I've been posting my own college hockey rankings on my blog since late January. Chris has been kind enough to afford me the opportunity to bring these rankings-- and other statistical meanderings I may indulge in -- to SBN College Hockey. Today, I present my new top 20 teams in the college hockey landscape.
First, let's go over how I calculate my rankings. I'm calling these rankings but I don't consider them a poll. A poll should include recent data, subjective opinion and scouting to place teams in a particular order. I'm not doing that. What I've done is created a ranking system that knows only the following about each team:
- Goals scored
- Goals allowed
- Shots on goal
- Shots allowed on goal
- Schedule
That's all this system knows. It doesn't take into account wins, losses or ties -- though it indirectly does because the only way to win games is outscoring opponents. It doesn't look at how a team has performed lately, nor does it know injuries or suspensions. I'll go over my methods a bit more below, but if that doesn't interest you you can jump straight to the table with my updated top 20 teams in America. Just note that there will be results that seem puzzling if you're looking strictly at wins and losses, so I'd recommend powering through what shouldn't be too much boring math.
The Math-y Details
I take this data and I've determined, using the method laid out by Tom Tango on determining how much skill and random variance is in a particular statistic. Tom Tango is perhaps the most prominent modern day baseball Sabermetrician and he's a consultant for numerous Major League Baseball teams, as well as doing some past consulting with NHL clubs. In short, he's a math wizard.
What I've found is that even with a full season of data, a team's shooting percentage for and against really only amounts to, at most, 50% skill and 50% variance -- most people in sports analysis call that 'luck' but I think that has a bad connotation. I'll call it variance. For instance, last season Minnesota finished second in shots on goal at 1,380. I then use that number to determine how much to regress their raw shooting percentage (10.1%). Turns out to be about 49% towards the Division 1 average (9.0%). This gives the Gophers a regressed shooting percentage of 9.5%. I do this for both goals/shots for and goals/shots allowed.
Using the new regressed goals scored and allowed I can create a Pythagorean winning percentage for each team. I use these first rankings to create a scheduling component, including a small home/road adjustment -- it's easier to win games at home, after all. This gives me a new set of Pythagorean winning percentages which allows me to re-run the scheduling adjustment. That gives me a new set of Pythagorean winning percentages and so on and so forth..
That's it. It might sound complicated but it's not. These crossover very well with the KRACH college hockey ratings carried by College Hockey News, though they do use wins and losses in that system as well, which means I do have some large differences in results when compared to KRACH's ratings. That said, the rankings between the two have about an 83% correlation.
Less Math-y Details
At it's core, I'm using regressed shooting rates to somewhat attempt to answer "what is the real talent level of this team?" I adjust for as much as I can -- schedule played, location of those games, skill and variance of the team -- and these are the results that I get.
The Ratings
Rk | LW | Change | Team | KRACH | Pyth% | SOS | SOS Rk |
---|---|---|---|---|---|---|---|
1 | 1 | 0 | Minnesota | 1 | 0.902 | 0.553 | 15 |
2 | 2 | 0 | Boston College | 3 | 0.864 | 0.566 | 9 |
3 | 3 | 0 | Vermont | 13 | 0.821 | 0.588 | 3 |
4 | 5 | 1 | Mass.-Lowell | 4 | 0.790 | 0.563 | 12 |
5 | 4 | -1 | Minnesota-Duluth | 20 | 0.772 | 0.566 | 10 |
6 | 7 | 1 | New Hampshire | 16 | 0.755 | 0.574 | 5 |
7 | 6 | -1 | Notre Dame | 9 | 0.749 | 0.564 | 11 |
8 | 8 | 0 | Quinnipiac | 7 | 0.741 | 0.505 | 36 |
9 | 9 | 0 | Maine | 24 | 0.735 | 0.577 | 4 |
10 | 11 | 1 | Providence | 10 | 0.717 | 0.554 | 14 |
11 | 10 | -1 | Michigan | 17 | 0.717 | 0.543 | 20 |
12 | 12 | 0 | Union | 2 | 0.707 | 0.500 | 40 |
13 | 13 | 0 | Massachusetts | 43 | 0.663 | 0.603 | 1 |
14 | 14 | 0 | Nebraska-Omaha | 27 | 0.647 | 0.546 | 18 |
15 | 16 | 1 | Miami | 31 | 0.637 | 0.544 | 19 |
16 | 15 | -1 | Northeastern | 18 | 0.637 | 0.568 | 8 |
17 | 17 | 0 | Merrimack | 46 | 0.630 | 0.590 | 2 |
18 | 18 | 0 | Ohio State | 22 | 0.629 | 0.526 | 31 |
19 | 19 | 0 | St. Cloud State | 8 | 0.613 | 0.536 | 24 |
20 | 20 | 0 | Colgate | 12 | 0.609 | 0.534 | 25 |
Note: LW is last week’s ranking. Change is the difference between last week’s ranking and this week. KRACH is a team’s ranking in KRACH ratings, found on College Hockey News. Pyth% is my Pythagorean Win Expectation. SOS is strength of schedule faced expressed in a Pyth% form. SOS Rk is a teams strength of schedule ranking.
As I stated above, this will differ from the conventional wisdom as my system doesn't include wins and losses. Yes, this top 20 doesn't include teams like Wisconsin and Ferris State who are highly rated clubs. Yes, it does include Merrimack and Massachusetts, two teams that finished in the basement of Hockey East. Let me walk through an example of how my system rates a team.
Massachusetts scored 76 goals this year and allowed 106. This gives them a -30 goal differential. The Minutemen put 1,053 shots on goal and allowed 1,025, meaning they scored on 7.2% of shots and allowed goals on 10.3% of their opponents shots on net. So far, I've found a team with shot totals that Massachusetts have should be regressed about 56% towards the national average offensively and about 46% defensively. This would put their regressed shooting percentage for and against at 8.2% and 9.7%. That doesn't seem like a big deal, but applied to their shot totals and Massachusetts regressed goals scored and goals allowed totals become 86 and 100 for a -14 regressed goal differential. That's pretty substantial.
Their raw Pyth% at this point is a 0.411 team -- meaning that they'd win around 41% of games played against an average team in my system. I use their schedule to figure out the strength of the teams they played, and they played a tough schedule. I adjust their raw Pyth% for the schedule. When I'm done re-running the ratings, Massachusetts' schedule comes out as the toughest in the nation. Their opponents Pyth% comes out at 0.603 which, in my system, is on par with playing a top 20 team every single night.
Why?
If I were filling out a ballot for the top 20 teams in the country, Merrimack and Massachusetts would not make my ballot. But what my system implies is that they played better than their results and did it against an incredibly tough schedule which boosts their overall rating. However, that's not my aim with this ranking system. When we watch sports, we're seeing a varying degree of skill and variance or luck or good fortune or something else of that ilk. It depends on the sport. Tom Tango has surmised that an 82-game NHL season is equivalent to 69 games of a MLB season. That means we know as much about a baseball team after 69 games as we do about an NHL team after a full regular season. In general, hockey is a tough sport to gauge.
The college game presents it's own set of challenges. The schedule is about half that of a full length regular season in the NHL and the talent spread in college is wider. Teams also play schedules with varying degrees of difficulties. The KRACH rating system cuts through a lot of that mathematically, and is very good. I still didn't want to take surface-level results into account. I wanted to attempt to account for everything I could, which meant shooting percentages for each team.
This system isn't perfect. In some cases, for many of you, it may be counter-intuitive to see teams like Ferris and Wisconsin outside the top 20 while Massachusetts and Merrimack resides in the back-half of it. It's all about the lens through which you're looking at the college game.