Jeff WuerthAssistant Athletics Director for Media Relations(904) 620-4027
Assistant Athletics Director for Media Relations
NCAA Tournament Selection Committee selected its 34 at-large teams
Sunday night who will participate in the 2010 NCAA men’s Division I
Basketball Tournament, and the "Dance Card" correctly predicted 33 of
the 34 actual selections.
Not only did the Dance Card predict 33 of the 34 teams right, they narrowly missed getting all 34 at-large picks correct.
Visit www.DanceCard.UNF.edu to see the correct predictions and how close they were to going a perfect 34-for-34.
34 at-large teams join 31 other teams who earned automatic bids to make
up the 65-team tournament field. The tournament begins with one game on
Tuesday, March 16, and kicks off in full gear on Thursday, March 18.
Card formula was developed by Jay Coleman of the University of North
Florida, Mike DuMond of Charles River Associates and Allen Lynch of
Mercer University (Lynch is a UNF grad). Dr.
Coleman is a Richard deR. Kip Professor of Operations Management &
Quantitative Methods in the Coggin College of Business at the
University of North Florida, Lynch an Associate Professor of Economics
& Quantitative Methods at the Stetson School of Business and
Economics at Mercer University, and DuMond is a principal with Charles
River Associates in Tallahassee, Fla.
Dance Card is a formula derived by Coleman, DuMond and Lynch as an
estimate of the decision rule that the NCAA Tournament Selection
Committee uses when picking the teams that will get at-large bids.
Using past decisions of the NCAA Tournament Selection Committee and
numerous pieces of information (e.g. RPI rankings, number of wins and
losses, conference records, etc.) for all teams that were candidates
for at-large selections in those years, Coleman, DuMond and Lynch
devised the Dance Card formula as an estimate of which pieces of
information were most important to the Selection Committee, and the
weights that the Committee placed on those pieces of information.
original version of the Dance Card has missed on more than three spots
in only two seasons (2007 and 2008). Over the 15 year period from 1994
through 2008, the original Dance Card correctly predicted 476 of the 512 available at-large Tournament slots (or 93 percent).
Over the nine-year period (2000 through 2008) for which the original
Dance Card was used since its initial development in 1999, it correctly
predicted 284 out of the 307 available at-large Tournament slots (92.5 percent). The formula's best years were in 2001 and in 2005, when it correctly predicted 33 of the 34 available at-large Tournament slots (or 97 percent accuracy).
A newly updated and modified Dance Card correctly predicted 32 of the 34 available
at-large slots in 2009, the first season in which it was used to make
predictions. Over the 10 years of data (1999 through 2008) on which its
development was based, it would have correctly predicted 329 of the 341 available at-large Tournament slots, or 96.5 percent accuracy (i.e., it would have averaged about one at-large slot missed per year).
The Dance Card can only be as accurate as the Selection Committees are consistent; it is an estimate of the Selection Committees' (not
the authors') decision criteria. The high level of accuracy and
consistency of the model is strong evidence that the Selection
Committees (which differ in composition each year) are actually quite
consistent from year to year.
explains some of the history of the Dance Card and the adjustments that
have been made over the years by himself, Lynch and DuMond.
in 1999, Allen Lynch and I looked at the at-large decisions of the NCAA
Tournament Selection Committee from 1994-1999,” said Coleman. “Using
those decisions, and estimates of the so-called “nitty-gritty report”
data on each team that was provided to the Selection Committee in those
years, we developed a statistical formula that we’ve used ever since to
predict who’s going to get a bid. In 2001, we published the formula –
which we called the Dance Card – and the research behind it in the
academic journal Interfaces. The formula we developed back
then was about 93 percent accurate when making predictions during
2000-2008, meaning it missed on average about two at-large slots a
followers of college basketball may well know, the nitty-gritty report
provided to the Committee contains a myriad a team performance
statistics for each team under consideration for an at-large bid. And
many of those stats get a lot of play in the media this time of year.
In essence, what the Dance Card does is assimilate all of that data
into just one number that best reflects whether
the team will get a bid. That one number reflects which stats are most
highly related to past committee decisions, and the degree to which
they are related.”
“Thus, the Dance Card is our estimate of the implicit formula that the Committee essentially follows when it makes at-large decisions. That’s not at all to say that the Committee has an actual formula that it uses – it doesn’t. However, Committee decisions are
highly related to certain team statistics (e.g., the RPI, conference
records, etc.). Again, the Dance Card reflects those stats that are
related, and the degree to which they are related. Oddly enough, it’s
likely that past or present committee members don’t even realize how
much their decisions are related to certain team performance
statistics. Also, it would likely be very helpful to have something
that could show them the patterns of past committees as they relate to
those stats. Again, that’s in essence what the Dance Card does.”
using the original Dance Card to make predictions for several years, we
noticed that the older the formula got, it started to miss more
frequently. (That’s a common phenomenon for statistical formulas like
the one we developed.) It actually missed four slots in 2007 and 2008,
which was well outside the norm for the preceding years. That’s one of
the reasons that, over the last couple years or so, Allen and I (along
with a third author, Mike DuMond) developed an updated Dance Card in
which we use more data (10 years worth, from 1999 through 2008), more
inputs (i.e., more team statistics), and more refined variable
definitions to derive what we believe is an even better model of the
committee’s decisions. The new Dance Card, which would have been
96-97% accurate over those 10 years (or only about one slot missed a
year), was used to make predictions for the first time last year, when
it missed two slots.”
The Dance Card can be seen at www.DanceCard.UNF.edu.
Copyright © 2012 University of North Florida1 UNF Drive | Jacksonville, FL 32224 | Phone: (904) 620-1000
Contact | Emergency | Privacy | RegulationsDisability Accommodations