[eDebate] Seeding tiebreakers

A Numbers Game edebate edebate
Tue Jul 28 17:53:50 CDT 2009


Summary:
Most tournaments use adjusted points as the first tiebreaker for elim
round seeding. Total points more accurately predicted elim ballot
winners, but only by a small margin. Both gave more accurate
predictions than opponent wins, at least as a first tiebreaker.

In outround ballots between teams with the same preliminary record and
the same total points, the negative team won almost 55% of the time,
and the team with the higher adjusted points lost almost 55% of the
time. However, there were only a few of these rounds, so the results
are probably not statistically significant.

Details:
The debateresults.com data includes 542 tournaments. As a first
tiebreaker to use for seeding teams before outrounds, [*]

510 tournaments used adjusted points
21 tournaments used total points
10 tournaments used opponent wins
1 tournament used ranks

One could rank a tiebreaker's accuracy by how often it correctly
predicts the result of a round between otherwise tied teams. [**]
debateresults.com records 21,662 elimination rounds ballots. 4971 of
these ballots are between teams who had the same preliminary round
ballot count. 4244 of those were at tournaments that recorded speaker
ranks in preliminary rounds.

The tie breakers I tested, listed in order of their accuracy, were:

58.7% total points
58.1% adjusted points
56.1% ranks
55.4% adjusted ranks
53.5% opponent win percentage [***] (takes into account opponents with byes)
53.1% total opponent wins [***]
51.4% opponent points
50.6% opponent ranks
50.5% just assume the negative team will win
50.0% flip a coin

238 ballots were between teams that won the same number of ballots and
earned the same total speaker points in preliminary rounds. 186 were
from tournaments where ranks were recorded.

The secondary tiebreakers I tested, listed in order of their accuracy, were:

57.8% adjusted ranks
56.5% ranks
54.6% just assume the negative team will win
50.0% flip a coin
49.6% opponent points
48.7% opponent wins [***]
48.1% opponent win percentage [***]
45.4% adjusted points
44.1% opponent ranks

Notes:

[*] Some of the debateresults.com data is erroneous. For instance,
according to NDT standing rule V.C.2,

"2. Seeding. The criteria for determining the seeding for elimination
rounds shall be administered in the following order: (1) wins, (2)
ballots, (3) adjusted combined speaker points (dropping high and low
ballots), (4) continue dropping high and low ballots down to twelve
remaining ballots, (5) flip of a coin."

The debateresults.com data, on the other hand, says that the tie
breakers for NDT seeding are:

1. wins
2. ballots
3. 1x adjusted points
4. 2x adjusted points
5. 3x adjusted points
6. 4x adjusted points
7. opponent wins
8-13. random

[**] A tie in a statistic is counted as half of a correct prediction,
no matter the outcome

[***] Opponent wins were calculated to exclude the result of the
opponent vs. the team in question



More information about the Mailman mailing list