# The Value of Player Movement in the NBA

By Harrison Chase and Carlos Pena-Lobel

In a post we wrote last spring, we looked at the teams with the best passing and movement over the course of the 2013-2014 NBA season. Having done this, we will now attempt to determine whether passing and movement actually improve an offense, and if so, how.

Intuitively, it seems as if passing and movement would have a positive effect.  At the youth level, practices are devoted to teaching players how to pass and move off the ball. The idea behind this is that these are fundamental skills that are easy enough to learn, but still majorly contribute to the success of an offense.  In the NBA, the San Antonio Spurs under Greg Popovich have been the prototypical passing and moving team; in our last post we discovered that they ranked first in both categories.

However, our last post did not indicate any benefits from passing and movement. That was intentional; there is no correlation between a team’s pass or run score and their winning percentage. Same thing for offensive efficiency, field goal percentage, three-point percentage, or any other advanced measurement of offensive efficiency. The main reason behind this is that other factors, like team talent, will have a much larger impact on winning percentage than movement and passing. Obviously, a team like the Heat last year (who were playing four future HOFers) were going to have a better offense than last year’s Bobcats no matter how much Charlotte runs or passes. Therefore, when determining the effect of passing and movement we need to control for talent, which we do by including a dummy for each team in our model.

First, we looked at all games played in last years regular season (except for the eight games which did not have player tracking data) giving us 2,444 observations. We then estimated possessions in a game using the basic possession formula found here, and used that to calculate distance run per possession (DPP[i]) and passes per possession (PPP) for both teams in every game.  Since we are controlling by team, this effectively measures the amount a team passes/moves relative to its own average.

The data we had for distance in a game was in terms of miles, but in our model we converted it to hundreds of feet for ease of interpretation. We did not have a teams’ time of possession for individual games, so we had to rely on PPP and DPP relative to a team’s season average to see if they were running more or less than usual.  We ran a multiple linear regression with four factors initially: a categorical variable for the team’s offense (can be thought of as a measure of offensive skill), a categorical variable for the opposing team’s defense (can be thought of as a measure of the opposing team’s defensive skill), DPP, and PPP.

First and foremost, we observed that PPP and DPP were significantly correlated with a team’s distribution of shots. Namely, as teams passed the ball more they shot more 3-pointers, while when they ran more they shot less from behind the three-point line and more at the rim. Intuitively, these findings makes sense; the more teams are swinging the ball around, the more likely it is that they will find an open player behind the three-point arc.  Meanwhile, if teams are cutting through the paint, or running lots of pick-and-rolls, it makes sense that both their distance traveled and shots the rim would go up. The table of how PPP and DPP affect both three-point rate and shots at the rim is below:

 How PPP Affects a Teams Distribution of Shots Coefficient Standard Error t value Pr(>|t|) Three Point Rate 0.0320914 0.0050443 6.362 2.38e-10 *** Two’s not at the Rim -0.0138844 0.0113676 -1.221 0.222055 Shots at the Rim -0.0261 0.008316 -3.138 0.001721 ** Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

 How DPP Affects a Teams Distribution of Shots Coefficient Standard Error t value Pr(>|t|) Three Point Rate -0.0166115 0.0070267 -2.364 0.018156 * Two’s not at the Rim -0.0059964 0.0081606 -0.735 0.46253 Shots at the Rim 0.0305 0.01158 2.633 0.008525 ** Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

It is interesting to note that the percentage of a team’s shots that are neither at the rim nor from behind the 3-point arc (i.e. midrange and long twos) aren’t correlated with either running of passing; Rather there seems to just be a trade off just between three pointers and shots at the rim.

So PPP and DPP are strongly correlated with a team’s distribution of shots, but what else are they correlated with?  One main thing that both of them affect is the percentage of shots that are unguarded. Even after controlling for a team’s shot distribution (which is important to do as three’s are more likely to be unguarded and shots at the rim are more likely to be contested) is evident that both passing and movement lead to more unguarded shots. The table of coefficients is below.

 Effects on a Team’s Percentage of Open Shots Coefficient Standard Error t value Pr(>|t|) PPP 0.010771 0.00498 2.163 0.03065 * DPP 0.022062 0.006891 3.202 0.001385 ** Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

DPP is also highly correlated with another statistic: effective field goal percentage, a metric for gauging a team’s shooting effectiveness. However, PPP shows no correlation (remember, this is after controlling for a team’s distribution of shots), which could go to show that unguarded shots at the rim are still better than 3-pointers.

 Effects on a Team’s Effective Field Goal Percentage Coefficient Standard Error t value Pr(>|t|) DPP 0.0231 0.007642 3.037 0.002418 *** PPP -0.00498 0.00523 -0.903 0.366715 Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Now all of these little stats are cute and everything, but the goal of an offense is ultimately to score points.  This brings us back to the central question; do passing and movement actually help teams to score more? To answer this we calculated a team’s offensive efficiency in each game (basically a team’s points per 100 possesions), and ran the regression on that. However, we also needed to control for a few more things. Our formula for pace took into consideration not only shots but also free throws, turnovers, and offensive rebounds. All of those are correlated with either passing or running per possession. For example, as the formula for possessions does not count offensive rebounds as a separate possession, when a team gets a higher percentage of offensive rebounds will have more time to run and pass the ball around. Meanwhile, when teams that shoot a higher rate free throws will have their possession cut short, and therefore have less time, and when teams have a higher turnover rate they will have less time to move on offense. Of course, there is a possibility that passing and movement is actually causally correlated with one or more of these factors, but to be safe we controlled for all three. Therefore, we added offensive rebounding percentage, turnover percentage, and free throws per possession to our model and ran that regression. To our surprise, only DPP was significant.

 Effects on a Team’s Offensive Efficiency Coefficient Standard Error t value Pr(>|t|) DPP 7.15416 1.26470 5.657 1.73e-08 *** PPP -0.16034 0.90863 -0.176 0.859941 Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

So wow.  Much significance.  Such effect. There is a huge correlation between the distance teams ran per possession and their offensive efficiency. The difference of 7 points is roughly the distance between the Heat’s offense and the Lakers’ offense last year. Now does that mean that if a team increases their DPP by 100 feet they will transform from Swaggy P and Jodie Meeks into LeBron and Dwayne Wade? No. For one, any team committing to this must be in incredible shape, or they will tire out quickly. Furthermore, it has to be purposeful running, not just running around in circles like this. Also, they will probably face diminishing marginal return.

Finally, and most practically, it would be very hard for a team to increase their DPP by 100 feet – the difference last season between the average distance run per possession by the most active and the least active team was only 45 feet.  While it is pretty obvious that the most active team was the Spurs (here’s another), Minnesota was more of a surprise for last place (although this could be due to Kevin Love’s outlet passes which causes Minnesota to still have half the team behind half court during a possession.)

Still, even though it is uncertain how much exactly a team could improve its offense by running more, the fact remains that it probably wouldn’t hurt them (unless they have players like these). By running more a team can increase its percentage of shots that are unguarded, it can increase its effective field goal percentage, and ultimately can significantly impact its offensive efficiency. In conclusion, this might be the only analytically sound thing Byron Scott will do all season.

[i] We only had data for the distance run by a team in the whole game, so we divided that by two to get miles run on offense.

View all posts