November 21, 2015: Vancouver Canucks Center Henrik Sedin (33) [1917] and Vancouver Canucks Left Wing Daniel Sedin (22) [1911] talk during a break in action against the Chicago Blackhawks on Saturday night in Vancouver, British Columbia.
One Timers

Using NHL passing data to uncover playing styles

Recently, I’ve focused on exploring NHL passing data compiled by Ryan Stimson and the volunteers involved in the passing project. I spoke about the value of shot assists specifically and took a closer look at some exceptionally good (and exceptionally bad) passers, along with offering some thoughts on using the data to test hypotheses when watching live action.

Here, I’ll shift focus and propose a simple system for classifying playing “styles” for skaters based on their total contribution to their team’s offense.

Before we start, a little groundwork. In the past, Stimson has done work interesting work centred on the New Jersey Devils (check here and here) incorporating passing statistics with shooting metrics to gather an understanding of how much offense flows through a given skater when they are on the ice.

Eventually, Stimson settled on primary shot contributions (PSC) as the name for the stat measuring a player’s primary passes that lead to shots (shot assists) and their individual shots (that is, Corsi events – any shot on goal, shot that misses the net, or shots that are blocked).

PSC (and PSC rate, expressed as PSC60 here) is valuable for a number of reasons. Most important, perhaps, is that PSC60 can be useful in predicting the number of primary points a player will accumulate. From Stimson’s piece:

stimson chart 1

Here, Stimson shows the value in tracking a player’s PSC60. This metric is interesting, in part, because it is pretty good at predicting a player’s primary points rate. In fact, you’re better off predicting a player’s primary points rate with PSC60 than you are using that player’s own primary points rate. As Stimson notes in the image above, “points can be deceiving. Shot Contributions are more exact.”

We can also look at PSC60 in relation to the player’s involvement in all of their team’s offense while that skater is on the ice. This “offensive involvement” is expressed as PSC percentage. Here are last season’s results:

Dashboard 1(445)

*for a clearer view, click here to interact with the viz. You can select a single team’s players by clicking the team logo on the right side of the graph.

For forwards, their rate of shots/shot assists is highly related to their PSC percentage, as we would expect (r^2 = 0.65). The better the skater’s rate of PSCs, the greater their involvement in their team’s offense. An average forward last season produced nearly 22 PSCs per hour and was directly involved in about 40 percent of their team’s offense while on the ice.

You’ll also notice that the full view is very messy. Fifty percent of all forwards fall between 18-24 PSC60 and 36-43 percent offensive involvement, so there is a lot of overlap at the centre of the graph. However, some outliers do catch the eye even if this full league view.

At the strong end, Eichel, Crosby, Couture, Tarasenko, Spezza, Pacioretty, Gallagher, and Kadri stand out for contributing high PSC rates and being in as much as ~57 percent of their team’s offense while on-ice.

At the negative end, Rinaldo, Kulemin, and Stempniak stand out. Rinaldo is a good check to ensure that this stat really works — he contributed virtually nothing to Boston’s offense. Kulemin and Stempniak are bigger surprises as each have played offensive roles at times in their career.

Here’s the same graph for defensemen:

Dashboard 2(34)

*again, please follow the link to interact with the viz by selecting teams to filter the view.


Last year, the average defender contributed 14 PSCs per hour and was directly involved in ~26 percent of their team’s offense. Positive outliers include many of the usuals – Hedman, Karlsson, Letang, Muzzin, Doughty, and Faulk. Erik Johnson, Matt Dumba, Morgan Rielly, and the Nashville version of Seth Jones (passing data last year is separated by team if a player moved) all make appearances.

At the negative end (few shot assists + shots, low involvement in on-ice offensive activty), we find Mark Borowiecki, Nate Prosser, Nick Schultz, Nate Guenin, and Team Sweden’s Niklas Kronwall. Marek Zidlicky, Paul Martin, and Mark Pysyk surprise with their poor appearances here.


Bringing the pieces together

So, let’s sum up what we know:

  • primary shot contributions are good at predicting a player’s goals and primary assists (Stimson’s work)
  • shot assists are a repeatable skill.
  • Corsi is an important player stat and a player’s Corsi For percentage is repeatable.
  • individual Corsi for rates (iCF60) are very repeatable as well (shown below):

Dashboard 2(31)

With the knowledge that shot assists and individual Corsi are talent-driven, repeatable, and important, I’ve created a “playing style” viz to classify forwards and defensemen according to their usage. This idea is based on a neat article by Seth Partnow (now an analytics consultant with the Milwaukee Bucks) at Nylon Calculus.

Here’s a look at how forwards sorted out:

Dashboard 3(9)

*viz here. Click on team logos, as usual.

In this view, players are sorted into four categories based on the ways in which they contribute to their team’s offense:

  • Playmakers – above average shot assists rate, below average ‘shots’ (Corsi events) rate
    • featuring Henrik Sedin and Joe Thornton as quad representatives
  • Goal Scorers – above average shots contribution, below average shot assists
    • feat. Rick Nash, Alex Ovechkin, James Neal and Matt Beleskey
  • No Offense – below average in both measures
    • feat. Brandon Prust, Paul Gaustad, and Zac Rinaldo
  • All Around – above average shot rate and shot assists rate
    • feat. Max Pacioretty, Vladimir Tarasenko, Jason Spezza, Jeff Carter, Nazem Kadri, Logan Couture, Sidney Crosby and Brendan Gallagher

Beyond the theoretical justifications we’ve laid out so far, this view largely passes the smell test. “Passy” Sedin and Thornton are the game’s clearest examples of gifted passers. That shows here. Nash and Ovechkin are known shooters. No one (except the Leafs, apparently) expects any contributions from Prust.

The potential for quibbling is in the “all-around” category (which is what makes this worth discussing at all). It’s important to remember that we’re talking about rate stats here, which will elevate some players well beyond their standard counting stats. It’s also worth noting that shooting percentage and all its year-to-year variance is ignored here, paving the way for Nazem Kadri to stand out.

Here’s the same view with defensemen:

Dashboard 4(15)

*viz here.

While it’s best to dig into the viz for yourself and try some team-filtered views, I’ll note a few outliers here too.

  • Passers – Thomas Hickey, Jared Spurgeon, and Chris Tanev stand out most here.
  • Shooters – Johnny Boychuk, Darnell Nurse, and Justin Faulk are in this quad, joined by some surprising, lower TOI names.
  • No Offense – names we’ve heard before…Borowiecki, Guenin, Schultz, etc…
  • All Around – Erik Karlsson, Victor Hedman, and Kris Letang are here, joined by Jake Muzzin, Morgan Rielly, and Torey Krug.

*due to a mistype in the data set, note that Brent Burns is listed on the forwards chart. If here, he’d be off the chart in the shooters quad, posting ~19 shots/60 to go along with his ~5 ShA/60.


So, what are the useful applications of classifying playing styles?

In a grand sense, this information is another layer in understanding the many ways that a player is contributing on the ice. Confirmation that Thomas Hickey is a strong passer helps us to note the value he provides, even when his counting stats leave something to be desired.

Jakub Nakladal and Jake Gardiner in the “all around” category help us understand their above average contributions to team offense when on the ice as well. These types of insights can be key in uncovering players that have useful skill sets that aren’t necessarily obvious or appreciated.

Next, we’ll turn to a look at some rankings based on total usage scores. For now, please feel free to dig through any of the viz I’ve included by visiting here or dig through the latest release of Stimson’s passing project data here.


Using NHL passing data to uncover playing styles
To Top