>
> FROM:  Shawn Walsh
>
> RE:  College Hockey Computer Rating
>
> I guess the best way to put in perspective the skewedness of the College
> Hockey Computer Rating is the fact that in the most recent rating
> (2/15/92) eight of the CCHA teams are in the top twenty in the nation!
> As well, eight of the WCHA are in the top twenty-two in the nation!
> With all due respect to Miami, Illinois, Bowling Green, Ferris State,
> Colorado College, St.  Cloud, etc., it is hard to consider them better
> than New Hampshire, Clarkson, and some of the other teams they are ahead
> of.  Because of the skewedness of the strength of schedule, then
> supposedly eight of the very best 20 programs in the United States this
> year are located in the CCHA.  We certainly can't justify this because
> the CCHA doesn't even have a winning record against HOCKEY EAST, but
> somehow this rating system has set this up.
>
I have to agree with Coach Walsh...these ratings look out of whack
(adding another opinion to a spirited debate).  It has been
years since I worked with models remotely close to the complexity
of TCHCR, but it appears that too much weight is being given to
losing to other "good" teams -  presumeably by relatively small margins
(leading to relatively high rankings for teams w/ bad records like BG).
How were the weights for things like margin of victory/loss, site of game,
etc. determined?
 
I suspect also that the problem is exacerbated by the previously
discussed scarcity of interleague games.  In the extreme, suppose that
only one team in a league played a non-league game and that it beat the
"#1 team" in the country.  Suppose further that every game in that
league ended in a tie.  One could easily envision a weighting scheme
that ranked every team in that league as being tied for the new #1
ranking.  Now suppose that instead of tying all its games, one team lost
every game by one goal.  I'm sure I could come up with a set of weights
that still placed that team near the top of the charts.  Perhaps it
should be, perhaps not (life's a puzzler ain't it?).............
 
It is popular to laugh at subjective rankings such as polls in favor of
"objective" measures such as computer models.  The truth is that more
often than not (always?) the model just reflects the subjective biases of its
creator via the weightings it assigns to various events.
 
Having said all that, I admit that I know little about the internal
workings of this particular model.......I'm sure many people will
enlighten me if I'm off base...
 
Jon Greene
[log in to unmask]
RPI '82, Cornell '84