Bracketology 2016-17 Style

[KGR11]

KGR11, a 90% chance of beating RPI is extremely high--is that indeed the chance the model us?  

I read fivethirtyeight and similar blogs religiously.  I don't need to be lectured on probability/results-oriented thinking.  At the same time, I'm genuinely in the dark about this predictor and could definitely be wrong in this particular case.

Yeah, as of Feb. 23 our KRACH rating is 246.9 and RPI's is 28.6. The result is that our H2H "KRACH-modeled" record against RPI is 246.9-28.6, or an 89.6% winning percentage (this is my understanding of KRACH as defined by USCHO. Someone can correct me if I'm wrong). KRACH is based entirely on game results so it doesn't care HOW a team is winning (squeaking out like we often do, or blowing out like Harvard often does).

As a fun exercise I found the average goal differential of all Cornell and RPI games. Out of 60 games played, the average goal differential is 1.2 goals with a stand deviation of 2.15 goals. If you assume that ties happen when the goal differential is between -0.5 and 0.5 goals, we'd have a 21.5% chance of losing, 15.8% chance of tying, and a 62.7% chance of winning. Winning percentage then becomes 70.6%. Our Strength of schedules are pretty similar so I don't think you'd have to make any adjustments for that.

This feels a little drastic to me: that theoretical winning percentage against one of the worst teams is below our overall winning percentage against all teams. But it does take into account the goal differential of games by both teams.

[upprdeck]

how do the systems deal with goal differential as related to how the game was played?

team 1 is losing 3-2 and pulls the goalie with 4 min to go and loses 5-2 with 2 Empty net goals
team 2 is losing 3-2 and plays it out losing 2-1
team 3 is tied and loses 3-2 in 5 ot
team 4 is losing 3-0 all game and scores twice in the last 30 secs on a  5x3 major with the goalie pulled

can it tell who played better

[andyw2100]

how do the systems deal with goal differential as related to how the game was played?

team 1 is losing 3-2 and pulls the goalie with 4 min to go and loses 5-2 with 2 Empty net goals
team 2 is losing 3-2 and plays it out losing 2-1
team 3 is tied and loses 3-2 in 5 ot
team 4 is losing 3-0 all game and scores twice in the last 30 secs on a  5x3 major with the goalie pulled

can it tell who played better

Team 2 played better, obviously, as they were apparently able to have scored goals removed. (They just removed one too many.)  

[upprdeck]

yeah those dang replay reviews can really change a score

[BearLover]

You interpreted what I said as suggesting that you think the model gives Cornell a 100% chance of beating RPI, then you addressed your arguments to that position instead of my actual one. So I clarified my actual position with smaller words.

Then why did you make this point?:

I don't think you understand how this works, then. If the predictor didn't take into account that there's randomness in a single game of hockey, and weaker teams can beat stronger ones, our chance of making the tournament would be much higher than 85%.

I'm not lecturing you on your understanding of probability, I'm lecturing you on your reading comprehension.

Except you did then make a snide remark about results-oriented thinking:  

Anyway, as all bad poker players do, after the Cornell-RPI game is over, some people will take the results as "proof" their predictive analysis was correct, and none of them will be right.

And the "probability" part was in response to KGR11.

Anyway, as KGR11 just posted, the 90% likelihood of beating RPI seems way too high.  I stand by what I've posted in this thread.

[Beeeej]

You interpreted what I said as suggesting that you think the model gives Cornell a 100% chance of beating RPI, then you addressed your arguments to that position instead of my actual one. So I clarified my actual position with smaller words.

Then why did you make this point?:

I don't think you understand how this works, then. If the predictor didn't take into account that there's randomness in a single game of hockey, and weaker teams can beat stronger ones, our chance of making the tournament would be much higher than 85%.

Are there no percentages much higher than 85 other than 100?

I'm really tired of this conversation. I have no interest in continuing to defend things I didn't say. Let's just win the fucking games.

[ugarte]

Anyway, as KGR11 just posted, the 90% likelihood of beating RPI seems way too high.  I stand by what I've posted in this thread.

I agree with this.

The output of a formula doesn't need to be bowed down to if you think the inputs stink. KRACH is a useful number, of course, but there's a reason scouts still have jobs in the post-Moneyball era. I don't give a shit if recursive analysis of win/loss data says we're 90% likely to win. This team has enough fallow stretches that I wouldn't lay close to those odds and we play bad teams close a lot.

This, Cornell players, is your cue to make me look stupid.

[Jim Hyla]

So after reading through all this, I've come to the conclusion that we don't yet have a system that can correctly predict the results of all games, or at least not one that everyone agrees with.

Is that right?

Well glory be, how about that.

[Trotsky]

So after reading through all this, I've come to the conclusion that we don't yet have a system that can correctly predict the results of all games

Score with 1 second left is pretty good, though.

[Beeeej]

So after reading through all this, I've come to the conclusion that we don't yet have a system that can correctly predict the results of all games.

Good Lord, I certainly hope we never have that system.

[Hooking]

I like the part of the playoffs when a win gives Cornell a 100% chance of advancement. I best like the part of the playoffs where a Cornell win means the season's over.

[Beeeej]

I'm a vegan. And I don't own a television.

FYP.

[Trotsky]

I like the part of the playoffs when a win gives Cornell a 100% chance of advancement. I best like the part of the playoffs where a Cornell win means the season's over.

2008 wasn't that great.

[KenP]

I like the part of the playoffs when a win gives Cornell a 100% chance of advancement. I best like the part of the playoffs where a Cornell win means the season's over.

2008 wasn't that great.

touche.

[Hooking]

Mea culpa. I assume "when Cornell wins and the season's over" it's over for EVERY D1 college hockey team. I keep forgetting the prevailing mood of number crunchers.