clock menu more-arrow no yes mobile

Filed under:

Hockey Analytics Workshop at CMU

On Saturday at Carnegie Mellon University in Pittsburgh, a hockey analytics workshop hosted by Andrew Thomas and Sam Ventura from war-on-ice.com featured a wide array of presenters to an audience with a varying degree of understanding as it relates to hockey analytics.

On Saturday at Carnegie Mellon University (CMU) in Pittsburgh, a hockey analytics workshop hosted by Andrew Thomas and Sam Ventura from war-on-ice.com / @war_on_ice took place featuring a wide array of presenters to an audience with a varying degree of understanding as it relates to hockey analytics.

Opening the conference was War on Ice co-founder Sam Ventura as he went over the survey responses on what the attendees thought their knowledge level was with hockey analytics. Asked why people wanted to come to the workshop, 47 said they wanted to better understand the game, 18 writer/want to include hockey analytics in writing (print, blogs, other media), and surprisingly (ok, not really), none said 'for the girls'.

Then the other half of War on Ice, Andrew Thomas continued to kickoff the day by asking who cares and asks about competitive analytics.. 'gamblers, ownership/front office personnel, coaches, players, and fans'.

Jesse Marshall, PensBlog and Faceoff Factor, started things for the guest speakers reviewing the basic terms and understanding today's hockey analytics such as Corsi Relative %, Fenwick For %, Offensive Zone Starts, Score Close, and PDO. Marshall showed a graph displaying the Penguins and where the players plotted in relation to Offensive Zone Start % and TOI Competition % with a special mention of my attendance in the audience and pointing out how bad Zach Sill is with such a high offensive zone start but a low TOI Competition %. It probably wouldn't be a smart thing to make short term investments based on PDO numbers as Marshall called it the equivalent of playing the stock market.

Sean Gentille, NHL writer for the Sporting News, talked about how journalists "have a responsibility to provide truth, fact based information to your readers" and finding the right mix of statistics to merge into an article. It was interesting to acknowledge the older generation of writers might find it more difficult to quickly accept and incorporate analytics as part of their coverage of the team.

James Santelli talked about the differences between baseball and hockey analytics and nailed the exact point that held back the numbers game being used in baseball due to communication between the guys analyzing the data and those on the field. Santelli said 'communication is key to take data to coaches and players'. He hopes in five years, like baseball with pitch-f/x, hockey will have comparable statistics like shot-f/x or save-f/x.

Matt Cane, puckplusplus.com, had a very interesting presentation about shot location as it relates to the handedness (right-hand shot on RW as on-hand vs right-handed shot on LW as off-hand, same applies for lefties). The statistics showed a player had a better shooting percentage from the off-hand than on-hand, 9.9% vs 8.6% respectively. It is important to note the difference could be due to a player's opportunity to use his one-timer or how a goalie to move in his crease going from side to side. It would be wonderful to see how shot location and rebound chances related to each other. Defensively,

Sam Ventura, war-on-ice.com, publicly for the first time unveiled his current work with Zone Transition Times (ZTT). It came about as he wondered what effect players like Rob Scuderi, who struggle moving the puck, had on the transition game for a team. Scientifically, Ventura used the idea of Markov chains, which is basically transitions between states. ZTT is measuring the time it takes to move the puck from one zone to another. It is a small sample size but for 2014-2015 season, teams with a good offfense and good defense as measured in ZTT were Pittsburgh, Montreal, St. Louis and Columbus.

Stephen Burtch, hockey analytics writer at Sportsnet.ca, explained Delta Corsi and keying on that Corsi For and Corsi Against are weakly correlated, Time on Ice (TOI) and zone starts are significant factors, faceoffs don't matter as much as people think, and age effects due to exist but hard to find. dCorsi 'represents the residual (differential) between a player’s Observed Corsi and their Expected Corsi resulting from the discussed regression.' Burtch believes this is a better measurement than Corsi Rel "because it is determined directly from contextual factors while the player is ON the ice (the OFF Ice results aren’t weighted equivalently to the other factors)."

Joe Walsh presented on hockey fight analytics, though it should be noted, he's not a fan of fighting and believes it shouldn't be a part of the game. Walsh's findings found no correlation with a team's total points and the amount of fights. The data used by Walsh came from hockeyfights.com, a little concerning some data in his discussion was based on voting on this website. It is funny that the numbers show Corey Perry as the second worst fighter of the 1,700+ as part of the analysis. The worst was former Penguins forward Tom Kostopolous.

Ventura returned to talk about more information will be coming to their website with models beyond just Corsi and Fenwick.

Jen Lute Costella, Analytics writer for Yahoo's PuckDaddy and only female presenter, freely admitted she's not one who creates the metrics but takes the information and merges the data to come up with a way to measure a team's likeliness to make the playoffs based on shot suppression. It was interesting to see the 2007-2008 Penguins were the lone Cup winner or Cup loser to not plot in a positive shot suppression.

Nilesh Shah, A Biostatistician and post-doctoral fellow at University of Pittsburgh, reviewed predicting playoff outcomes based on regular season data. His expected numbers are skewed a little bit based on having to consider the performance of non-playoff performances like a backup goalie, which is why his model didn't consider the Chicago Blackhawks as strong Cup contenders since Antti Raanta posted horrible save percentage numbers at even-strength.

Benjamin Zhang excited many hockey analytic followers with his ESPO - "Expected Shot Probability Outcome" based on chess scoring systems called ELO and Glicko. I'll need to do some additional research to understanding more of the information Zhang was providing.

Kody Van Rentergem, Robert Morris University hockey operations, had my favorite presentation explaining how the coaching staff reviewed video of every goal scored against them in a season tracking the data and they learned many of the goals were the result of losing their man one-on-one. This basic statistical tracking allowed the staff to make changes in their systems and by having the data to present to players backed up by film, it helped achieve the changes needed and RMU made good improvements.

Ryan Stimson, a contributor to the New Jersey Devils site 'In Lou We Trust', is working with others to track passing statistics from one zone to another or internally within a zone. This will be interesting to watch how these numbers translate over a period of time in greater detail as I believe strongly that players who can pass well or take passes will help teams transition from one zone to another. This information with the work that Ventura is now about to publicly unveil with ZTT would be wonderful development, especially since many of the old guard complained about rule changes after the year long lockout were going to dumb down the passing skills for defensemen.

The final presentation was by Jeremy Kanter of YinzCam, a mobile app developer, who explained if you want to grow the utilization of this information on mobile platforms like an App, then complex presentations on a web browser won't work on much smaller screens like an iPhone 6 plus.

To view the presentation materials, check out War on Ice's Blog for more information. Thank you to Andrew Thomas and Sam Ventura for their time putting together a great workshop.