Basketball Player Movement Between International Leagues and the NBA

This directed graph, created with R and igraph, shows how professional basketball players transition between leagues around the world. The size of the edges is based on the number of players that transitioned between the new nodes (leagues). 

Pro Basketball Players Transitioning Around the World

The following is a heat map that shows the same information, but in matrix form. It’s a little easier to compare transitions between two edges, but is less useful to understanding the network formed by the transitions.

These graphs were created for a talk that Shane Sanders and I conducted at the Midwest Sports Analytics Meeting 2018.

  • S. Sanders and J. Ehrlich, “Around the World: Rating Men’s Professional Basketball Player and League-Quality by Estimating Player Win-Value Changes across Leagues,” presented at the Midwest Sports Analytics Meeting, Central College, Pella, IA, 17-Nov-2018.

How soon do you know when an NBA game is done?

Looking at play-by-play data of every game of every season between 2004-05 to 2015-16, I was able to determine at every second whether the team that is currently winning will win the game. I averaged every second of every game (regular and post-season) to come up with the following visualization:

I then averaged each minute to come up with this visualization:

The data can be download here.

NBA Shooting while Winning versus Losing

It has been hypothesized that younger players may shoot threes better when their team is winning [1]. To see if this is true at the NBA level, I analyzed every shot of every game from the 2004-05 to the 2015-16 seasons. I found no statistical difference when shooting threes while losing or winning. I found the same to be true of free throws, but I did find a difference (p=0.037) when shooting twos. Interestingly, when the team is losing, the average player shoots 46.28%, but when winning the average goes down to 45.78%. However, the effect size is d=0.03, which is small.

Here is a density map of the three-point shots, along with the analysis:

t = -0.80358, df = 11738, p-value = 0.4217
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-0.009613867 0.004023235
sample estimates:
mean of losing mean of winning
0.2918484 0.2946437

Here is a density map of the two-point shots, along with the analysis:

t = 2.0842, df = 15112, p-value = 0.03716
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
0.0002954158 0.0096292267
sample estimates:
mean of losing mean of winning
0.4627997 0.4578374

Here is a density map of the free throws, along with the analysis:

t = 0.38493, df = 13994, p-value = 0.7003
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-0.004826125 0.007184844
sample estimates:
mean of losing mean of winning
0.7277869 0.7266075

The raw data can be download here.

[1] A. Glockner, Chasing perfection: a behind-the-scenes look at the high-stakes game of creating an NBA champion, First DaCapo Press edition. Boston, MA: Da Capo Press, 2016. p. 85

2014-2015 NBA Regular Season Team-Level Data Parallel Coordinate System

The following D3 visualization demonstrates a parallel coordinate system, which is useful for high-dimensional data. Try dragging and dropping the axes to rearrange the visualization.

The following paper provides an introduction to parallel coordinate systems:
Inselberg, A. (1997), “Multidimensional detective”, Information Visualization, 1997. Proceedings., IEEE Symposium on, pp. 100–107