Adjusting pace

[andylarsen]

Probably the most accepted change in how NBA people look at stats is adjusting for pace across the board: most serious analysis happens right now on a per possession basis. This is good. But the basic calculation of pace (simply, average # of possessions per game) itself has always frustrated me, especially when used as a proxy for how fast teams play:

1. Teams with more close games will have more possessions at the end of them, sometimes significantly so, due to free throw battles. Ditto to teams with poor FT% shooters, when teams use a Hack-a-Player strategy.
2. The categorization of an offensive rebound as a continuation of a possession. Teams with more offensive rebounds will be characterized as playing slower, even if that might not be the case: they're simply having their possessions extended by good rebounding.
3. Teams with more turnovers, or that force more turnovers, will have a high pace statistic, even if they're not necessarily getting their own shots up faster.
4. That pace alone doesn't tell the story. Let's say a team shot on offense on average after 10 seconds, but then played defense so that they forced an opponent shot after 20 seconds have gone off on the shot clock, that team would have a pretty average pace statistic. But we'd be missing the story of the fastest offense in NBA history.

Has anyone here explored anything like an "adjusted pace" that would accommodate some of these issues? Or pace split up into offensive or defensive sides? This seems like it would be useful in characterizing more accurately how teams play.

[J.E.]

I think pace does what it's supposed to do.

If you want to you can create new metrics like
- "average time before team takes a shot"
- "average time of possession, ignoring possessions that end in turnovers"
- etc.

Obviously, you'll need to parse PbP to get most of these

[schtevie]

A few observations on the subject:

I sometimes think things would be a bit more intuitive if the pace statistic were presented as its inverse: average seconds per possession. This, for example, would show that the average team this season has a "pace" of 15.3 sec/poss, and the the fastest and slowest teams (GSW and Utah Jazz) are at 14.6 and 15.9, for a deviation from average of -0.7 and +0.6, respectively. Apparently no big difference.

However, the measure we really want is the average time of offensive possession per team, and one can estimate this by assuming each team plays the average opponent. In which case, GSW's and Utah's "pace" becomes 13.9 and 16.6, with deviations from average of -1.4 and +1.2. Well, that's a bigger difference, but what explains it?

I don't think as a general rule (and specifically in the instance) it is differences in turnover or offensive rebound percentages or foul incidence that explains most of the observed variation. Between there simply not being much variation in these statistics and the individual effects not being large, pace differences are largely explained by the most obvious factor: the propensity to fast break.

What is the duration of an average fast break possession? 5 or 6 seconds. As such, it can be calculated for an average team that each additional percentage point of fast break possessions in a team's mix will decrease "pace" between 0.1 and 0.2 sec/poss.

So from 82games, we see that this year there is a 13% difference between the Warriors and the Jazz in fraction of shots taken between 0 and 10 sec. Let's suppose that that is the difference in fast break attempts, what implies that over half the difference between the fastest and slowest paced teams are accounted by that factor.

Then a final observation, what this also implies is that true half-court offense are all about the same, pace-wise, with perhaps only a second, on average, separating the fastest and slowest. And when you think about it, that alleged fact shouldn't be surprising at all.

[Crow]

Some of these points have been raised before but some are new and this is perhaps more in sum than I can recall on the topic than anywhere else. Plenty of ways to break it down. At minimum, it would be nice to see some database with offensive and defensive pace in terms of avg. seconds per possessions. Hope to see it, here, at Nylon Calculus or somewhere. An example of something that has been possible for awhile but just hasn't been pushed thru yet.

I asked about individual pace years ago but that line of inquiry hasn't seen much or maybe any public development, or has it somewhere that I am forgetting?

[andylarsen]

J.E.: I think Pace does a good job of one of its goals, which is to give us an idea of how to adjust the stats up or down based on the number of possessions. What is up in the air, IMO, is if it does a good job in describing how a team plays. This is where it falls short to me.

Schtevie: I like how you broke it down here, and I agree: the difference in the fastest pace and slowest-paced teams is probably at least 50% due to how quickly they shoot the ball, which is good. But that leaves a lot of play, and there are a lot of assumptions made here. The biggest is probably the simple doubling of the gap between the two teams in seconds per possession to describe the offensive split, when that may be in fact caused by in difference between how GSW and UTA's defenses play, not their offenses.

Crow: Agreed, I think this is a natural fit for Nylon Calculus. I should probably just scrape up a PBP database, have a first shot at trying to make this work, and send the work to Ian.

[NateTG]

...
Obviously, you'll need to parse PbP to get most of these

A bit of a digression:
There's a bunch of work involved in compiling play by play databases, and some copyright concerns. There's potential for huge duplication of effort if everyone parses and scrapes their own data. Do we have some notion about best practices with respect to sharing that kind of work?

[italia13calcio]

Some of these points have been raised before but some are new and this is perhaps more in sum than I can recall on the topic than anywhere else. Plenty of ways to break it down. At minimum, it would be nice to see some database with offensive and defensive pace in terms of avg. seconds per possessions. Hope to see it, here, at Nylon Calculus or somewhere. An example of something that has been possible for awhile but just hasn't been pushed thru yet.

Real easy way to get an estimate of avg. seconds per possession is to just divide teams time of possession per game from the SportVu data by a team's pace

[sndesai1]

...
Obviously, you'll need to parse PbP to get most of these

A bit of a digression:
There's a bunch of work involved in compiling play by play databases, and some copyright concerns. There's potential for huge duplication of effort if everyone parses and scrapes their own data. Do we have some notion about best practices with respect to sharing that kind of work?

copyright concerns?

[NateTG]

...copyright concerns?

Sure. Also questions about format and distribution method.

[boooeee]

New post up that touches on this topic. I'm using standard play by play data to measure offensive and defensive pace separately - no SportVU data needed.

Offensive versus Defensive Pace in the NBA

The Warriors' have by far the league's lowest seconds per possession on offense, 13.4 seconds per possession - there's a chart in the post showing what an outlier they are. They actually have the third slowest defensive pace, but their offensive pace is so quick, they still lead the league in overall pace. I'll eventually add these numbers real time to my site, allowing for real time updating and drill downs by date range and season. I also have stats on average time to shot that I'll add soon as well.

[Crow]

Alright, alright. Progress. Nice work.

[schtevie]

boooeee, I've got a righteous and proper rant pending on the subject at hand, and in service of which might I make a data request that hopefully will clarify issues?

As you have the underlying PbP data, could you present the decomposition for each team between fast break possessions and "other" (i.e. half-court sets), both for offense and defense? You could let the data speak to select a time cut-off for however many seconds or less for a fast break (or is there a generally accepted standard?)

An initial observation (and what merits extended comment) is that there is extraordinary little variation across teams in terms of defensive possession time (what, of course, makes sense). But this said, it would be good to know if this story changes at all in terms of half-court defense.

Thanks.

[boooeee]

I've got things broken out by how the possession began which I think may get at what you need.

There are three main ways a possession can start: after your opponent makes a shot (or their final free throw), off of a turnover, or off of a defensive rebound. I will share how each team's pace varies by how the possession started. For GSW, it turns out they are fast regardless of how the possession starts. Possessions off of turnovers and defensive rebounds are much quicker than ones off of opponent made shots (as expected). But there isn't much difference in pace between possessions off of turnovers or possessions off defensive rebounds.

[schtevie]

Thanks for the comment, and I don't want to presume, but possession origin isn't really a relevant factor for understanding the particular issues at hand.

Whatever the origin, either the offense fast breaks and beats the defense down the court so as to get a higher quality shot (in expectation) than otherwise or it doesn't. Yes, some origins facilitate fast breaks relative to others, but that doesn't address the issue of the degree to which half-court offenses and defenses differ significantly in terms of average elapsed time of possession.

I am supposing that your data indicates elapsed time at first scoring attempt and that the overall distribution should show a "trough" identifying the demarcation between fast breaks and half-court offenses.

A rational discussion of pace begins with identifying these key distinctions, on both the offensive and defensive ends: division between fast breaks/half-court offenses and efficiency of both.

[boooeee]

Thanks for the comment, and I don't want to presume, but possession origin isn't really a relevant factor for understanding the particular issues at hand.

Whatever the origin, either the offense fast breaks and beats the defense down the court so as to get a higher quality shot (in expectation) than otherwise or it doesn't. Yes, some origins facilitate fast breaks relative to others, but that doesn't address the issue of the degree to which half-court offenses and defenses differ significantly in terms of average elapsed time of possession.

I am supposing that your data indicates elapsed time at first scoring attempt and that the overall distribution should show a "trough" identifying the demarcation between fast breaks and half-court offenses.

A rational discussion of pace begins with identifying these key distinctions, on both the offensive and defensive ends: division between fast breaks/half-court offenses and efficiency of both.

I took a look at the distribution of possessions, and there is no clear break point in terms of length. The distribution looks roughly bell shaped with a peak at 16 seconds. See the end of this post for the details: Team Pace and Efficiency by Possession Type.

However, if you had to pick a "fast break zone", 4-6 seconds seems to be the most likely candidate. There is a slight bump in the distribution, and a clear bump in efficiency (see last chart in the blog post).

Another callout - Possessions off of turnovers (1.12 points/poss) are significantly more efficient than possessions off of made shots (1.00 points/poss) or defensive rebounds (points/poss). But the Brooklyn Nets haven't got the memo. Their efficiency off of turnovers is not better than efficiency off of made shots.