Re: 2019 NCAA Tournament Thread - Regionals are the best weekend of hockey all year
Really..? Make a cogent argument to the contrary.
I'm waiting, but not expecting much, given your history.
Lump yourself into the same mass of humanity that has no idea how a statistical model works, and go tee it up, or whatever.
Fishman,
I won't answer for the other poster here, but I would like to engage this conversation a bit more.....
These are my assumptions:
1- The NCAA (coaches, ADs) desires a system which is objective rather than subjective, since subjective led, in past years, to supposition of favoritism toward name brand schools
2- Any subjective system is going to be based on math in some way
3- Since it's entirely subjective, the math needs to be somewhat easy to understand.
I would argue, actually, that the current PWR is too complicated, but that's another discussion. For now, I think it's safe to say the above are true.
Then, we have the question:
What makes a predictive system differ from a system which merely evaluates the prior results?
This is a good question. Obviously, a predictive system must have some basis in data, or it can't predict anything. In this way, of course a predictive system is based on results. It can't be based on anything else (pulling rabbits out of a hat?).
I would suggest, however, that predictive systems likely make use of a far greater collection of information....for example, although I have not looked at any other system other than the 2 most commonly discussed here, I would assume that other systems use things like:
Margin of victory
What the opponents record (or RPI or whatever) was in the day you played them
Injuries to key players
Goalie rotation
ENGs
etc.
There are 2 issues with such systems, to me.
1- There is no way of knowing exactly how to weight all of those inputs, so you are guessing, in some sort of way
2- There is no way of knowing how good the predictions are
For these reasons, I would personally prefer a system which uses only KRACH, adjusted for home ice. That particular system does have this one advantage:
Given the rankings, one can replay the entire year and get the records of all of the teams correct. That is a very powerful argument.
Of course, there are issues there as well. The nonconference season weighs too heavily in some sense, because 8-10 games on average for each team decides a range for an entire conference to fit into....
So, it's not perfect. But, to me, it's better than anything we've got going.
Would you like to flesh out a predictive system which you prefer?
