PASS Networks

Bob Donnan-USA TODAY Sports
Bob Donnan-USA TODAY Sports /
Bob Donnan-USA TODAY Sports
Bob Donnan-USA TODAY Sports /

According to Merriam-Webster’s Dictionary, a network is:

noun net·work ˈnet-ˌwərk, a group of people or organizations that are closely connected and that work with each other

There are social networks (Hi, Twitter).

There are professional networks (Hello, LinkedIn).

And, there are team-based networks. In your office and on the court. While these can be seemingly forced, particularly as a function of the lineups that a coach puts together, they are best measured by how a collection of players work together. Lineups are mandatory; networks are created from the successful relationships within a lineup. When that cohesion works in coordinated fashion to achieve a goal, scoring and defending become natural products of the connectivity among that group. If we had to pick one (and only one!) measure to determine how well a lineup collaborated, the choice is clear: it’s the pass. More significant than the points scored, a pass signifies teamwork; it’s a measure of trust and relationships. Passes get a lineup moving towards a common goal: creating opportunities to score points.

If passing is important (and clearly it is), then the amount of passing a team does should indicate a stronger network cohesion. This has been done before, and this prior work by Andrew Bergmann highlighted how well each starting lineup shared the ball. It was an essential first step of network-based passing analytics.

So let’s take that second step, to PASS Networks.

PASS_1415Warriors /

How is this different? For starters, we are examining the entire roster, not just the starting lineup. But more importantly, rather than looking at just total pass activity (or even normalized pass activity), this network is based on Points Generated Per 100 Passes. By normalizing by points scored, we get a sense of how valuable different passes are. Further, instead of displaying all connections across the roster, this network only shows above passes above league average points per 100 (which last season was roughly 54 points).

So as you’re digesting this chart, here are some of the things to pay attention to:

  • lines: pass activity weighted by points per 100 passes (thicker means more points) – notice the direction of the arrows
  • nodes: connectivity – the larger the dot, the more teammates that player delivers passes to
  • colors: communities – players exhibiting common characteristics

Armed with that knowledge, what’s good about last season’s Warriors PASS Network? First, it’s full of large nodes: Golden State had a lot of players making above-average passes for scores. A lot. Second, there’s a fair amount of bidirectionality: two-way communication or, reciprocity. Third, there’s only two communities, which organize roughly into main contributors and reserves. But above all, look at the levels of “super-facilitation” from Draymond Green, Andre Iguodala, Shaun Livingston, and (to a lesser extent) David Lee. And, look how many players feed Klay Thompson (and not Steph Curry) at an above-average level!

Now let’s compare these historic Warriors to a less-historic team, last season’s New York Knicks:

PASS_1415Knicks /

The differences are astounding. First, there’s completely unconnected nodes throughout the roster: players who have no part whatsoever with pass activity at above average scoring levels. Second, none of the nodes are particularly large – they had no “super-facilitators.” Third, there’s three disjointed communities. Sure, some of that is attributable to midseason trades, but plenty of it isn’t.

If we reasonably believe what good and well, not-good, look like, let’s fast forward to this season and take a look at Golden State thus far:

PASS_1516Warriors /

Truth, there are currently four communities; this is likely a function of only a fraction of the season completed. Judging by the size of the nodes and the width of the connections (not to mention an undefeated record), the process for getting buckets is strong. Very strong. However, an interesting note: notice how many more teammates are feeding Steph Curry this season, compared to last. This does not mean that Leandro Barbosa is passing to Steph Curry a ton. It does mean that when Barbosa passes to Steph, the result is oh-so well above average.

As for this year’s Knicks, their PASS Network appears to be a step in the right direction:

PASS_1516Knicks /

Still three communities…but that should be the least important takeaway here. They are feeding each other and working together. Look at all the players assisting to Carmelo, at above average levels!

So what – how is this helpful? At first glance, it may seem that this is no better than just looking at Assists statistics. I would disagree – aggregate stats are great, but not at the nuance. PASS Networks show the connectivity (or lack thereof) across teammates. Look back at last season’s Knicks – who would’ve thought Sweet Lou Amundson and Cole Aldrich were such a dynamic duo?! This isn’t just about generally creating shots; networks get you analyzing the best dynamics across the entire roster.

More to come throughout the season.

Data and photo support provided courtesy of and data extraordinaire Darryl Blackport.