Everyone’s favorite college hockey ranking scheme, KRACH, is also a favorite way to predict the odds of each team winning a game. I used KRACH to predict each tournament team’s odds of winning it’s first round game, advancing out of the regional, winning a semifinal game, and winning the national championship.
The full data is below, but here are a few interesting observations:
- Quinnipiac and St. Cloud St, facing auto-qualifiers, have the easiest paths out of their regions (each predicted over 50%) and to a national title.
- The Northeast is the most balanced region, with the four teams’ likelihoods of winning the region ranging from 14%-38%.
- The 2-3 matches are, as always, anyone’s game. The most favored #2 only has a 61%-39% advantage, while the tightest is 52%-48%.
- The top teams are strong — with KRACH predicting a 69% chance that a 1-seed will win it all, the highest such prediction since 2009.
- They play the games for a reason — the last three champs were 2015 Providence (1.82%), 2014 Union (15.61%), and 2013 Yale (3.17%). A Northeastern or a Duluth win wouldn’t be shocking, but a RIT or Ferris State probably would.
|Game 1||Game 2 (Region Champ)||Game 3 (Frozen four semifinal)||Game 4 (National Champ)|
|1. St Cloud St||86.03%||54.93%||33.56%||18.68%|
|4. Ferris St||13.97%||3.16%||0.68%||0.12%|
|3. Boston U||38.92%||13.96%||5.71%||2.03%|
|1. North Dakota||72.21%||48.93%||26.99%||14.61%|
|3. Notre Dame||47.53%||17.81%||6.48%||2.29%|
|4. Minn. Duluth||33.85%||14.09%||4.69%||1.67%|
|2. Boston College||59.41%||30.43%||13.51%||6.37%|
A few people on Twitter speculated that Quinnipiac might not be as good as their KRACH suggests (injuries plus a Spring slump), so I ran some numbers.
You may want to follow along with the real KRACH:
I then ran a KRACH including only games played in 2016. First, some teams KRACH rank and ratings:
1. Northeastern 2314
2. Boston College 843
3. Mass.-Lowell 740
4. Denver 739
6. St Cloud St 530
7. Yale 500
8. Boston U 489
9. Providence 457
10. Mich Tech 444
11. North Dakota 387
12. Michigan 373
13. Minn.-Duluth 368
14. Notre Dame 367
15. Harvard 337
20. Ferris St 196
48. RIT 27
Wow. That’s different at the top. Using this faux-KRACH for predictions would make an insanely dominant Northeastern an 86-14 favorite over North Dakota, with a 46% chance overall of winning the national championship. Quinnipiac’s chance of a national title would fall to 11%.
I really like this idea and find it interesting. I worry that the limited number of non-conference games in 2016 cause these numbers to be too biased by a limited number of games. Any idea how the Pairwise would look if you added a category for games in 2016? or more heavily weighted 2016 games compared to 2015 games in KRACH?
Yeah, the 2016-only is misleading in two ways–1) we cut out a lot of the inter-conference play that helps calibrate the relative strength of the conferences, 2) we’re down to a small enough number of games that some teams are near perfect, which which probably leads to an overestimation of their actual strength (does anyone really believe that if Northeastern and North Dakota played 10 times that NU would take 8-9 of those?)
I did it because people on twitter pointed out that Quinnipiac had a much stronger first half than second half, so may be overestimated by KRACH; it occurred to me that Northeastern was in the opposite situation.
PWR used to have a last 20 games criterion back when it also had a “teams under consideration”. Adding it now, given its likely strong correlation with RPI, probably won’t do much, but you can play around with adding it back in on CHNews’s PWR calculator on the “modify pairwise criteria” tab:
I forgot they had that capability to add it. I definitely remember the TUC and last 20 games. I’ll give it a shot but I bet your right that it won’t do much.