KRACH predicts the NCAA tournament

[jimdahl]

I've been doing this for about 5 years now and it's always fun...

KRACH predicts the NCAA tournament

But this year it's downright crazy. We've never seen the PWR-based tournament seeding and KRACH deviate so much.

* UND’s predicted 53% chance of winning the opening round game is lower than any 1 seed in the previous 4 years I’ve been doing this

* Union is even worse off, not even favored to win its 1 vs 4 matchup

* Similarly, 2 seed Minnesota is an underdog to 3 seed Boston University

* The largest line in any 2 vs 3 matchup is 54%-46%, Miami over Mass.-Lowell

[Fetch]

Did they just pick them out of a hat ?

[luapsided]

hmmm why even play? Just give Boston the trophy already. :/

[GeauxSioux]

hmmm why even play? Just give Boston the trophy already. :/

Kind of what it looked like for the Sioux last year, but look what happened.

[jodcon]

Is it KRACH or on crack?

Hard to tell.

[Snake]

KRACH was not on crack; the PWR was. The teams and seedings were messed up by the PWR.

[jodcon]

KRACH was not on crack; the PWR was. The teams and seedings were messed up by the PWR.

You might be right, I guess we'll know in a couple days.

[sagard]

KRACH does a nice job of putting numbers to it. I know if the Gophers beat BU, UND, BC, then Michigan in route to the title they will have earned it.

[The Sicatoka]

If the autobids and KRACH were used to seed the tournament (and not accounting for conference match-ups, etc.):

BC v. USAFA

Maine v. BU

Mich v. Union

WMU v. Minnesota

UND v. NMU

UMD v. DU

Miami v. Mich St

FSU v. UMLowell

Better?

[jimdahl]

KRACH does a nice job of putting numbers to it. I know if the Gophers beat BU, UND, BC, then Michigan in route to the title they will have earned it.

Yep, region of death. Minnesota's chances of winning it all, per KRACH, are lower than all the 3 seeds (BU, Maine, Mass.-Lowell, and Denver). Only Air Force, Union (a 1-seed!), and Cornell stand a lower chance of winning it all.

[Snake]

Yep, it all depends on whether or not you feel KRACH or PWR does a better job of choosing a representative field of 16. I think more "numbers guys" lean toward KRACH as it is completely objective.

[jimdahl]

In the KRACH vs. PWR comparison, there's no more stand-out difference than Union this year. #3 and a strong 1-seed in PWR, #15 and not even favored to win its 1 vs. 4 opener by KRACH (though just barely).

Unfortuantely, a single sample won't really tell us much, but you can bet a lot of people armed with KRACH bet against Union in their office pools.

[BigGreyAnt41]

Yep, it all depends on whether or not you feel KRACH or PWR does a better job of choosing a representative field of 16. I think more "numbers guys" lean toward KRACH as it is completely objective.

How is the PWR not objective? It's completely numbers. They just choose different numbers.

[Dave Berger]

Yep, it all depends on whether or not you feel KRACH or PWR does a better job of choosing a representative field of 16. I think more "numbers guys" lean toward KRACH as it is completely objective.

Yes. The following was posted by "Rich" in one of the USCHO forums regarding the ranking systems, and gives a point-by-point argument for embracing KRACH...

1. If a team wins a game, they always increase their rating in KRACH. PWR can drop a team for winning, and it can happen in RPI too.

2. A team's rating always increases more for beating a higher-ranked team than beating a lower-ranked team. That's not always true for PWR.

3. A team's KRACH rating always drops if they lose, and by more if they lose to a lower-rated team. Losing to the No. 1 team cannot increase a team's KRACH rating. If KRACH rewarded SoS somehow, then this wouldn't be true. KRACH does not reward losses to good teams. Period.

4. Tying a better team increases a team's KRACH rating, and tying a worse team decreases it. This makes perfect sense, but is not always true with PWR.

5. A .500 team's KRACH rating is exactly equal to their strength of schedule.

6. If two teams have the same record, the team who has played the toughest schedule is always ranked higher. If two teams have played the same schedule, the team with the better record is ranked higher.

7. If a hypothetical team goes .500 against teams in a given place in the rankings (say, 1-5), they will be ranked within that group (assuming those are the only games that team has played). Neither RPI nor PWR guarantee this.

And most importantly:

8. Given every team's KRACH ratings and the schedule they have played, you can calculate the team's record exactly! If the strength of schedule component was exaggerated in any way, this wouldn't work. This is the most convincing evidence that SoS is properly weighted in KRACH. You sure can't say that about PWR or RPI!

If you look at all of those reasons, you can see that they are exactly what one would hope a rating system would do. I find it amazing that you can take all of the given KRACH ratings and the schedule any given team has played and work out what their winning percentage is. Well, amazing until I understood how it's calculated.

I hope this helps.

Dave

[Dave Berger]

How is the PWR not objective? It's completely numbers. They just choose different numbers.

It's objective, yes, but for the reasons I've summarized above, the numbers don't always add up. For example, if Team A has a 4-0 record against Team B and meets Team B in a best-of-3 playoff format, Team A may not want to sweep Team B if Team B's record is perliously close to the Teams Under Consideration "cliff". If Team A wins Game 1 , Team A may benefit from losing Game 2 and then winning Game 3. Any system which rewards a team for losing a game has serious flaws.

Furthermore, Head-to-Head and Record vs. Common Opponents are often restricted to a game or two throughout the whole season.

I would argue for using KRACH to rank the teams (or RPI, if you're held to that, but KRACH is better mathematically and does a better job, imho, of handling strength-of-schedule issues) and saving the other critera (Head-to-head and Common Opponents) if the KRACH is very close (say, within 5 points...)

I want to say for the record that I like the fact that the Pairwise is objective, and that's a good thing, but there are more mathematically sound ways to be objective.

Dave

[Snake]

It's objective, yes, but for the reasons I've summarized above, the numbers don't always add up. For example, if Team A has a 4-0 record against Team B and meets Team B in a best-of-3 playoff format, Team A may not want to sweep Team B if Team B's record is perliously close to the Teams Under Consideration "cliff". If Team A wins Game 1 , Team A may benefit from losing Game 2 and then winning Game 3. Any system which rewards a team for losing a game has serious flaws.

Furthermore, Head-to-Head and Record vs. Common Opponents are often restricted to a game or two throughout the whole season.

I would argue for using KRACH to rank the teams (or RPI, if you're held to that, but KRACH is better mathematically and does a better job, imho, of handling strength-of-schedule issues) and saving the other critera (Head-to-head and Common Opponents) if the KRACH is very close (say, within 5 points...)

I want to say for the record that I like the fact that the Pairwise is objective, and that's a good thing, but there are more mathematically sound ways to be objective.

Dave

How is the TUC "cliff" determined? Does someone set it at 0.500, and therefore make it non-objective? KRACH considers all games against all teams.