There’s no shortage of data out there when it comes to college football, so I decided to take the time to create a least squares model, based on the same principals I have been using for the NFL, and outlined here.

The idea was to take all 752 schools that played college football, cross that with all 4138 games that were played (up to last weekend) and see how they predict the games (especially versus the Vegas lines).

When you try to create a model with 753 variables (I also included home field advantage for the regular season games as a variable), you quickly begin to test the limitations of Excel and Solver. After a little digging, I discovered that Frontline Solver, the company that developed Solver for Excel, limits the version included with the Microsoft product to 200 variables.

Not to be stopped, I found they have some advanced pro models and engines that will handle significantly more and downloaded a 16-day trial of these.

You set it to pick the best engine and it runs. For this particular problem, it happened to run the Gurobi Optimization engine.

From there, it is fairly plain and simple. You run the solver and it calculates a ranking for each team. I just plugged in the bowl games and here’s what we get.

Team ATeam BExpVegasDiff% Diff
Colorado StateWashington St7.685.52.1840%
Southern CalFresno State-8.22-5-3.2264%
BuffaloSan Diego St-6.86-1.5-5.36357%
Louisiana-LafayetteTulane-0.671.5-2.17-145%
Ohio U.East Carolina16.48142.4818%
Oregon StateBoise State-0.74-3.52.76-79%
PittsburghBowling Green5.6950.6914%
Utah StateNorthern Illinois-5.462-7.46-373%
MarshallMaryland-5.11-2-3.11156%
SyracuseMinnesota6.4151.4128%
WashingtonBrigham Young-8.75-3-5.75192%
RutgersNotre Dame20.75146.7548%
CincinnatiNorth Carolina5.9332.9398%
Miami FLLouisville3.603.50.103%
MichiganKansas State3.233.5-0.27-8%
Middle Tennessee St.Navy9.6063.6060%
MississippiGeorgia Tech3.71-3.57.21-206%
TexasOregon18.33144.3331%
Texas TechArizona St20.1914.55.6939%
Boston CollegeArizona12.467.54.9666%
Virginia TechUCLA11.227.53.7250%
Mississippi StateRice-9.02-7-2.0229%
DukeTexas A&M9.6913-3.31-25%
NebraskaGeorgia10.8291.8220%
Nevada-Las VegasNorth Texas7.916.51.4122%
WisconsinSouth Carolina-2.330-2.33-233%
IowaLSU7.218-0.79-10%
Michigan StateStanford9.2463.2454%
Central FloridaBaylor24.64177.6445%
OklahomaAlabama14.3116-1.69-11%
Oklahoma StateMissouri-1.741.5-3.24-216%
ClemsonOhio State2.492.5-0.010%
HoustonVanderbilt-5.733-8.73-291%
Arkansas StBall St10.529.51.0211%
AuburnFlorida State19.817.512.31164%

Here’s what the table means. Vegas has Auburn as a (+7.5) underdog to Florida State, however, the model shows FSU as an almost 20-point favorite, an almost 12-point difference.

One of the most interesting lines is Clemson as a 2.5-point underdog to Ohio State. The LS model has that game as a 2.49-point spread, almost exactly spot on.

Several others are extremely close like that: Miami-Louisville (3%), Michigan-Kansas State (8%), Iowa-LSU (10%), Oklahoma-Alabama (11%) and Arkansas State-Ball State (11%).

Here’s a quick rundown of which side the computer is on for each game (including the Vegas lines): Washington State (-5.5), USC (-5), Buffalo (-1.5), Louisiana-Lafayette (+1.5), East Carolina (-14), Boise State (+3.5), Bowling Green (-5), Utah State (+2), Marshall (-2), Minnesota (-5), Washington (-3), Notre Dame (-14), North Carolina (-3), Louisville (-3.5), Michigan (+3.5), Navy (-6), Georgia Tech (+3.5), Oregon (-14), Arizona State (-14.5), Arizona (-7.5), UCLA (-7.5), Mississippi State (-7), Duke (+13), Georgia (-9), North Texas (-6.5), Wisconsin (0), Iowa (+8), Stanford (-6), Baylor (-17), Oklahoma (+16), Oklahoma State (+1.5), Clemson (+2.5), Houston (+3), Ball State (-9.5) and Florida State (-7.5).

Of course bowl games are strange for a variety of reasons, (hitting the buffet circuit, poor practices, lack of focus…) so there always seems to be a lack of consistency.

We will check back in a few weeks on how well this did.