>>Lets suppose team A and team B finish 12 and 13 respectively in the
>>PWR winning 10 and 9 PWR comparisons respectively.
 
>>Now lets suppose that if one looks at the individual PWR comparison
>>between A and B that B beats A
 
>>Who goes to the tournament?
 
>B
 
        To be totally pedantic, there could be other teams in the mix
that change that outcome; if A wins one more comparison than B, but B
wins the comparison with A, A must win *two* more comparisons than B
with all the *other* teams.  If those comparisons involve other teams
on the bubble, it could complicate matters.
 
        But yes, all else being equal, team B would go to the
tournament in the scenario described above.
 
[...]
 
>They get an idea going by RPI -- of who is strong .... and most of the
>time, you can choose the same top teams through RPI and PWR ---- for the
>bubble teams, they check each team against each other that are sort of
>close together ...
 
        Okay, you've said this before, that they identify the bubble
by looking at the RPI rather than the PWR, but going by your Joe Marsh
interview from last year
<http://www.uscollegehockey.com/tournament/032097.html>, that doesn't
appear to be true.  I quote:
 
" The other teams are determined by looking at the Teams Under
" Consideration, and comparing them to each other using the criteria
" outlined above. Schools like Minnesota, Miami, New Hampshire and
" Vermont were obvious choices once the committee could easily see those
" teams were beating all the rest under consideration.
"
" The committee then took all the "bubble teams," like Michigan State,
" St. Cloud, Denver, RPI, Princeton and Colorado College, and compared
" them all to each other.
 
        It sounds to me that they identify the "lock" teams, and hence
the "bubble" ones by using the PWR--or more precisely the comparisons
among teams not receiving automatic bids--and *not* by the RPI.
 
>Now -- in essence, they are doing the same thing
 
>It's just not as methodical as PWR.....
 
>The only difference is in Charlie's scenario above.   In PWR, A goes ---
>in the committee's mind, B goes.
 
>This is where I wish the committee would adopt PWR as strictly as we
>once layed it out to be.
 
>It's not often that it will make a difference, but it could ...  And I
>think if the committee was explained PWR -- they would see how on the
>money it is, and that it's logically better.
 
>Cause, heck, they don't give the regular season ECAC title to Team B -
>just because it was 2-0 vs. Team A --- when Team A had more points
>overall --- right?
 
        I beg to differ.  Equating pairwise comparisons to individual
games is a bit off since they factor in a lot of criteria such as
record vs common opponents and record vs Teams Under Consideration.
In that way, it's more like the ECAC tiebreaker system.  If two teams
are tied for a position in the ECAC standings (their won-lost-tied records
playing a balanced schedule are identical), do you apply the ECAC
tiebreakers between each team and each other team in the standings,
total up the number of tiebreaker wins, and compare those totals?  No,
you consider only the tiebreaker between the two teams in question.
 
        Now, since two teams vying for an NCAA tournament spot haven't
played the same schedule, you can't just compare their records
(although if you wanted to stretch the analogy, you could think of
"PWR with errorbars" as corresponding to ECAC record).  Instead, you
compare the two teams using the selection criteria, just as you'd use
the tiebreakers in the ECAC situation.  (In fact, the pairwise
comparison concept grew out of the tiebreakers they used to use if two
teams were considered too close in RPI to separate on that basis
alone.)
 
        If Northeastern and St. Cloud are competing for the last
playoff spot, why should anyone care how they stack up against
Michigan State or Princeton?  The Northeastern-SCSU PWC already
*compares* the two teams based on a host of factors, most of them
relating to how the teams did against *other* teams.  Similarly, if
you want to decide whether New Hampshire or Clarkson gets the last
Eastern bye, you compare UNH and Clarkson; you don't compare them each
to each other, plus Michigan, Miami, North Dakota, BU, etc.
 
        Using individual comparisons to seed the tournament is utterly
straightforward once the field is chosen, when the number of teams
involved is, at absolute most, eight; it can get a little bit subtle
when trying to pick 12 teams out of 22 or so if the comparisons become
really non-transitive, but it could easilybe converted to a
deterministic algorithm (like the one I use for the "automatic"
feature on my selection script
<http://www.slack.net/~whelan/cgi-bin/tbrw.cgi?tourney>).
 
                                         John Whelan, Cornell '91
                                     Official Scorer/PA Announcer
                                        U of Utah Ice Hockey Club
                                               <[log in to unmask]>
                      <http://www.cc.utah.edu/~jtw16960/joe.html>
 
HOCKEY-L is for discussion of college ice hockey;  send information to
[log in to unmask], The College Hockey Information List.