Measuring Half Court Pace

December 23, 2015; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) dribbles against Utah Jazz guard Rodney Hood (5) during the first quarter at Oracle Arena. Mandatory Credit: Kyle Terada-USA TODAY Sports
December 23, 2015; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) dribbles against Utah Jazz guard Rodney Hood (5) during the first quarter at Oracle Arena. Mandatory Credit: Kyle Terada-USA TODAY Sports /
facebooktwitterreddit
December 23, 2015; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) dribbles against Utah Jazz guard Rodney Hood (5) during the first quarter at Oracle Arena. Mandatory Credit: Kyle Terada-USA TODAY Sports
December 23, 2015; Oakland, CA, USA; Golden State Warriors guard Stephen Curry (30) dribbles against Utah Jazz guard Rodney Hood (5) during the first quarter at Oracle Arena. Mandatory Credit: Kyle Terada-USA TODAY Sports /

By now, everyone seems to agree, the only way to beat Golden State is to not let the Warriors control the pace, and rather play the game at your own speed.

This makes a great deal of intuitive sense. After all, the Warriors are experts at accelerating the speed of the game, capitalizing on opponents’ mistakes to rip off huge scoring jags. The Spurs played last Monday’s much anticipated[1. And wholly anticlimactic.] Clash of Presumptive Titans at Golden State’s speed, letting bad offensive possessions reflect back at them with dizzying speed as the Warriors sprinted over, around, through and in-between them.

Speed of the game, its energy, tempo, is vitally important. Controlling that tempo is choosing the field on which the battle will be fought, a huge advantage to whichever team is able to grab it. But for all the understanding of the importance of pace, we do not have a very good way of describing it in statistical terms.

Sure, there is the “pace” statistic, which is not a bad stat, so far as it goes. But number of possessions in a game[1. “Pace” as commonly expressed is simply possessions per 48 minutes of play.] is not quite the same thing as the “speed of the game.” Innumerable quirks can raise or lower this possession count, when exact possession counts are even used.[2. Our “Team Ratings” page uses as close as we can get to “true” possession counts – counting actual changes of possession rather than simply estimating from box score totals.] Offensive rebounds, turnovers, intentional fouls can skew this total to make a game appear to have been played much faster or slower than would be described by the naked eye. Indeed, the 2015 NBA Finals was played at a relatively constant pace as measured in possessions/48 minutes, but even a casual viewing of proceedings would indicate how much more quickly the game was moving as the Warriors rolled to victory in games 4-6, swinging the pendulum back into the Warriors’ favor after Cleveland had dictated a three-yards-and-a-cloud-of-dust feel to the first three games of the series. Even the SportVU speed and distance data doesn’t help much, as only Game Five saw Golden State move appreciably more on offense (around 1130 feet per 24 seconds of offense compared with around 1070 feet per 24 seconds the rest of the series. In fact, the second and third most the Warriors moved were in Games 2 and 3, the contests which were quite visibly played in a grind-it-out style more favorable to the short-handed Cavs.

So how to go about a better measure of speed of play? For one thing, beginning to disaggregate the various pieces of “pace.” By observation, teams don’t play at one speed in all circumstances. Some teams “run” opportunistically, while others prefer to slow it down even when the opponent makes a mistake, and commits a live ball turnover. To some degree this aggressiveness is what we think of when we think pace. But as Mike D’Antoni likes to describe[1. As he did during last season’s Sloan conference.] the quickness of play which characterizes his Seven Seconds Or Less philosophy isn’t simply sprinting up and down the floor, so much as it is constant brisk movement. Get the ball up court quickly, but don’t hurry.

Capturing this “half court” pace is surprisingly difficult, at least with public data. SportVU movement data quite naturally correlates with possession counts, as transitioning from one end of the court to another, and doing so quickly is about the most and fastest players move during the course of the game. While there may be moments of sprinting to cut, rotate or recover in the half court, that distance is miniscule compared to the 50+ feet every player runs every time possession changes. Another stat I’ve been playing with, “defensive movement forced[3. Discussed a little here, but more to come soon.]” suffers similar transition running related issues.

Using play-by-play or shot log data to determine average possession length is promising, but their is still a lot of noise. First of all, which possessions? A sideline out of bounds play out of a timeout (SLOB) is a very different situation than walking the ball up fullcourt after a made free throw. Further, simply averaging possession time introduces several other things into the mix – though the wonderful Inpredictable those aggregate stats here.

A team may want to play slow, but if they turn the ball over frequently, those are shortened possessions (which also tend to lead to short opponent possessions, i.e. fast breaks.) Similarly, early, non-shooting[4. ESPECIALLY HACK-A!] fouls can break up the flow of play. What I really want to capture is some sort of measure of the intended speed of play for an offense.

To do this, we need an apples-to-apples comparison of possessions, so the first thing to do is to remove all of the ‘half court” possessions from the equation. Inbounding after a defensive foul not in the penalty, or a SLOB after a timeout, or a reset shot clock after a kicked ball or technical foul are out, at least for now. Similarly, plays immediately following offensive rebounds (either a live ball rebound by a player or a team offensive team rebound such as after a blocked shot.) or steals are excluded as playing against a scrambled defense represents more the “opportunistic pace” I described above than the more deliberate half court pace under discussion here.

Next, we want to remove all play outcomes except for field goal attempts. This is based on the assumption that for the most part teams are using their offense to try and make field goals, and that fouls drawn are more of a happy accident than the intent. Similarly, turnovers represent something interrupting the normal “flow” of an offense. This leaves us a set of field goal attempts which come in “full court” situations – where the now-offense has the ball at one end of the floor and must dribble up and attempt to score in the allotted 24 seconds.[2. Some of the categories possibly include plays where the ball is inbounded at midcourt such as from deadball turnovers and the like, but for sake of this initial analysis, those are assumed to be a small percentage of plays and thus can be ignored for now.]

Previous research indicates these plays can be broken into four rough categories based on prior action: liveball rebounds from missed field goal attempts; made field goals; free throws made or missed and true deadball situations arising from a team rebound or a deadball turnover. According to  NBA play-by-play logs for this year here are the average lengths of time between prior action and ensuing field goal attempt:

avg timeshots
avg timeshots /

The greater average time elapsed after a made field goal is initially confusing, because the ensuing shots are generally similar efficiency to those after free throws or deadball plays, and later average shot clock times would usually suggest lower efficiency, However, recall the clock runs on made field goals for most of the game, so that extra second or two is mostly just the time it takes to inbound the ball before the shot clock actually starts.[3. This possibly gives an opportunity for further research, as the speed with which a team inbounds a ball after an opponent make might tell a great deal about their intended speed of play. For example, one of the big changes in Washington’s play this year has been their commitment to get the ball out, get it in bounds and go after opponent makes, which has quite naturally increased their pace of play considerably. Perhaps greater or smaller average gaps between average time off made field goals and other full court plays could indicate something about this intentionality.] The quicker time period between defensive rebounds and ensuing shots is the result of transition play, which certainly is a part of determining pace, but it seems as much a part of opportunistic pace as half court play. So we’ll exclude that as well.

Finally, this leaves us with a dataset, through Sunday’s games of just over 50,000 shots, or about 1,700 per team, for this season. The average time is what I’ll describe as a team’s half court pace. When things go the way they intend their offense to work – that is, they get a shot – this is the average time taken. One final change has to be made to prevent “advantaging” teams that play better defense. Squads which end up taking the ball out of their own hoop much more often are naturally going to be at a disadvantage on this measure just by virtue of the extra second or two of running clock after a make. So the final adjustment is to scale each team’s average to what it would look like with an standardized distribution of actions preceding these shots.

Below is the average half court pace for the NBA this season:

Though this list tracks fairly closely with more traditional measures of “pace” using simple possession counts, there are enough differences for the exercise to have felt worthwhile. Further, this is only a first step towards developing a more robust understanding and definition of pace, possibly combining this information with some sort of quantification of “opportunistic” pace, though that is an exercise for a later time and post.


Thanks to Zach Harper for asking the questionswhich inspired this post.