What we are really discussing here is the relative weighting of the
factors involved in the RPI and why they are what they are.  This is
something I have been interested in for a while.
 
Ralph says,
>I agree with John that 1/3, 1/3, 1/3 seems better than the .25, .5, .25
>weighting that is used in the RPI calculation.
 
But why?  I don't necessarily disagree with either the current RPI or
any other revised format...but I want to see strong reasoning behind
any rating that people come up with.  What would make (1/3)*3 better
than the current weighting?   Similarly, why should the NC$$'s RPI
have weights of .25-.50-.25 as opposed to anything else?  I am
inclined to believe that the NC$$ itself does not really know the
answer to this either.
 
John said,
>Second comment: I prefer 1/3, 1/3, 1/3 to .25, .50, .25. :-)
>If anyone is interested in number crunching and has plenty
>of time on their hands, I'd like to see .50, .25, .25 and .40, .30,
>.30 as well.
 
Sounds like you're going to look at various weightings to try to
determine which one you think is the best - not a bad idea.  But
here's the brick wall I run up against whenever I think of doing this:
what in heck are you going to use to compare it with?  One person may
look at the output of several different weighting and say, "*This* one
makes sense because it has TeamA above TeamB," yet this is really a
value judgment based on one's own personal opinion that TeamA *is*
better than TeamB (in other words, you didn't need a rating to tell
you that).
 
And of course, another person may think the rating is hogwash because
he/she thinks that in actuality, TeamB is better than TeamA.
 
One of the many things I liked about TCHCR was that it had what I
considered to be a strong foundation for its method such that without
even seeing any results, I could say yes, this makes sense.  That is
what I would like to see done with other ratings, RPI/RPICH included.
 
Dick writes:
>        Basically the idea is that the won-loss percentage should be
>multiplied by a "degree of difficulty" to obtain a corrected won-loss
>percentage.   How is the degree of difficulty to be calculated?   Well,
>the most obvious way would be the opponents won-loss record.
 
Actually, my understanding is that this is not the case at all in the
RPI (that win% should be multiplied by a degree of difficulty to
obtain a corrected win%).  I'm not aware that any attempt is being
made to create a corrected win%.  Rather, factors such as win% and
opponents win% have been determined to be factors that the committee
wishes to look at, and the RPI gives them some combination of these
factors that is supposed to be most fair.
 
In the past, the committee has discussed "strength of schedule" as if
it were a completely different entity from win%, and in fact win% and
SOS were considered two separate and completely unrelated factors in
choosing teams as recently as a few years ago.  While Dick's method
may make some sense (again, more reasoning and info would be helpful),
I still believe that the committee wishes to keep these factors
separate (thus not counting games played by the team in question in
OppWin%); they just wish to determine some relationship between them,
and that is why the RPI exists.
 
So, let me suggest that we keep separate the two issues that have
arisen here: 1) should the RPI have different weightings; 2) are there
other ways of rating teams that are valid and how can we determine
their validity.
 
BTW, I'll note that I like to see RPICH the most because it is "the
only rating or poll that really matters".  But I am also curious to
see what other people come up with.  I just ask that you support it
with a well thought out explanation and description of how the numbers
were crunched.  I don't think it helps to see a list of teams and
numbers without knowing how that list was created.  It isn't any
better than a poll in that case, and in many ways it can be worse -
and the goal is supposed to be to build something that eliminates the
regional and emotional biases of polls.
---                                                                   ---
Mike Machnik                                            [log in to unmask]
Cabletron Systems, Inc.                                    *HMM* 11/13/93