On Unpredictability II (was Re:Scores)

ZAMM_Phaedrus · Mon, 11 Feb 2002 22:24:17 -0800

Now, on to predictability.
1) A game of
quizbowl is, in effect, a quasi-random sampling of the
entire corpus of human knowledge. A distribution is an
attempt to direct this sampling, although each
distribution has built-in assumptions of what is worth asking
about.
2) Each packet, therefore, is a different sample. It
stands to reason that success on one packet does not
guarantee success on another packet. I would argue that the
realm of knowledge is sufficiently wide to allow for
this.
3) The goal that the outcome of a particular
match-up ought not to vary no matter which packet is used
from a give set is an unreasonable. Although wide
variations in talent level make certain matchups likely to
be totally one-sided, for many match-ups, the
possibility of what one would call an upset exists as a
non-negligible probability. That is, one team may be clearly
better, but will win, on average, eight games in a series
of ten. 
4) Writing to explicitly prevent an
upset is as bogus a proposition as writing to
explicitly cause one. In either case, you are playing
favorites.
5) Even if outcomes were not affected by packet,
other variables exist. One team may play well in ne
game and poorly the next game, due to causes unrelated
to the packet used. Players are prone to
psychological and physiological effects, for example the effect
of eating (or not eating) lunch or being in the same
room as someone you dislike.
6) Therefore, while
winning a prior game has some predictive value, it does
not have a total predictive value for a
rematch.
7) Bottom line: get tossups and just win,
baby.

>CWRA > OSU: 465-0
>PitA > OSU:
330-155

Each packet is different. Teams match up differently.
using scores against a single common opponent suffers
from the error of using a small sample size. For
example, both teams cited played OSU twice. The scores
left out are, according to the website PitA 370-OSU
110 and CWRA 330-OSU 55. 

>PitA > CWRA:
485-85
>CWRA > PitA: 310-160

In ameliorating the
problem of small sample size, each team cited played the
same teams, including the same teams twice, save for
having each other on the schedule. A decidedly better
and statistically sound, though less dramatic,
argument would be to cite overall points per game or to do
a stricter analysis on an opponent by opponent
basis.

[snip a few other scores]

>Honestly, our loss
to Case in the playoffs gave me a surreal mental
feeling. It was like a parallel universe, or like I
couldn't tell what was a dream and what was reality. To be
more specific, I felt that way at halftime when we
were trailing 210-20.

So, basically, you felt
out of it because you were having your collective
asses handed to you in that game. Did you go in
overestimating yourself or underestimating your opponents? Is
quizbowl a game or a short story by Borges? 

The
good teams can stop someone every once in a while. The
great teams are the ones who can come from behind when
they're down big at the half to a good
team.

Anthony de Jesus, deconstructing quizbowl