The Oklahoma City Thunder were one of 10 NBA teams to use SportsVU technology to gain an information advantage over their opponents. (Mike Ehrmann, Getty Images)

The future of advanced statistical analysis in hockey is happening right now. The only problem is that it’s happening in basketball.

There are two massive obstacles to the advancement of statistical analysis in hockey, in my opinion. One is the speed and complexity of the game, which makes it difficult to track and isolate specific events. A game like baseball lends itself well to advanced statistics because it is so easily parsed into individual events that can be quantified and analysed. In hockey, it’s extremely difficult to isolate events beyond the obvious, like faceoffs and shots. It’s one of the reasons why a lot of advanced statistical analysis revolves around attempted shots: they’re one of the few isolated, individual events that are available to be analysed.

The second obstacle is the accuracy of the data. Because statistics are compiled by human scorers at each rink, human error and bias enter into the picture, making it difficult to trust their accuracy. For instance, Colorado Avalanche scorekeepers tend to record significantly more shots on goal than league average, while the New Jersey Devils scorekeepers significantly undercount shots. Madison Square Garden scorekeepers consistently get shot location wrong, sometimes by as much as 20 feet, and NHL scorekeepers as a whole apparently can’t even decide where the faceoff dots are.

The technology for overcoming both of these obstacles already exists: it’s called SportsVU and it is currently invading the NBA. While it has a lot of potential to revolutionize basketball statistics, the potential for the technology in hockey is even greater.

SportsVU is based off what was originally military missile tracking software, but is now used to accurately track the location of players, recording their every move 25 times per second. It was originally applied to soccer in Europe, but when Stats brought it to North America, they moved to applying it to a more popular sport on this continent: basketball. Currently, 10 NBA teams are using SportsVU technology in their home arenas and making use of the data, including the Dallas Mavericks, Oklahoma City Thunder, and San Antonio Spurs.

In order to track player and ball movement, 6 cameras are installed in the arena. With just those 6 cameras and the high-powered SportsVU software, they can create a mountain of data. ESPN the Magazine did an article on SportsVU a year ago and highlighted how far down the rabbit hole goes:

Against Denver on Dec. 25, for example, [Kevin] Durant’s box score line read 44 points, seven boards and four assists. SportVU, meanwhile, detailed that the Oklahoma City small forward held the ball for a total of 2:51, averaging 2.3 seconds on his 75 touches. He was good for 0.6 points per touch, just up from his season mark of 0.5. He ran 2.8 miles in all, averaging 4.1 mph. But the really interesting discovery was that Durant dribbled 96 times, or 1.3 dribbles per touch, and that the more he put the ball on the floor the worse he shot: 55 percent with zero dribbles vs. 3 percent with six or more.

The technology can easily track a player’s most frequent shot locations and from where he’s most likely to score. It can track exactly where a player is most likely to turnover the ball, how much time he spends in specific areas of the court, who he passes to most frequently, and how likely a player is to score after receiving the ball from a specific teammate.

Really, that’s barely scratching the surface. But what gets me excited is imagining the possibilities when applied to hockey.

At a basic level, tracking puck possession would become far more precise, both on an individual level and on a team level. The shot location problem that plagues attempts at assessing shot quality, as the software would provide an exact location on the ice unaffected by scorer bias and human error. From there we can get into player and team shooting tendencies, not to mention better assessing how well a goaltender does on stopping shots from different locations.

Do you want to prove that a certain winger does have a tendency to stay on the perimeter instead of driving hard to the net? Do you want to know which defencemen are the best at moving the puck out of the defensive zone with possession? We know which players spend the most time on the ice, but what about the players that cover the most distance in a game? Want to know if your team is more likely to score if one player controls the puck more often on the powerplay than another player? Which defenceman is on the ice for the most completed passes by the opponent? Which defenceman is on the ice the most frequently when the puck is in the slot for the opposition?

These are all questions that could be answered by incorporating SportsVU technology into hockey. Sure, some of the answers might not end up being useful, but the ability to slice hockey up into individual events would be unprecedented. In addition, the accuracy of the data would be immensely improved. For teams, it would reduce the amount of time spent analysing video: instead of watching game film after game film to see what breakout a team uses the most often, what if you could just call it up on your laptop and immediately see the most frequent paths the puck and players take when heading out of the defensive zone?

Currently SportsVU is only available for soccer and basketball, as well as a limited application to football. It seems like it would be simple to adapt the technology for hockey, though I wonder if the small size of the puck might cause difficulties, as well as the fact that hockey players normally touch the puck with the stick rather than their feet or hands. Considering everything else the software can handle, however, that doesn’t seem like an issue that couldn’t be resolved.

Instead, I’m guessing the main issue will be money. Hockey just isn’t as big as basketball and football in the US, so Stats will target those two sports first with SportsVU. They’re avoiding baseball entirely, simply because there are a lot of companies doing similar work in baseball already, but that still means some time before they’re likely to put much time, money, and effort into targeting the hockey market.

With that said, a rich NHL team looking to gain any advantage possible might be able to convince them otherwise. Might a team like Vancouver, with its deep pockets and willingness to go outside the box(see: mind room), look to use SportsVU in the future?

Comments (22)

  1. Very interesting read, Daniel. This could definitely open up some fascinating lines of analysis in hockey stats.

    Though to your comment “There are two massive obstacles to the advancement of statistical analysis in hockey” I would add a third: the old boy network that results in management and coaches made up of “hockey guys” who have particular views of how the game should be played to be successful, a belief in the importance of “character,” etc. I am thinking for example of Brian Burke’s disdain for analytics (I know he’s not in management, but can you imagine trying to raise the idea of advanced stats with Don Cherry?!?).

    I think maybe one of the reasons Mike Gillis has employed some of the unconventional tactics he has is that he was a (relative) outsider to the hockey establishment, having come into management because of his experience as an agent rather than as a player. So he may, therefore, be more open to new ideas for player evaluation and development than some of his peers.

    • The “old boys’ network is alive and well in most other sports as well. It isn’t unique to hockey. What might sway the “character” guys about this technology is its ability to tell who actually backchecks (and how hard they backcheck) and who is floating.

      While yes someone like Burke (sigh) might resist more than other management types, eventually the benefits will become undeniable as teams that use advanced analytics begin to outclass those who do not.

  2. This would be revolutionary. The blogosphere’s favorite metrics, corsi and fenwick, are pretty poor, and certainly not comparable to the best stats in baseball and basketball, or even football. That’s mostly a paucity of data, like you said (though I think there probably are better metrics that could be studied if there hadn’t been this wasted effort).

    The “heat maps” that are widely available in European soccer are glorious, let me tell you. And hockey has much, much more data available than soccer (though not quite as much as basketball), so SportsUV would be even more useful. I cannot wait for this to take off

  3. While I don’t doubt the power of the tool and the data, my understanding is there are already several tools doing some of this on some teams. For example, Calgary has been pumping this PUCKS system Feaster has brought in that is supposed to do video review and event quantification work of some persuasion.

    Also, with a camera based system, I query how much the boards and nets would affect the ability to track the puck. I also wonder how the system would adapt to the uniquely hockey-esque challenge of on the fly substitutions.

    Very interesting article and I am sure we will all be watching to see if anyone does pick up this system.

    • I wonder if the system is so advanced it can tell individual players apart – either through number recognition or through learning each player’s individual physical characteristics.

      American football has unannounced substitutions (often just before the ball is snapped) which would require an extra level of recognition.

  4. It would be pretty epic from a player’s stand point.. Not that the player’s don’t already have each one of their shifts already que’d up on a flashdrive ready for their device for the plane ride, but this would also help in so many other ways when looking at opponents, etc etc..

    As a fan, I can’t wait for it to enter the NHL, or some form of Hockey somewhere!

  5. Interesting and somewhat timely article. I was just listening to The Basketball Jones Overdose podcast yesterday and they were talking about the SportsVU cameras.

    I think this also might be interesting in analyzing questionable hits and if there are trends by certain players and give the disciplinary body much more evidence to draw upon.

    I think it would also be interesting to see how much a specific defender can change the shot of an opposing player. Where would Ovechkin shoot from against Brian Lee as opposed to Zdeno Chara?

  6. Was wondering the other day if any of the NBA teams that use this also have a NHL team in their arena. Apparently the Caps do (and probably a few others). Apparently they’re quite expensive and new programs would probably have to be developed for Hockey, but it would give a huge edge to certain teams if they could, uh, borrow them.

    • That’s actually a really good point: the more NBA teams that adopt this technology, the more NHL teams will have it already implemented in their home arena. From there it’s a software problem rather than a hardware one.

  7. The problem of tracking a puck is vastly more difficult that tracking the larger soccer or basketball balls, in particular since many players use black tape that can hide it. You’d also need extremely high-speed cameras and correspondingly fast computers (to handle the extra frames) in order to accurately detect shots, where the puck is moving 100mph+, and looks like a blur on most video.

    It would be great, and I’m sure its coming, but hockey is a much harder game than basketball to adapt machine vision solutions to.

    • That’s definitely an issue, perhaps the issue. The puck moves quickly and is often obscured, so that’s a major technological hurdle to overcome. From what I’ve been reading about SportsVU technology, it’s not an insurmountable problem, and some articles claim that it would be relatively simple to adapt it to other sports, with hockey specifically mentioned.

      • Perhaps some kind of partnership with the NHL, in which they could fit each puck with a chip that would relay it’s location 25 times per second in sync with the cameras would help with this problem? Unfortunately, I’m not sure how feasible that would be.

  8. The speed of hockey would make this an epic chore for the code writers. In football, baseball and basketball and especialy soccer the shots or hits are for lack of a better term telegraphed. The fan and camera can tell well in advance who is shooting or kicking or passing the ball and from exactly what point on the playing surface. A vast amount of goals scored in hockey are from a deflected puck that requires the highest resolution HDTV in super slow motion to determine who the actual goal scorer was. Also with the puck having to cross completely over the goal line which is often also determined by the aforementioned tvs the real time value of this system might be in question. The NHL’s current systems value is in the real time aspects. As to the shot location in hockey that is a human error factor that will always plague it because of the real time on the fly entry system. The disparity in the number of shots is also a human error problem which will also be hard to over come as the shot statistician needs to determine from an elevated perspective wether the shot actually would have entered the net without goalie intervention. A puck shot at the net and caught by the goalie is not necessarily a shot if his catching or blocking glove is not in a direct line with the mouth of the net. If the puck is high or wide even by an inch it is not a shot but a missed shot. There would need to be a lot of camera technology specifically allocated just for this one stat on both sides of the rink. Oh and by the way the way if I’m paying several hundred dollars for my lower bowl seats I don’t want these cameras to obscure my view . This technology sounds great but for the cash poor NHL owners I think their current system while flawed in some ways gets the job done for the average fan and scout.

    • The system requires 6 cameras for the NBA, allowing precise tracking of all the players and the ball. The cameras are in the rafters and are quite small, so there’s no chance of them obscuring vision or distracting fans.

      The cameras capture 25 frames per second. I’m not sure if it would be enough to track minute changes in direction from a deflected puck, but I think it would be. I could be wrong, though.

      It should be able to accurately tell whether a shot is going wide or not rather than on net. If it takes just 6 cameras to do the type of analysis that they’re already doing in the NBA, it does seem like much more would be necessary for shot tracking.

      • Love this post… Great job Wag’s. I was one of the thousands of nerds who attended the MIT conference this past March, and it was super duper cool (glaven). Seriously though, the advancements in statistical analysis and correlating technologies is going to increase exponentially as these methods become more commonly utilized across different sporting platforms. I appreciate your reference to basketball and the scattered embrace of this type of tech across the NBA. It was entirely too obvious that the old boys club in the NHL front offices are drastically lagging behind in their appreciation for advances analytical methods and technologies. I attended all of the NHL related talks and I can assure you that many teams sent some sneaky personnel (all of which sat ‘covertly’ at the back with laptops open and no key strokes in sight). They are paying attention to advanced analytics, but I not convinced that most of them know what to look for.
        As for issues with tracking the puck, I believe that a safe step would be to hire a 7 year old Wayne Gretzky who infamously used to watch HNIC and trace puck movement on a piece of paper in order to identify puck path tendencies. Perhaps Wayne was onto something… As advanced as SportVU seems to be, and I’m sure that its capabilities far exceed its current usage within sports, I believe that all of these video based data collection system’s require human oversight, as a form of human (sober) second thought. Most of the technologies currently being utilized are three-part in nature: Video, Software, and Human. The human element is necessary for insuring that the data is correctly applied to what the video and software reads. Without it, the random nature of sport can lead to inaccuracies within the data collection process.

      • saw something like this in Calgary that’s being used for youth hockey. They use 12 robotic cameras inside the glass. The data is transferred to a memory stick right after the game and coaches can work with their team in a coaches room with 2 HD monitors. The coach has telestration capabilities and can show the same play from 2 different angles. Very cool and awesome teaching capabilities. From a company called InThePlay.

  9. I have no doubt it can track a large orange ball and indeed a puck for most of the time but scrums in front of the net may be are a different manner and in the case of a legaly shot puck, height is a critical factor. An overhead camera would have to have puck altitude determining capabilities which with today’s technology would probably require a puck chip or lower cameras. Fox tried the puck chip back in the infamous glowing puck coverage days and one of the many reasons they scraped it was because of the cost. Another reason that the puck tracking technology of fox tvs failed was that the sensors they used to track the puck which were elevated lost track of the puck behind the boards on the tv viewers perspective side. Basketball has far fewer blind spots than an NHL rink does. I’m not saying I wouldn’t want the NHL to give this a look but I’d rather see the money spent on player safety or lowering ticket prices.

  10. It’s quite evident that this writer does not understand the fast pace of the game of Hockey. He doesn’t know how many different stats that are recorded by the Off-Ice Scorers (12 events). Sometimes in a 10 second span you could have 8-10 different events to record. Each event has drop down screens for futher explainations. For instance a missed shot not only requires an approx. location, but what type of shot; slap, snap, wrist, wrap, backhand, tip and redirection. Then you need to add in the why it was missed;hit crossbar, hit goal post, wide, and over net. Who really cares if the missed shot was from 40 ft or 45 ft, thats why it’s an approximate location?

    As for face-offs you do not explain how “Phantom Locations” are determined. If its not right exactly on the DOT as you mention then you get you Phantom Location. So it sounds like 1 -2 feet off would creat this Phantom location. A little tweek to the current system could take care of this.

    Now try to adding in takeaways, giveaways, hits and block shots…. Your system would create more problems than the one used now.

    Good luck Mr. Know-it-all. What are you on commission for the company if you sell this software?

    • I have been doing this very work for the last two years. My focus has been the Montreal Canadiens. I have been using video to track each and every puck-possession play occurring in a game. I have this data.

      • Hey man,
        Some of these guys you are competing with (and I mean posters) have been doing this for an excess of 25 years. I myself as a tech savvy person has been up with these guys tracking the multitude of stats and the amount of REQUIRED human interaction would render this system moot. I’m just saying that it’s good to know who you’re going up against before you try to put on the big boy pants.

        • Thank for the advise. I’ll keep plugging away. How many of those big boys have been hired by professional or junior organizations? Or is it just me?

      • to add to my previous post, they DO use a multitude of cameras (about 5 angles) to track this puck visually and as the puck size is a factor, a lot of people forget that most of the uniforms/gear have an excess of black. This would make it a little more difficult to track without some sort of computer tech in the puck itself

Leave a Reply

Your email address will not be published. Required fields are marked *