For some reason this bounced on Friday afternoon.  Note dated end line.
 
>Ralph Baer suggests using a 50-25-25 weighting in calculating the RPI ndex,
>rather than the 25-50-25 that is currently in use.  I whipped up a modfied
>version of my RPICH program to see what the results would look like, ad
>they are startling.  They look very close to the way many of us would ate
>the teams non-mathematically.  The results of this experiment should
>probably be brought to the attention of the Ice Hockey Committee and te
>league offices.
 
Fascinating.  This seems to offer empirical evidence to my theory regarding
the problem with this year's seedings: either something is wrong with the RPI
or it has been over-relied upon by the Committee.  Apparently something IS
wrong with the RPI.  So then, why does the RPI apparently work well when used
for seeding the basketball tournament, but fails when applied to hockey?
 
My hypothesis: in basketball greater weight for your opponent's strength is
necessary to determine the strength of your own team.  That is explained by the
very great variation in  strength between teams.  This becomes evident when you
look at the great disparity in strength between conferences.  For example, a
team in the Patriot League  and  a team in the ACC both sport identical 20-2
records.  Yet a win in the ACC and a win in the Patriot cannot be counted the
same when you compile a rating system attempting to show the relative strength
of all teams.  Obviously a team with a 20-2 record in the Patriot League is not
as good as a 20-2 team in the ACC.  In fact a team that was 10-10 in the ACC
would probably be stronger than a 20-2 team from the Patriot.  The current
RPI formula which allocates .25 weighting to the won/loss record then seems
logical for basketball-it's not how many wins you have, butwho those wins are
against.
 
But the formula breaks down when it is applied to hockey.  Why, because there
isn't as much inherent fluctation in strength from the best team to the worst.
College hockey (Division 1) has only four conferences.  While one may argue
that one is better than the other, or that one is always the stepsister to the
others, you can't say that the conferences and their member schools are
completely different strength-wise.  Results during the season bear this out:
an Ohio State ties a LSSU, a Princeton beats an RPI, a Michigan Tech nearly
steals a conference tournament championship.  The top four teams nationally
according to most polls include representatives from three of the four
conferences.  If you use the NCAA seeding, you come out with four teams, four
conferences.
 
Returning to the consideration of RPI I would suggest that a win in college
hockey should be considered to indicate more regarding a team's strength
than a win in college basketball.  Thus underweighting the value of the won/
loss record in the RPI for hockey IMO skewed the individual power rating for
each school, and ultimately resulted in an invalid seeding.
 
I'm not smart enough to propose whether Ralph's proposed weighting scheme is
right, or whether an alternative should be considered.  I do think however
that Ralph's inspired evaluation does show that the RPI needs to be changed
before next season, and that the faulty RPI was a major, if not the major,
factor in this year's bizarre seedings.
 
I would welcome any opinions from the stats mavens out there on the list.
                    _
            "NYS   // Hockey"
        Go 'Gate  //   Brian Morris
          Go RPI //      Albany, NY
          ______// [log in to unmask]
         (______/
************************On to St. Paul!*****************************