarter calculation than we realize. It also justifies the urgency with which a lot of point guards command their teammates get their butts moving after a defensive rebound. There is so much more to talk about here. We’re just scratching the surface, which is scary and exciting.[/rquoter] http://nba-point-forward.si.com/2011/03/04/boundless-potential-with-nbas-advanced-metrics/ http://twitter.com/#DraftExpress/status/43764285339471872
The modern technology is really fascinating and it's clear it's changing the way the game is coached (eg, Tibs' defensive stat-based schemes). I do wonder if this could be put to more beneficial use though. For example, why are refs still needed to decide if a player is out of bounds, or behind the three point line, or if a shot is gotten off in time? It would be trivial to put sensors in the ball, the hoop, the shoes, etc, or use this camera technology to detect these things. The less ref intervention, the better IMO.
Great read. Two thoughts jumped to me: 1. I also wonder why not more NBA teams are interested. 2. Catch and shoot being better than dribble and shoot might indicate why Adelman's offense is good.
I was referring to Adelman's offense being predicated on movement and passing to the open guy who ends up shooting immediately after catching the ball.
Steve Francis Disagrees. Wasn't this pretty obvious even back even in the JVG days? Maybe it's just because I'm a geek over this stuff, but it seems to me our coaching decisions have been heavily influenced by advanced metrics for a while now.
It wasn't made clear to me what these cameras actually do. My impression is that it tracks players/refs and captures the image at 24 images per second, but i'm not sure if it is up to an person to analyze the data or does it do that on its own. Does anyone have any additional insight on the cameras?
what would be more interesting but highly controversial would be the referee calls and how they can be manipulated. Example, everyone knows the makeup call, one ref calls a dubious turnover on one team, and the next possession, a very ticky tack foul is called to "make-up" for the foul call on the other team. What effect could you reduce the makup call if you immediately subbed the player off? would this reduce the makeup call %, what if you called a timeout, would this also give the refs time to cool off and also reduce makeup call %?
a lot of these stats are really only useful if you isolate them to each individual player, most players have strengths and weaknesses. well except Manu who i remember reading is about the same % at everything, no favourite side, catch and shoot, dribble left, dribble right, whatever, his %'s are basically the same for everything.
[rquoter] Players are tracked, with an exact location in x,y coordinates, as are referees and the ball. 25 times a second, software analyzes the video, and stores information about where everyone is and what is occurring. 1,000,000 entries per game with 60 some games added to their databases so far. Basic things like field goal attempts, blocks, rebounds, and assists are part of what is tracked. And that's where it starts getting really cool -- player position and defender proximity to a player are now available without having to chart and track a game by hand and eyeball estimates.[/rquoter] http://espn.go.com/blog/truehoop/post/_/id/25735/the-data-holy-grail I'm not sure if this means the rest of the stats are still charted by a person, though.
Sorry for the bump, but I just found out that some of the numbers that are tracked by those cameras actually get published on STATS Hoops' blog. http://sportvuhoops.stats.com Some interesting stuff there. A few things regarding the Rockets as well.
the number of variables that goes into a singular player's performance is huge. Put that player on a team, against X other teams and variance is incomprehensible. Anyone who thinks they can get meaningful results crunching numbers on basketball is seriously deceived. See stock market, and Daryl Morey for present day examples.
Anyone else think these advanced metrics are better for defenses than offenses...in terms of scouting opponents. Is there any advanced metrics regarding how successful/unsuccessful defenses are. Everything seems focused on the offensive stats versus anonymous defenders. That's why I say these stats help defenders probably more than offenses players; albeit, I can see why they help GM.
not that intersting to me. i'm not an anti stats guy, but if you watch enough basketball none of that was overwhelming except that you have exact percentages. for instance, catch and shoot, its obviously better than trying to create your own shot. steve francis never learned how to catch and shoot and it was part of his demise. i have stated that numerous times on this site
Completely agree. Anyone saying that collecting these stats is meaningless is foolishly ignoring this. I would say that it's possible that the main reason certain teams (Boston and Chicago come to mind) are as good as they are is because of the ability of their coaching staffs to interpret this data and form successful scouting reports and defensive schemes based off of it.
I'm sure there are several companies that do this by now but I know IBM had a very prominent Smart Planet campaign several years back that talked about Smart Grids for city planning and Smart Security. In essence this is the same technology - video data being tagged, tracked, and organized without the need for human input. I have seen some of those demos in person and it's truly amazing the way these cameras can pull logical physical data out of these moving images that we can then scrutinize however we see fit. Given some of the capabilities I have seen, I imagine hand charting is a thing of the past if teams are willing to spend money on the technology to do it for them.
