Editor's note: Dean Oliver is the author of "Basketball on Paper" and is ESPN's director of production analytics. He was formerly the director of quantitative analysis for the Denver Nuggets.
Everyone is hungry for it -- an impending landslide of data that tells us where every player and the ball is during every point in every NBA game. It's a data set that answer these questions, among other things:
• Who can shoot a jump shot with defense in his face?
• Who isn't getting back on defense?
• Who draws the defense the most?
• Who makes effective drives to the basket?
This is SportVU, a product of Stats LLC that the NBA purchased to be used across every arena starting this season. It's the culmination of a decade of the basketball analytics revolution, where different people, different methods and different data have advanced the status of basketball as a thinking man's game.
But let's back up a little bit.
Before the previous decade, much of basketball analytics focused on player valuation metrics -- numbers that essentially ranked players from best to worst. The motivation for this was simple: Players are hugely important in basketball. My old boss in Denver, Mark Warkentien, used to talk about how "I'll give you the best coach along with a bunch of guys who play hard and play together, but I'll take LeBron James, thank you." Having a metric that summarizes a player's value in one number has been viewed as the Holy Grail, theoretically allowing decisions about personnel to be easy.
"For years, Warkentien has evaluated players with an approach he calls 'eyes-ears-numbers.'" -- Benjamin Hochman, Oct. 12, 2009
The problem with player value metrics is that there is little to validate them, meaning that no metric has established itself as clearly the best. Metrics couldn't even be fairly compared. As each new metric has been developed, it has served mostly to complement traditional scouting, a way to reality check when subjective opinions formed by watching and hearing about players were going too far astray.
Player value metrics are not the only basketball analytics, though. It was 10 years ago this week that my book "Basketball on Paper" was released. That book focused on teams first and players second. It broke down the game, dividing points into efficiency and pace, with a focus on how teams and individuals could be efficient. The Four Factors break down efficiency into shooting efficiency, rebounding percentage, turnovers per possession and getting to the line. Partly because of this perspective, there has been a more broad emphasis on shooting (and defending) layups and 3-point shots, which have the highest effective field goal percentage of any shots on the floor. This did not require my book, but it highlights the book's approach of using analysis of readily available data -- box scores and shot charts -- to suggest simple ways to influence the game.
But consider this quote:
"The information that appears in box scores, television graphics and many scouting reports often does a poor job of explaining a game." -- David Leonhardt, Jan. 9, 2005
There have been attempts to gather new analytical information to augment what has been in traditional box scores. A service called Synergy was automatically chopping and tagging video clips so general managers, scouts and analysts could see post-ups, pick-and-rolls and isolations. Synergy was and still is used throughout the NBA to combine some analytical statistics with video to evaluate both tactics and components of player value.
Play-by-play data also became more broadly available, allowing for some context-specific evaluation, particularly who is on the court together. Michael Lewis, author of "Moneyball," wrote an article for The New York Times highlighting how teams with Shane Battier, a player who doesn't accumulate a lot of traditional stats, did better with him on the court.
Plus-minus is a simple fact that tells you whether the team does better or worse while a player plays. Though not a player rating, people often get tempted to look at it that way, and a method, adjusted plus-minus, was developed to try to use that number to estimate player value. That method was occasionally hyped as being able to "account for everything that happens on the court."
Today, even with metrics that claim to account for everything, the analytics community still craves the SportVU data. If metrics already incorporate everything, why would even their creators want this?
Because it will tell stories in greater detail than has ever been done before. The data is now there to not only explain player ratings but also to provide information that has never been presented. Players who "draw the defense" can be quantified as ones who pass the ball after having two or more players around them. "Getting good spacing" can become a visualization illustrating which players control parts of the court. "On-ball defense" can be isolated to see how often a player defending the ball cuts his opponent's chance of scoring.
"Every action on a basketball court is influenced by nine other players, not to mention a coach. For this reason, there is no 'holy grail' in basketball equivalent to baseball's on-base percentage." -- Chris Ballard, Oct. 21, 2005
That quote rings true. Even though we have all this data, that doesn't mean that player value becomes easy and obvious. Data is still just a step toward information, and information is just a step toward knowledge. Player tracking data is a step toward information about drawing the defense, spacing and on-ball defense. Those are steps toward player value, which is the knowledge owners pay for.
Imagine comparing Kobe Bryant and LeBron James based on several pieces of information, such as:
• Man defense on wing players
• Creating better shots for teammates on the perimeter
• Creating better shots for teammates at the rim
• Help defense on drives
• Jump shots off the dribble, adjusting for location
This would be fascinating. What if James is just the 44th-best wing defender and Bryant is third best? For fans, it's a simple debate. For coaches, it is a management tool for how to use Kobe or how to use LeBron. For management, it suggests what kind of player to bring in to complement a core player's strengths and weaknesses.
"It's one thing to have a ton of data. It is entirely another thing to know what to do with it." -- Patrick Minton, Sept. 17, 2013
Indeed. As interesting as these snapshots can be, these details can potentially blind you, a coach or a GM from the bigger picture, which is that LeBron is just really good, regardless of whether he ranks low in some reasonable-sounding statistics. A player metric that doesn't put such new statistics into the perspective of winning and losing is going to fail as much or more than older metrics that don't have those statistics.
This also is SportVU, a data set that introduces a new kind of uncertainty. The data will be there for every team to use however it wants -- including making mistakes. The uncertainty in player evaluation, tactical evaluation and general team construction comes less from lack of data than from lack of analytic power. Decisions will continue to be made, some using "gut," some using data and the best ones using the right blend of both.