For anyone who follows quantitative sports analysis, player tracking cameras are not news. Along with the NBA, soccer teams use them (even in the MLS) and rugby teams use them. They give x-y-z coordinates for each player at a high frame rate, which can be processed into a variety of statistics. Many think that this approach will revolutionize sports analysis. I stumbled across an article at ESPN today spreading this view to the masses.
Tracking data can help with many things, but it won’t save analysts from themselves. Here’s a point-counterpoint from the article linked above.
Point: “Paul Pierce averaged 4.5 assists this season, which is pretty good for a scoring wing. But that number doesn’t tell the whole story. According to SportVU, Pierce’s teammates shot a higher percentage after his passes than any other player in the NBA. This shows Pierce is passing at the right time — he’s giving his teammates mostly layups and open shots.”
Counterpoint: Pierce might be making great passes, but it’s just as likely that Pierce plays with better than average shooters or better than average cutters/floor spacers, or that Pierce commands a strong defender so other defenders are weaker, or that the Celtics run good set plays through Pierce. Correlation does not mean causation! There are tons of omitted variables other than Pierce’s skill that could explain Pierce’s passing “effectiveness.”
Point: “Nikola Pekovic’s breakout season was largely helped by Ricky Rubio. Pekovic made 76 percent of his field goals off Rubio passes, compared to 56.4 percent overall.”
Counterpoint: Again, there’s no way to know for sure that Rubio is the key causal factor here. Maybe Pekovic is a great finisher at the hoop or excels at the pick and roll — plays where Rubio is the likely passer — but he has to take some less favorable shots in other circumstances when the offense breaks down and he gets a pass from someone else.
Point: “The NBA-wide shooting percentage is significantly higher when the shooter doesn’t take any dribbles. This confirms what any basketball observer suspected: ball movement equals offensive success.”
Counterpoint: This one is just silly. I’m going to go out on a limb and guess that shots taken off the dribble tend to occur on possessions where no one is open, and shots off the pass tend to occur on possessions where someone comes free. In other words, defensive position (actual and anticipated) is omitted from the analysis. On top of that, you probably don’t get to the line much by shooting off of a pass, and shooting off the dribble forces defenses to collapse and respect penetration.
Point: “Throughout the history of basketball, the players considered the best rebounders were the players who averaged the most rebounds per game. But that doesn’t tell the whole story. What if there’s another elite rebounder on a player’s team hogging all the rebounds? Or what if a guy plays for a bad defensive team that doesn’t produce as many missed shots? SportVU allows teams to deeply analyze rebounding by generating never-before-seen stats such as rebounding chances (described as when a player is within 3.5 feet of the ball and, yes, that measurement is exact) and rebounding in traffic (when opponents are within that 3.5 foot circle).”
Counterpoint: This misses so many important aspects of rebounding. One is right there in the paragraph: “What if there’s another elite rebounder on a player’s team . . . ?” Looking at “rebounding chances” won’t adjust for that, since the elite rebounder can still take the rebounds, even when you’re close. It doesn’t measure how often players get themselves within the 3.5 foot circle or tell us how often they should get there, either (they shouldn’t run to the ball if they can help more by boxing out, getting back on defense, or leaking out for a fast break). It also doesn’t reward players who are excellent at boxing out but not great leapers, who still help their team rebound, especially if they can tie up a good rebounder on the other team (in fact, this is often a conscious strategy).
Point: “It’s tough to analyze passers from team-to-team because two players can be in such different situations. Maybe Steve Nash was a better passer than Rajon Rondo in 2012 — although that’s like picking between the cheerleading captain and dance team captain — but Rondo has better teammates who hit a higher percentage of their jumpshots, which leads to more Rondo assists. But SportVU provides statistics that give context to assists, such as total passes, secondary assists (if Derrick Rose passes to Joakim Noah who quickly finds a wide-open Carlos Boozer under the hoop for a layup, Rose would receive a secondary assist), and shooting percentage after a particular player’s pass.”
Counterpoint: So, it matters that Rondo’s teammates shoot a high percentage? That’s interesting, since this didn’t come up when considering Pierce’s passing ability. I’m not sure how secondary assists help, either, since Rondo likely benefits from being on a team with other strong passers. All of the variables listed above regarding Pierce apply here as well.
No doubt, the tracking data will allow analysts to address some of my complaints. I’m a good complainer, though, so I can probably come up with more. There are many pitfalls when analyzing a fast-moving sport like basketball, where every decision is based on what other players are doing, have done, and will do. I have a feeling teams won’t be using this technology nearly as much as people think. Sometimes there’s no substitute for the human eye-brain combo.