Origin of Possession


Mar 3, 2015; Cleveland, OH, USA; Cleveland Cavaliers guard J.R. Smith (5) reacts after a 110-79 win over the Boston Celtics at Quicken Loans Arena. Mandatory Credit: David Richard-USA TODAY Sports

For the last four years, I have been tracking efficiency and length of possession for a college team, separated by the origin of possession (the action which immediately preceded the change of possession). It works out to roughly 8,000 possessions in total.

The idea originated from something very basic — the ‘points off turnovers ‘stat that is frequently mentioned during a game. Obviously, live-ball and dead-ball turnovers produce much different ensuing possessions, and even live-ball turnovers produce much different offensive possessions as a function of where the turnover occurs.

That led me to wonder if there were disparities in the efficiency of other possession starts at a lesser scale, what I called “possession coupling.” That team’s PPP off of each type of possession start over the last four years suggests there might be:


The 2014-15 season data does not include all games played yet, which may explain the “Missed Layup” and “Make FT” outliers. Even if those outliers remain, there appear to be some loose patterns beyond the scoring boost provided by live turnovers.

  • Possessions starting off a blocked shot produce offensive possessions at an efficiency rate more similar to live turnovers than any other possession start. This passes the eye test as well.
  • The tightest pattern appears to be possessions starting out of a timeout (which included the first possession of each half, and all media and team timeouts). There are intuitive narratives that seem to fit with this, although it is far too early in the process to give them voice.
  • Taking three-pointers appears to be a good way to defend this team in years where it is not very efficient. Making three-pointers generally appears to be a good way to defend this team.
  • Missing layups is a bad way to start a defensive possession

This is the same team’s average time of possession to first shot (or turnover prior to a shot):

Whereas the average output of each type of possession start varied season to season, the average lifespan of each type remained strikingly, incredibly similar.

If that similarity:

  • Proves to be statistically significant
  • Holds true across all teams playing with a :35 shot clock
  • Is found across all teams playing with a :24 shot clock and having SportVU in their arena

Then player tracking can be used to better understand why. The line of thinking goes something like this:

  • An attempted shot is a function of being open AND/OR a function of the shot clock (or a function of being J.R. Smith)
  • Openness is a function of where the shot-taker is located and where the other nine players on the floor are located
  • Where those players are when the shot is taken is a determining factor of where they will be when the shot is either made or rebounded by the defense
  • That location, and the alignment relative to all other teammates, and to the five opponents, dictates the possible future movements and locations throughout the ensuing possession
  • So that, to some degree, the shot a defense allows creates a defined set of possible player alignments on the ensuing possession, and likewise the shot an offense chooses to take creates that same set of possible player alignments.

An extreme example: a team is protecting a late game five-possession lead. They are less concerned with scoring than with giving up a quick score, so they choose to run the shot clock all the way down and simply get a shot on the rim after :35. In this scenario, what is the optimal shot to miss, and where should the four non-shooters be to ensure their best defensive performance on the ensuing possession?

Against this particular team that I track data for, it would appear that taking a three-pointer would be the baseline answer. Making it would be even better. Visualizing the ideal shot to create the highest likelihood of a defensive stop on the ensuing possession, it would be a three-pointer from the silo, with ideally four, and at least two, additional players outside the arc above the break and any players inside the arc retreating as the shot is released…which begins to sound something like what NBA teams who forego offensive rebounding do.

Beginning to think more about offensive and defensive possessions as a chain of interconnected events could unlock all sorts of knew basketball understandings.