This season the NBA stats website started using exact possession counts instead of the estimate they had previously used. While exact counts had previously been available on some smaller sites, including my site, PBPStats , it was good to see what I would assume is the most visited site for NBA stats finally use exact counts. If you have compared numbers on the NBA site and on PBPStats you have probably noticed that the possession counts are different. How can the counts be different if both are using exact counts? Well it turns out defining a possession is not as simple as one might think. I will use game 1 of the Raptors Magic series to show a few examples of this.
At the end of the second quarter Danny Green makes a three with 0.4 seconds remaining. Orlando inbounds the ball to Aaron Gordon who fails to get a full court shot off before the buzzer sounds. Should this count as a possession for Orlando? Their percentage chances of scoring here are probably in the low single digits so does it make sense to count it the same as we would count a normal mid-quarter possession with a 24 second shot clock.
At the end of the third quarter Pascal Siakam misses a corner three. In the play-by-play Michael Carter-Williams is credit with a rebound with one second remaining in the quarter. Does that mean Orlando gets credited with a new possession after the rebound? As you can see from the video, it would take a miracle for them to score after the rebound.
At the end of the fourth quarter Kawhi Leonard missed a three pointer that would have tied the game and the ball went out of bounds with 0.5 seconds remaining. Orlando just needed to inbound the ball and have time expire to win the game. Should this be counted as a possession for them?
If you look at the stats for this game on the NBA site it has Orlando scoring 104 points on 99 possessions. On PBPStats it has Orlando scoring 104 points on 96 possessions. I don’t know exactly how possessions are counted on the NBA site, but the most logical guess I can make is the three possession difference is because the three examples above are all counted as possessions for Orlando. None of them are counted on PBPStats. Any possession that starts with two or fewer seconds remaining in a quarter isn’t counted as a possession, unless points are scored. I arbitrarily picked two seconds, but the goal was to limit counting end of quarter possessions with little to no chance of scoring.
This is why you will see the team efficiency stats 1-2 points per 100 possessions higher on PBPStats. This isn’t the only place where decisions need to be made on how to count a possession, flagrant fouls are another place (I count the free throws and possession following the free throws as two separate possessions), but it is the biggest one that would be causing most of the difference. I understand why one would count possessions the way they are on the NBA site - it’s easy to understand and easier to code, but I think not counting those end of quarter possessions give a better representation of a team’s efficiency. There is no perfect way to do it. Working with play-by-play data is always going to be messy and there will always be issues you hadn’t considered that come up. I think it’s good to have options available that make different choices and let the user decide which one makes more sense for what they are looking for.