Saturday, December 19, 2009

Last Minute Bowl Game Preparations

The first bowl games are only hours away, so it's time to get ready! No, I don't mean chili and guacamole (though that sounds good). I mean our third annual attempt to use the rankings to predict the bowl games. We'll once again be playing over at ESPN's College Bowl Mania (not endorsing it, it's just where I'm signed up), which requires predictions of outcomes and rank-ordering of confidence in those predictions. The 2007 random walker bowl predictions turned out pretty well, placing above the 75th percentile of all entries, which I believe is good for a fully automated system. The 2008 bowl predictions, however, really missed the mark. But that doesn't stop us from coming back to try again.

The big caveats up front are that we definitely do not recommend these rankings for anything beyond entertainment. The underlying rankings are not intended to be best in any sense, except that they are relatively simple to explain. And the methodology used to assign predictive probabilities here is fundamentally flawed. The general principle we use is that we assign a probability to a victory by the higher-ranked team such that the addition of a mythical fractional-outcome game split according to those probabilities does not change the existing rankings. However, this simple principle is fabulously erroneous for many reasons, not the least of which is that it can algebraically return a greater-than-100% chance of victory. You might think a mathematician would bother to fix such a problem, but there are so many other flaws in trying to use these rankings to predict upcoming games that it's unclear whether further detailed work would yield anything useful.

Like past years, we used the full-component RWFL predictions at p=0.75 (see other parts of this site for those details). Each game selection is specified by predicting the higher-ranked team will beat the lower-ranked team. The games are presented in order of confidence (highest to lowest), determined by the mean of the stated probability spread (obtained individually by the first-place votes and the last-place votes). I'm happy to discuss such details further, though I warn again that it is a relatively crude model. And since I'm assuming visitors just want the results, let's get to the predictions.

----
International
South Florida (7-5) over Northern Illinois (7-5)
70-82%
----
AdvoCare V100 Independence
Georgia (7-5) over Texas A&M (6-6)
67-79%
----
Chick-fil-A
Virginia Tech (9-3) over Tennessee (7-5)
69-76%
----
Emerald
Southern Cal (8-4) over Boston College (8-4)
65-70%
----
Rose Bowl Game presented by Citi
Oregon (10-2) over Ohio State (10-2)
61-71%
----
Brut Sun
Stanford (8-4) over Oklahoma (7-5)
63-69%
----
Insight
Minnesota (6-6) over Iowa St (6-6)
65-67%
----
Konica Minolta Gator
West Virginia (9-3) over Florida St (6-6)
65-67%
----
New Mexico
Fresno St (8-4) over Wyoming (6-6)
65-66%
----
Pacific Life Holiday
Arizona (8-4) over Nebraska (9-4)
65-66%
----
EagleBank
UCLA (6-6) over Temple (9-3)
56-70%
----
Outback
Auburn (7-5) over Northwestern (8-4)
61-62%
----
Papajohns.com
South Carolina (7-5) over Connecticut (7-5)
61-62%
----
Valero Alamo
Texas Tech (8-4) over Michigan St (6-6)
57-66%
----
Champs Sports
Miami FL (9-3) over Wisconsin (9-3)
56-67%
----
Sheraton Hawaii
Nevada (8-4) over SMU (7-5)
56-65%
----
Bell Helicopter Armed Forces
Houston (10-3) over Air Force (7-5)
47-75%
----
FedEx Orange
Georgia Tech (11-2) over Iowa (10-2)
56-61%
----
Citi BCS National Championship Game
Alabama (13-0) over Texas (13-0)
47-70%
----
Texas
Missouri (8-4) over Navy (8-4)
54-60%
----
Gaylord Hotels Music City
Clemson (8-5) over Kentucky (7-5)
57-58%
----
St. Petersburg
Central Florida (8-4) over Rutgers (8-4)
50-65%
----
R+L Carriers New Orleans
Middle Tennessee St (9-3) over Southern Miss (7-5)
56-57%
----
S.D. County Credit Union Poinsettia
California (8-4) over Utah (9-3)
53-59%
----
AT&T Cotton
Oklahoma St (9-3) over Mississippi (8-4)
52-59%
----
Allstate Sugar
Florida (12-1) over Cincinnati (12-0)
52-59%
----
Capital One
LSU (9-3) over Penn State (10-2)
50-60%
----
Tostitos Fiesta
TCU (12-0) over Boise St (13-0)
53-55%
----
Roady's Humanitarian
Idaho (7-5) over Bowling Green (7-5)
50-58%
----
GMAC
Central Michigan (11-2) over Troy (9-3)
54-54%
----
Little Caesars
Ohio U. (9-4) over Marshall (6-6)
52-53%
----
AutoZone Liberty
Arkansas (7-5) over East Carolina (9-4)
44-55%
----
MAACO Las Vegas
Oregon St (8-4) over Brigham Young (10-2)
49-50%
----
Meineke Car Care
North Carolina (8-4) over Pittsburgh (9-3)
39-51%
----

Note, in particular, the toss-up nature in the last picks above. Meanwhile, even some of the more confident calls include probability spreads that dip below 50%, including the call for the BCS National Championship Game.

By the way, we get a LOT more visitors than comments on this site. Feel free to express your opinion about the accuracy (or better yet, the stupidity) of these predictions!

Labels: