Well let's say you have a numerator and a denominator. the higher the numerator/ denominator ratio, the higher the score, so what I mean is you accumulate the number in the numerator by wins or whatever metric you choose. Anyway, making that number higher is what I mean by accumulating points. I guess it's a bad choice of words. What I am getting at is comparing teams via how they did vs a common opponent, let's say Clarkson for example, is really fraught. Clarkson is playing way better now than they were early, yet the numerator in the equation you get by beating them at different times is the same. It was WAY easier to beat them in october than now. Those two games are no where near the same.
In the same way if HE plays zero games vs the CCHA, then how on earth is the result of common opponent games played in vastly different time frames relevant. IMHO then, common opponent weighting is too high. If league A wants to have higher priority over league B when it comes to the post season, league A should play them head to head, at least some of the time.Otherwise I'm skeptical of the decision making process.
Hey, I'm open to any and all ideas. What I know is that the KRACH system can, in reverse, predict the record of every team in the nation considering who they played, and the 'odds' that the system gives them of winning the game. That's pretty good.
PWR I don't like as much because it don't think it's mathematically sound. There are too many 'fudge factors'. Oh, we want a bonus for beating highly rated teams? OK, we invent one. We find out that certain games, even if you win them, decrease your RPI rating. OK, we'll just randomly delete those games. Oh, and why is it 25/21/54. That's an odd percentage. The whole thing is too fabricated for me, although in the end it comes out close to KRACH, which is pretty good.
Now, if you are only talking about the common opponents part of the PWR, I say that.......almost every comparison anymore is determined by RPI alone. The number of odd comparisons each year is very small, so that the CommOpp component doesn't factor in very much.
However, overall, I think that this system places too much emphasis on the non-conference games, because that is the only way to 'sort' the teams. That's big drawback. If you didn't have a big mass of teams (that's a conference) called A, another one B....all insulated and then playing a few scattered teams from the other conferences, and instead had a much more random mix, it would be better. However, you can't really force schools to play other schools, and the reason for conferences anyway is to make scheduling easy.
You could make an edict like: No more than 24 conference games. That would help.
But when, like this year, the Big 10 is 45-15 against OOC competition, it's hard to know how to adjust for that, even if it's only going to end up being, on average 25% of the schedule.
You could also do something like, before anyone has played a game, create an allotment, like:
B10 - 4 teams
NCHC - 4 teams
ECAC - 2 teams
HEA - 4 teams
AHA - 1 team
CCHA - 1 team
But that's not really right either.
The fact is that it is impossible to create a system which accommodates both the fact that, as you say, playing Clarkson in November and playing them in February are two different things, without also including something like the 'eye test', which makes things subjective, and we don't want that.
