Pages Navigation Menu

Assessing The New Toys: A Primer On The Uses and Lessons of SportVU Player Tracking Data

USA Today Sports

USA Today Sports

We are proud to feature this guest post from Seth Partnow. Seth is the creative mind behind the blog, Where Offense Happens, and can be found on Twitter, @WhrOffnsHppns.

In this two part series…special thanks to Daryl B. of BackBoardBlues for help with the charts and parsing the data. He’s a fantastic follow on twitter @BBallport for new and innovative ways to model and visualize NBA statistics in ways which make these complex concepts instantly digestible.

With the advent of the SportVU system in all 30 NBA arenas, the NBA has entered the realm of big data. Even before Kirk Goldsberry lit up the basketball blogosphere with his early attempt at a SportVU-based “one number” approach to player value, with its mix of points banal (Chris Paul is the most effective player in the league with the ball in his hands) to intentionally incendiary (Ricky Rubio is the worst!) teams and fans have been trying to make heads or tails of how to interpret the data gleaned from transforming X’s and O’s into X’s and Y’s.

The numbers the league has made available as “Player Tracking Data” can be extremely frustrating. Numerically-minded fans know what SportVU is, so they are desperate to get their hands on information, any information, which can possibly be gleaned from the system. In this two-part series, we hope to shed some light on how to best use the publicly available data. This post will focus on the how the data is already being used as well as the newly available, box score level tracking data, while part two will examine some of the larger evaluation work that we and others have done by analyzing these stats in concert with other, more traditional metrics to shed new light on various aspect of the game.


With their access to the full range of SportVU data and small armies (very small in some cases) of analysts and interns, the teams themselves are constantly seeking new interpretations of this information. We have to take it as an article of faith that insights gleaned are actually filtering through to the coaches and players, as aside from one well-received (excerpt in the hometown papers) feature on Toronto’s use of the technology last season, front offices are being extremely tight-lipped on the subject.

Understandably so, since the ability to properly parse and translate the information caught on camera into actionable information for personnel decisions and game planning purposes represents a huge opportunity for competitive advantage between teams. Nobody who knows anything really wants to give anything away. Any front office holding the key to unlocking the deepest secrets of basketball certainly isn’t going to turn it over to the league for public consumption. However, even the tiny sliver of data presented on can provide some new understanding to fans, despite Goldsberry’s closing speculation that “the days of bedroom analytics might be numbered.” (To be fair to him, Goldsberry was referring to the computational power required to parse the full data set rather than making a tired blogger-in-mom’s basement joke.)

Some of the information presented is actually fairly straight forward. For example, the Catch and Shoot and Pull-up tabs break down accuracy on these common shot types by players around the league. Nothing groundbreaking, but useful to have a centralized, sortable resource to see who is taking the most pull-up jumpers, and who is truly the most deadly on catch-and-shoot 3’s (Hint: You might have read about him recently elsewhere on Hickory High).

But, with no guide on how to navigate and interpret this information, much of the data presented on amounts is open to misinterpretation and misuse.  What exactly are we to take from a player’s average speed during a game? Doesn’t that tell us more about the player’s role and his team’s style than anything intrinsic to the player? How is a thinking fan to relate what he or she sees during a game to the rather opaque columns of numbers?


Luckily for fans, recently begun providing additional data from the cameras by including a “Player Tracking” tab in every box score. This box score presents some (but not all) of the information collected under the aggregate “Player Tracking” section of the website. The box score player tracking also presents some new information not previously available.

First, a quick caveat – at least as of the first implementation shortly before the All-Star break, the real-time updating of the tracking data in the box score is extremely inaccurate.  For starters, some of the stats which are included are inaccurate as initially presented, sometimes wildly so. For example in the February 13th Thunder/Lakers game, the last game heading into All-Star Weekend, the division of shot attempts between “contested” and “uncontested” changed at some point from the totals tracked during the same to the updated totals now visible. Further, Steve Blake was credited with 18 “hockey assists” during and immediately after the game.  This was hard to believe, as the league leader in the category, Chris Paul, averages around 2.2 secondary assists per game:


Finally, some information included in the “final” box score is omitted from the in-game box score. The defensive rim protection columns (more on this below) do not appear until the box score updates at some point overnight:


With luck, these issues are limited to the roll-out of the new box score, but if not, real-time applicability of the stats will be extremely limited.

Also unfortunately, some of the information presented is of little analytic value.  For example the “distance covered” and “average speed” stats are largely a function of playing time and roles.  Wings tend to run further during a game than bigs, and the greater “average speed” of these players simply captures they are running (as opposed to sliding and shuffling) more often in the half-court than are post players. Plotting distances run against possessions played essentially demonstrates as much — distance run basically measures the number of times a player runs up and down the court in an average game. Without further information it borders on pure speculation to make pronouncements about a player’s level of effort or lack thereof based on this information. Perhaps if two players playing similar roles have vastly different distances traveled and average speeds, questions can be asked as to why this happened, but in general these differences will probably be explainable as vagaries of a particular game.

Other information is, on its own, largely descriptive as opposed to holding much use in terms of evaluating whether a player played well or poorly. Without reference to other information, the number of touches and passes executed by each player falls into this category. A high disparity between number of touches and passes indicates the player shot or turned the ball over a lot, but there is no need for extensive player tracking technology to catch this, as traditional box score stats already account for such “possession ending” actions. As we’ll discuss in part 2, perhaps there is some insight to be gleaned by comparing these numbers with the player or team’s season-level pass and touch numbers. For example if a player seems to have been unusually involved offensively (or alternatively seems to have been frozen out), this might show up in terms of the individual game. An obvious example from Thunder/Lakers is Kevin Durant, who was credited with 95 touches, well above his season average of 67.5 per game. Though he did play a few more minutes than average, the disparity is much more attributable to the man who normally handles the bulk of Oklahoma City’s ball handling duties, Russell Westbrook (averaging just under 80 per game). While this is a fairly banal example, the touches stat could demonstrate a player simply not being involved in the game for whatever reason.

Much was made at the time of J.R. Smith taking only one shot in a December 13, Knicks loss to Boston, as if he was responding to complaints about his gunning by simply shutting down on offense. The player tracking data supports this to a degree, as he only had 31 touches, whereas his season average is a hair under 55 per game. Though Smith played about 6 minutes less than his season average in that contest, that alone would not explain a reduction in over 40% of the amount he handled the ball. There is certainly some game-to-game variation in the amount of times any player touches the ball but this player tracking stat can help confirm the eyetest if a player appears to be especially involved or particularly marginalized in a given game.

Continuing to move through tracking box score, secondary assists are tabulated. Long wished-for by those who wanted a more full accounting of playmaking ability, the secondary or “hockey” assist attempts to capture plays where a player makes the pass that directly leads to the pass for the score. From this Lakers-Thunder contest, Kendall Marshall recorded one hockey assist in addition to the 17 ‘traditional’ assists with which he was credited. This is a textbook example of what people have wished to capture when wishing “secondary” assists were tracked in the past:


Out of a high pick-and-roll, Marshall’s early pass to Kaman forces Kevin Durant to help off Wes Johnson in the corner. Kaman immediately hits the open man to get credit for the assist himself, but without Marshall’s quick recognition and on-target pass, neither Kaman nor Johnson would likely have benefited from this opportunity.

Similarly, “free throw assists” are being tracked as well. These are plays where the passer would have received an assist if the recipient had scored, but was fouled in the act of shooting instead and proceeds to make at least one of the ensuing free throws. The Lakers were credited with one such play, with Marshall pitching the ball back to Ryan Kelly who is fouled on a catch-and-shoot three point attempt:


However, a note of caution, as with much of the publicly released SportVU data, the tracking and tabulation of this category is still imperfect. The above was the only credited free throw assist during the game, but nothing was credited for this play:


Kaman hits Jordan Hill on the duck-in, and Hill is hacked while going up to shoot immediately.

The existence of some degree of error in terms of recording stats via player tracking technology should not be terribly surprising. It is an open secret that traditional stats are imperfect especially with respect to those requiring judgment calls such as assists, blocks and who receives individual credit for steals and rebounds. The “hometown assist” is a real thing, and many players have wide splits between their home and road production in these stats. In fact a cynical explanation for why “assist chances” are not included in the per game tracking box score is to save local scorers from the potential embarrassment of a player ending up with more assists credited than chances generated. Thankfully, that did not occur in the Lakers-Thunder matchup, as Marshall, for example had a faintly ludicrous sounding 35 assist chances on which he recorded his 17 helpers. (More on where this number came from in part 2).

Moving on, the next few columns present some fairly interesting data. One of the big buzz words around modern NBA defense is “rim protection” and with player tracking, this can finally be examined in detail. This will be discussed in much greater detail in part 2 (also here), but the “Defensive Impact” tab on the season long player tracking is actually capturing two crucial pieces of information: the commonly cited “opponents’ field goal percentage at the rim” but the roughly equally important number of shots at the rim the player contests. Thankfully this information is now included in the box score. With the addition of a little bit of background information, this data can illuminate a fair amount about the game that just happened.

Over the course of the season, one of the Lakers’ major downfalls has been allowing teams too many attempts at the rim (33.4 attempts per game, highest of any team in the league) and not doing well enough to contest those shots when they occur (teams shoot 60% on those attempts vs. the Lakers, 23rd in the league). Part of this is attributable to poor perimeter defense, which broadly speaking leads to the high number of attempts, and partially to the fact they tend to play smaller lineups with Shawne Williams and Wes Johnson manning the power forward,while regular centers Pau Gasol and Hill are average-at-best rim protectors. With Gasol out, they actually do a better job protecting the paint, as Chris Kaman is very much underrated as a rim protector both in terms of holding opponents shooting percentage down (40.7% at the rim through the All-Star break) and contesting a high number of shots (over 11 per 36 minutes played, whereas leaguewide big men average just over 8 contests at the rim per 36).

Well in this particular contest the Thunder were able to do a better job finishing versus Kaman than have other teams over the course of the season, scoring on 6 of 9 attempts. For example, here Durant is able to use his combination of skill and length to finish over, around and past Kaman’s contest:


On the whole, while 19 of the 26 shots at the rim the Thunder attempted were contested (73%, right around average), they were still able to finish 12 of them (63.2%, well over the league average of around 51% on contested shots at the rim). While it’s dangerous to ascribe everything to a narrative, the tale here might be that though the Lakers rotated well, the dearth of talent on such an injured squad showed through.

On the other end of the floor, the Thunder contested 23 of the Lakers’ 28 attempts at the rim, with L.A. shooting a slightly above expected 13 (56.5%). Serge Ibaka was particularly effective in this regard, contesting 11 of these shots. In fact over the course of the season, Ibaka has been one of the most valuable rim protectors in the entire league, on par with Andrew Bogut both of whom trail presumptive Defensive Player of the Year Roy Hibbert by a decent margin. Though effectively defending shots at the rim doesn’t always mean actually blocking the shot, all 5 of the misses forced by Ibaka where in fact blocks such as this one:


Continuing on, we can see rebound chances. Considering Oklahoma City out-rebounded the Lakers by a small 44-42 margin, it is potentially interesting to note the vast disparity in rebound chances. In particular, the Thunder’s large edge in attacking the offensive glass is probably indicative of the differenes in strategy and personnel. Though Ibaka is a more than capable perimeter shooter, he tends to play much close to the basket on offense than do any of Williams, Kelly or Johnson, who manned the power forward spot for L.A. for much of this game. Another factoid of note is that Durant grabbed 12 defensive rebounds on only 13 chances. Which isn’t actually that out of the ordinary as he is one of the more efficient rebounders in the entire league.

And finally we get to the look at the shooting breakdowns. While simply looking at contested vs. uncontested data obscures some information (for example, “contested” 3 pointers and attempts near the rim are almost always more efficient shots than uncontested midrange shots, though be aware that the data used in the linked piece uses a somewhat different definition of “contesting” a shot than does SportVU. As will be examined more fully in part 2, the lack of a Z axis measuring vertical distance from the floor is a limiting factor in analyzing the SportVU data), uncontested shots are axiomatically better shots. Leaguewide data supports this intuition as well:


So while it’s not perfect, looking at which team did a better job creating more uncontested looks will tell a great deal about the game. Alternatively, the oft-repeated description of the NBA as a “Make or Miss League” can be demonstrated here. Sometimes the offense purrs along creating open shot after corner three. And then the team can’t throw the ball in the ocean and loses.

Watching Lakers/Thunder for the first time, the impression was that LA was getting good shots, which they just couldn’t hit down the stretch. Shawne Williams in particular missed some great shots. Examining the tracking box score, he indeed went 2-8 on uncontested shots! In the last 8 minutes, he had 4 great looks at corner 3s and missed them all.

While it’s uncertain how many of these would be scored as “uncontested” in the SportVU tracking, it remains a useful tool to deciphering whether  something was wrong the offense, or the team just stopped making shots.  In this case, it was the latter from the Lakers standpoint.


So that’s a primer on how to use and interpret the new box score information available on Part 2 (available next week!) will go much further in depth on such topics as rim protection; a more holistic approach to measuring offensive involvement than “usage rate” taking account for playmaking; as well as a look potential applications in terms of rebounding, dribble drives and what the information on touches and time of possession can tell us about how various teams play.

  • Dodgson

    Great article! I look forward to more analysis going forward.

%d bloggers like this: