College hockey is not often publicly tracked, so this project came about from a desire to do something different. During the 2016-17 season, I ripped a bunch of Boston University men’s hockey games from the site they were being streamed on at the time. I figured it could be useful to have the video and I could find something interesting to do with it. Initially, I wanted to track entire games and get a spreadsheet up and running of everyone’s individual stats, as well as team ones. Long story short, I didn’t have the time to devote to that, so I narrowed my interest down to the power play. My goal is simply to quantify, illustrate, and analyze the BU men’s hockey team’s top two power play units from the 2016-17 season. This post will mostly serve as an intro to this project as well as a first look at what I’ve been doing and will be doing.
I have defined the top two units as the two most frequently used groupings of players (in the same formation) in the games that I had access to. That boiled down to a top unit of Charlie McAvoy (7), Clayton Keller (19), Jordan Greenway (18), Jakob Forsbacka Karlsson (23), and Patrick Harper (21) [4F1D, 1-3-1/umbrella-ish something formation they kept moving around] and a secondary unit of Bobo Carpenter (14), Kieffer Bellows (9), Dante Fabbro (17), Chad Krys (5), and Patrick Curry (11) [3F2D, split formation].
For those unfamiliar with power play formations, that means the two groups typically looked something like this:
One limitation of this project is sample size. With 31 power plays to look at over seven games for PP1, and 16 to comb through over five games for PP2, it seemed like I’d have a decent number of shot attempts to work with, and that’s mostly true. What I have can give a nice glimpse into the power play and shooting tendencies, but no one player had more than 15 shot attempts over the course of these games, making some individual charts look bare and patternless. I would have more, but not every game had video available for the public, or it wasn’t available for free, so I took as much as I could get. It should be enough for a snapshot of what the Terriers’ “ideal” setup was before tweaking it in the last handful of games of the season. There’s no shortage of passes, though, so that post will have fuller visuals to look at.
Another limitation is that the majority of the games I could snag were BU home games. This is a bit concerning considering the Terriers were 13-3-2 at home versus 8-6-1 on the road, but after calculating home-road splits for the power play itself, I found that BU operated with a 21.3% success rate at home (16-75) and an 18.6% success rate on the road (13-70), which is not a statistically significant difference at the α = 0.05 significance level. So, on the season, BU didn’t fare significantly better in their own rink versus others, despite the bump in percentage. Overall, their season power play conversion rate sat at 19.4% (33-170).
When it came to actual tracking, Ryan Stimson gave me access to a tracking tool he’s used with the Rochester Institute of Technology’s hockey team in the past that had been built by an RIT grad student. It allows the user to click where shots and passes occurred and then gives x and y coordinates of those events as well as shot distance, pass distance, shot angle, and more. I’m so grateful to Ryan for this because without it, I was going to track everything by hand and somehow digitize it. This way it’s much easier to work with and manipulate in Python. The image of the rink was also part of this tracker, and I have incorporated it into my visuals because, to be honest with you, I don’t know the exact measurements enough to draw the rink in Python. I can do the blue line and two red lines, but the others are tougher to align.
Now, to the specifics of this post! Because of all the introductions I had to get out of the way here, I’m only going to look at a very basic overview of shot locations in this post. The next one will be more player-specific and will even incorporate some NHL data (you can probably guess for who).
I just used matplotlib for these with the x,y coordinates of the shot attempts. Nothing is specified here in terms of whether the shot hit the net, went in, missed, or was blocked, and nothing is specified in terms of shot type either. Again, this is just a very basic look to get things started.
PP1 and PP2 shot locations
The above scatterplot indicates where each shot attempt from both units was taken. There’s a clear net-front presence here, which you might have guessed earlier looking at the formation diagrams. With Greenway and Curry stationed at the top of the crease at almost all times (Greenway had more leeway with where he could set up), there was usually a body making things difficult for the opposing goaltender. Not a lot of shot concentration in the high slot, and it looks like that’s the only real hole in the inner part of the ice, which is interesting, but understandable for the second unit specifically. Krys and Fabbro rarely ventured far past the blue line, which will be clear when I post the passing plots. For the first unit, Keller operated as the quarterback and would usually pass the puck around from that spot rather than shoot, so it makes sense that that area is a bit sparse as well. That one little blue dot in the corner there is a Carpenter shot that was not captured by the broadcast, but I wanted to credit him with a shot so it’s hanging out there for now.
Here we get a better look at PP1 itself. It’s a little messy at first glance, and will likely still be a bit messy when I go and differentiate between players, but the general shape is there. The biggest thing with this power play was that, even though everyone had their set roles, the players (especially Keller) were not afraid to slide around to other parts of the ice. It sometimes makes it hard to tell exactly what shape they’re going for, but they usually return to their “places”.
This is just very clearly a mostly stationary power play. Again, not a huge sample size so it looks a little empty, but the individual clumps are very distinct and show literally zero slot presence. PP2 was given a formation and instructed to stick to it, so much so that you can probably guess every shot without me even identifying which belongs to who. My personal favorite is that little Carpenter clump in the left circle there.
Now that we’ve gotten the basic shapes and locations out of the way, it’s time to get more specific with the next post.