APBRmetrics

[Ben F.]

I was working on a project counting possessions from the play by play, and I realized that possessions at the end of the quarters is a thorny issue.

Have we come to a consensus regarding whether we count possessions from their start or their end? It doesn't matter at any time but at the end of quarters, but that ends up being 4 possession a game going either way. It seems that most estimates count possession endings (what they produce) because that's the only way to measure them from the box score. But if a team gets the ball with time left on the clock before the end of the quarter, and they don't end the possession by turning it over, getting a shot off, or getting to the line, is that really a possession? It would seem like it should be, as long as the team made an attempt to score, but that of course is impossible to judge from the the play by play. If you do it by start, actual scoring attempts become fairly indistinguishable from one team missing a shot and the other rebounding it with 1 second left, and the team making no attempt to score.

Perhaps for that reason and because the standard has been to only count those possessions that produce results we shouldn't count them. But I thought it was worth bringing up, at least.

So any thoughts? I know it seems a bit trivial, but I do think it's important both for accurate possession counts and for having a sound definition of what a possession is.

[Chicago76]

Mathematically, a possession ends with a defensive rebound, a made field goal, or a turnover. For single game results, quarter-ending possessions may prove to be an issue, as one team can have 4 and the other could have zero. Assuming 100 possessions in a game and 100 pts per team, offensive efficiency could be over/underexpressed by +/- 4. Over the course of a season, we would expect end of quarter possessions to roughly equal out, so the issue is immaterial in that application.

On a game by game basis, how you deal with quarter ending possessions depends upon what you're trying to measure, I suppose. Assume you're trying to measure offensive efficiency between both teams. Should you include the following items:

-a 40 foot halftime heave that clanks off the rim. The shooting team has no timeouts and received the ball after a made basket with 2 seconds on the game clock.

-a defensive rebound with 0.2 seconds left on the game clock that results in no FGA or turnover

-a winning team dribbling out the clock at the end of a game

-a team driving and scoring at the buzzer, but the basket is waived off for a charge. Technically, the defensive team won possession, but time expired.


If I was trying to get an accurate measure of Off Rtg for both teams, I'd argue:

-don't count the FGA or the possession in #1,
-count the possession for the shooting team in #2, but not for the rebounding team
-don't count the possession in #3 when the winning team has zero incentive to do anything that would record one.
-in #4, record the possession for the team committing the turnover, but not the team taking the charge.

Ideally, possessions and points are recorded when a team has a realistic chance of getting a "reasonable" field goal attempt. That line becomes blurred though. What is reasonable? 4 seconds left, the length of the court and no timeouts? Inbounding at halfcourt post timeout with 1 second left? In example #2 above, what if the shooting team got the ball with 3 seconds left drove, tossed up a 40 ft runner at the buzzer, and the other team rebounded? Is that a possession that would measure the true Off Rtg of either team or properly reflect FG%?

You could go on and on with these types of examples. I think the important thing to recognize is that there is a little noise in everything we're trying to measure on a single game basis due to game clock issues. To me, it's not worth the effort.

FGM + constant * FTM + TOV + Opp DRB is good enough for me

[Kevin Pelton]

In our JQAS paper, Dan, DeanO, Justin and I used the following defintion:

[Ben F.]

OK, that seems like a pretty sound definition that takes into account what I was talking about, I was just wondering if it existed.