By the numbers: Taking a look at Michigan's player data from Pick 224
I discovered Pick224.com, fell down a rabbit hole, and this is what I learned.
Welcome to Fresh Ice, a data-driven college hockey newsletter. Each week, I’ll be publishing fresh (no pun intended) content that focuses on educating and explaining all aspects of the game. Haven’t subscribed yet?
I fell down a rabbit hole yesterday.
I learned about Pick224.com, Dave MacPherson’s invaluable collection of data from every amateur hockey league under the sun, a few months ago but hadn’t spent much time looking at the data. Then I figured out how to pull just Michigan’s data, and it was game over for a data nerd like me.
I’m working on a larger project that’s coming soon and uses some of this data, but for now, I just want to talk about some of the things I found most interesting and enlightening.
As someone who was in the press box for 31 of the Wolverines’ 36 games this season, I like to think I have a pretty good sense of the team. I know the players pretty well, I have a general idea of what Mel Pearson is trying to do systems-wise, and I just generally know what’s going on.
But some of what I saw in this data caught me way off guard. Let’s dive in.
Even Strength Goals For %
For some background, this is a simple measure of, at even strength, what percentage of the goals that were scored while a player was on the ice were goals for his team. So, if player X was on the ice for two goals for and one goal against, his EVGF% is 66.67%. It is by no means a perfect metric — it doesn’t account for things like ice time or who the player was matched up against — but it can give a general, baseline understanding of how a team performs with a particular player on the ice. Even more useful is the relative version of this metric, which I’ll get into later.
In the NHL, with much, much, much more available data, there are a lot of super cool metrics that attempt to quantify play-driving ability and a player’s impact on his team. I’d highly recommend this breakdown from Charlie O’Connor of The Athletic for the most commonly-used metrics at the NHL level, but in college hockey, this is pretty much what we’ve got to work with.
Here are Michigan’s top five players in terms of EVGF%:
We can see that two of the top three are defensemen, Keaton Pehrson and Cam York. Pehrson particularly stands out because in most games, I didn’t notice him very much. That’s not necessarily a bad thing for a defenseman, because it usually means he’s just taking care of business on the back end and not getting into trouble. But still, for him to be second on the team in this metric was unexpected.
Pehrson was on the ice for 24 goals for and just 14 against, which is the fewest goals against of any of Michigan’s main six defensemen. For additional context, both York and Pehrson’s most common defensive partners, Luke Martin for York and Jack Summers for Pehrson, rank in the top ten. Martin came in seventh at 60.42% and Summers finished tenth at 58.97%. Martin was on the ice for an eye-popping 29 even-strength goals for, which is the most of anyone on the team.
Will Lockwood and Jake Slaker ranking highly is expected, as they were two of Michigan’s most effective and productive forwards. Lockwood was on for 21 goals for and just 11 against, while Slaker was on for 22 for and 14 against. They finished first and second on the team with 23 and 31 total points, respectively.
I mostly associate Dakota Raabe with the penalty kill, but he’s clearly impactful at even strength as well. What’s extra interesting here is that he mostly played on a line with Garrett Van Wyhe and Nolan Moyle, both of whom finished around 50% in EVGF%. I have line charts from 31 of the 36 games, and in 17 of those games Raabe played with Van Wyhe and Moyle. In the rest of the games, he was a healthy scratch a few times and bounced between a handful of different lines, including dressing as the extra forward three times.
Quite frankly, I don’t have an answer for why Raabe’s more successful in this metric than his most common linemates, but I would venture to guess that it came from who he played with when he was away from them. Another example of how more available data in college hockey would make analysis easier and better.
Now, the bottom five skaters in this metric:
Eric Ciccolini is probably partially harmed by small sample size here, as he had shoulder surgery in February that ended his season early and didn’t play a ton of minutes when he did dress, but he still played in 72.2% of Michigan’s games, so it’s still a fair comparison to include him on this list. He was on the ice for eight goals for and 12 against, leading to the worst EVGF% on the team. Ciccolini bounced around lines a lot — his most common line was only used for four games. It could be that a lack of time to build familiarity with his linemates led to less-than-stellar outcomes, or it could be something to do with who his lines were matched up against, or it could be any number of things. More data would really help answer this!
Mike Pastujov is also probably a little bit hampered by smaller sample size both in games played and time on ice, but the skaters he spent most time with — Jack Becker and brother Nick Pastujov — came in around 60%, so it begs the question of where the negative impact came from. Again, more data would be super helpful in figuring out this disparity.
Morgan and Hayhurst were each on the ice for 17 goals against — the most of any forward on the team. They actually were on lines together for seven games of the 31 I have line charts from, which is an interesting data point. It’s entirely possible that they had challenging matchups in those games, gave up a few (or more than a few) goals against, and that impact dragged them down for the rest of the season.
Nick Blankenburg’s appearance here is interesting, because the general consensus that I’ve seen is that he’s Michigan’s best two-way defenseman after Cam York. But in reality, he was on the ice for the fewest even-strength goals for of any of Michigan’s defensemen and tied for the second-most goals against. His partner, Griffin Luce, also didn’t grade out particularly well, which could suggest that this pairing either got the toughest matchups or just weren’t super effective as a pairing. Say it with me now: More data would be helpful to answer this question!
EVGF% (Relative)
This is where things get fun! Raw percentages are useful for some things, but as the analytics world has evolved and more data has become available, relative metrics have solved a lot of the problems present with raw percentages. Again, I’d highly recommend Charlie O’Connor’s breakdown for more information on this.
This metric shows the change in EVGF% between when a player is on the ice and when he’s not on the ice. Essentially, it measures the player’s impact relative to his team because it demonstrates the impact of when the player is on and off the ice.
Here are Michigan’s five highest-ranking skaters in this metric, with the note that I excluded Emil Öhrwall because he only appeared in 15 games:
For the most part, this data matches up with the raw percentages. Martin comes into the top five and York slips out, and the same is true of Jimmy Lambert and Slaker. Raabe moves up from fifth in raw percentage to lead the team in relative impact, which, again, I just find fascinating. I wish I had more answers in this piece than just repeatedly saying things are interesting, but this really is so interesting to me. When Dakota Raabe was on the ice, Michigan’s goals for percentage at even strength was almost nine percent higher than when he was on the bench. Again, this could be due to the competition he was matched up against or any number of other factors, but it’s still a fascinating data point and one that invites further investigation (if the data was available).
Pehrson continues to stand out in this metric, and it’s making me want to reevaluate my opinion on his performance this season. I like him a lot as both a player and a person, but I didn’t think he was necessarily the most impactful player. I saw him as a guy who did his job, stayed out of trouble and didn’t make waves, which is undeniably valuable but isn’t always going to show up in statistics. While I don’t want to overreact to just two numbers here, I can clearly see that he had a net positive impact on the team, and the positive impact was much larger than I expected.
I’m intrigued by York slipping a little in the relative metric and his partner, Luke Martin, coming into the top five. York still ranks third among defensemen and seventh on the team in the relative metric, so it’s not like his raw percentage completely misrepresented his performance, but it’s a good reminder of how raw numbers and relative numbers can differ.
Let’s look at Michigan’s bottom five skaters in this metric:
On the negative side of the chart, it’s even more similar to the bottom five skaters in raw percentage. The only one who doesn’t appear here is Moyle, who finished at -7. Ciccolini’s relative impact is even more dramatic than his raw percentage. Michigan’s goals for percentage at even strength was almost twenty percentage points lower with Ciccolini on the ice than with him on the bench. I do want to note that Ciccolini played most of the season through repeated shoulder dislocations, and that would almost certainly impact a player’s effectiveness, so I don’t want it to seem like I’m being unfair to him here. Still, he’s at the bottom — by a lot — in both versions of the metric, which I don’t think can be fully explained by playing through an injury.
Morgan’s placement here is interesting because he’s thought of as one of Michigan’s more dependable, two-way forwards. Like Raabe, he’s a terror on the penalty kill and flies under the radar somewhat at even strength. But unlike Raabe, his under-the-radar impact is negative, rather than positive. I do wonder if changing positions has any impact on him here, because he played center a fair amount early in the year before moving to his more natural position on the wing. Centers have more defensive responsibilities than wingers do, generally, so it’s possible that part of what makes him less comfortable or successful at center is that defensive workload.
Again, I’m intrigued by Blankenburg here. Blankenburg ranks last among Michigan’s defensemen in both EVGF% and EVGF% (Rel), and what an excellent case study in how just using the eye test to evaluate a player can miss a lot of information. I would love to have more data available to see if this is caused by how he’s used in games, the quality of his competition, his defensive partner, or any number of other things it could be.
Estimated Time On Ice
The thing that excited me the most when I learned about Pick 224 is the estimated time on ice calculation. eTOI was invented long before the NHL tracked and provided ice time data and it predates essentially all of the analytics we use today. Here’s a good explainer of one of the first eTOI calculations.
In college hockey, like in most other amateur leagues I’m aware of, teams track time on ice but don’t provide it publicly. This forces people like myself to guess at how many minutes a skater played and leads to people creating eTOI calculations, which estimate a player’s ice time based on their contributions to events. Dave MacPherson uses one on his site, and I was so excited to see it.
Just for a quick conversation, let’s look at Michigan’s top three forwards and top two defensemen in eTOI.
Freshman Johnny Beecher just barely edges senior Jake Slaker for the lead among forwards, while freshman Cam York far and away leads the team in ice time. Again, to be clear, these numbers are estimates based on a calculation, but they fit with what I felt like was happening live. All three of these forwards leading their group in ice time makes sense given their production and also that all three play on both the penalty kill and the power play. I would love to take a look at PP and PK time as compared to time at even strength, but, again, that would require ice time data being public.
York played 27 minutes in his first college game — and he sprained his ankle partway through. It’s of no surprise, then, that he continued to play heavy minutes the rest of the way. He killed penalties, played on the top pair at even strength and ran the first power play unit. He was, in essence, the prototypical No. 1 defenseman. Blankenburg just barely beats Martin (20.84) in this calculation, despite playing on both the PP and the PK while Martin only played on the PK.
I don’t necessarily think there’s a ton of insight to gain from eTOI on its own, but I really liked that it was available and thought it was super interesting to see if it matched up with my impressions of Michigan’s usage. Overall, it did.
A final rant about data
A lot of what I said here can be explained by having more data available. Pick 224 has just about all of the available data for college hockey, and there’s still just not very much. To be quite frank, I’d work in college hockey forever if it was an option, so I would love to see the amount of data grow and the level of analysis increase. There are still plenty of people who do really cool things with college data, but it requires a lot more work than similar projects would at the NHL level.
Regardless, I thought it was fun to take a look at some different metrics than I typically look at and dive down the rabbit hole of data. Many thanks again to Pick224.com, which is a seriously awesome resource that I would highly recommend everyone investigate.
I hope you enjoyed this issue of Fresh Ice! It was a little different from what I’ve done before with my penalty kill series, but I really enjoyed just digging into the data and talking about what I thought was cool. Stuff like this is how I got interested in the data and analytics side of the game, and I hope it can do that for you too!
Still haven’t subscribed?