I spent last Saturday at the first Boston Hockey Analytics conference, a gathering of analytically inclined hockey faithful. For those unable to attend, here are a few highlights. Note that, to the best of my knowledge there was no audio or visual recording of this conference.
–Michael Schuckers gave a talk summarizing the state of goalie research, including material that he’s working on for an upcoming book chapter. For the unfamiliar, the most common metric in evaluating NHL goaltenders is save percentage, which is limited in part because different goaltenders face different distributions of shots over the course of a season. Indeed, you could even have a Simpson’s Paradox scenario, where Goalie A is better at saving each type of shot than Goalie B, but that Goalie B still ends up with a better save percentage overall. This upcoming book chapter will be a must-read.
-Schuckers also pushed for those in the audience to do whatever necessary to get the NHL to share its tracking data. IMO, this is a no-brainer. The world of basketball is better for the brief look into this rich information that the NBA shared during the 2014-15 season and parts of the 2015-16 one. See, among other examples, this excellent tutorial on how to scrape and analyze player movements. The NHL’s lagging behind, and given the well known flaws that the league has with scorer/rink biases, the potential is there for public analysts to answer some excellent questions and help grow the game.
-Rob Vollman gave a talk on roster construction, providing a glimpse into how rules of the CBA dictate who and what players are reasonable values. You can buy Rob’s book here, which presents this and other analytically driven research. The takeaway linking Rob’s and Schuckers’ talks: don’t give goalies massive contracts, as there’s too good of a chance they won’t be worth it.
-I gave a talk on how to use R for reproducible hockey research. Slides and code here (note: download the pdf of the slides if you are looking for links). There are very few hockey researchers who share both their code and data. It’d be better for everyone involved if we can change this.
-Cole Anderson presented work on an ELO-based player comparison tool, in which the hockey public can rank players. This makes sense, particularly given that traditional player rankings (say, a scale of 1-10) can lead to ambiguous numbers (like 7.7). Cole’s work appears similar to the surveys that 538 has used to, for example, rank James Bond villains or pick summer Olympic sports. Cole’s code is also in R, and available for your perusal here. Hope to see more out of this project.
-Eric Cantor gave a talk looking at roster construction, looking at the Tampa Bay Lightning’s experience with seven active defensemen in place of the usual six. Eric’s evidence suggests that Tampa performed slightly better with the non-traditional construction.
-Ryan Davenport and Edwin Niederberger looked at shot locations from world tournaments, including the recent World Cup and the Olympic seasons. Relative to the rest of the world, the USA’s backline looks particularly ineffective offensively.
-Brian Carothers and Joseph Nelson gave two talks. First, the pair looked at quantifying defensemen given their hit and blocked shot totals, the slides of which are found here. Second, the pair led a Python workshop, the overview of which is linked here and the code of which is found here. Python and R are both free and powerful. You should learn (at least) one of them.
-Rob was asked how many teams are doing appropriate due diligence with respect to analytics? Rob guessed six, with most teams “nowhere close.”
-Billy Jaffe (New England Sports Network), Neil Abbott (player agent), and Ron Rolston (coach) led a panel moderated by Babson’s Rick Cleary which summarized their perspectives on how analytics have changed the game. Perhaps unsurprisingly, their biggest take-home is that these practitioners don’t care about what model you may have chosen or what (statistical) tools you needed to employ, they just want immediate, actionable, and simplified recommendations. This begs the question – that wasn’t asked – what happens when those suggestions don’t match their prior viewpoints?
-Although each panelist likely loses more hockey knowledge in their sleep than I’ll ever learn, there was a bit too much selecting on the dependent variable for my taste. In other words, because Team X and Team Y have recently won the Stanley Cup, this is how all teams need to win the Stanley Cup. Hockey’s way to random for that.
-Perhaps given that the conference was held in Boston, the Bruins’ 2011 Cup winning team was held in particularly high regard. Two of the panelists, for example, praised the Bruins’ winning culture and development of a high character locker room as driver’s behind their success. Of course, if that Boston team was so good at hockey, why did it need seven games – and an overtime – just to get out of the first round? If Montreal had won that round one series instead of Boston, did Boston still have a winning culture and a high character locker room?
-I missed a few other talks, but if those researchers or anyone else wants to share materials, please send them along! And many thanks to Luke, Rick, George, Michael, Rob, and the rest of the organizers for their great work.
StatsbyLopez 