College Football Prediction Models

Welcome back to another edition of “TSI vs the Field,” my weekly article where I analyze where my T Shoe Index compares to other top predictive models in college football and provide betting advice based on where the models align and where they differ the most. This article was a sizzling 3-0 last week on games I wrote up where the models aligned most closely. Here’s what the computers are predicting for Week 2 in college football:

Note: SP+ Rankings are the work of CFB analyst Bill Connelly (ESPN), FPI is the Football Power Index from ESPN Analytics, and Sagarin is Jeff Sagarin, formerly of USA Today.

 

Week 2 College Football Model Alignment:

Maryland Terrapins (-17) vs Northern Illinois Huskies, O/U 46.5

Our biggest model consensus this week is on the Maryland/NIU matchup, where there is just a model variance of 2.3 points from the highest to lowest prediction, all of which align on the same side of the spread. Sagarin projects Maryland -13.7, while SP+ makes it Maryland -16, with TSI and FPI filling in the middle, good for a model average of Maryland -14.8, so the fact the spread is at a key-ish number of 17 is a “fire” signal from a betting standpoint.
College Football Week 2 Pick: Northern Illinois +17

Marshall Thundering Herd (-10) vs Missouri State, O/U 55.5

Another unanimous ATS play brings us to my old stomping grounds in Huntington, West Virginia, where Marshall is laying 10 points against Missouri State, who just got blown into another dimension by USC last week. Our model variance is just 2.5 points, with SP+ projecting Marshall -11.6 and all other models are around Marshall -14, all of which get us above the spread and key number of 10 with an average of Marshall -13.4. As a reference point, the game grades for each team in their Week 1 games – mind you, Marshall lost by 38 points to Georgia – indicate Marshall was about 32 points better than Missouri State last week, so take that for what it’s worth. 
College Football Week 2 Pick: Marshall -10

Duke Blue Devils vs Illinois Fighting Illini (-2.5), O/U 49.5

This is one of my favorite bets of the week. Our model variance on this game is just 2.8 points, with Sagarin calling for a pick ‘em and SP+ indicating Illinois -2.9, with an average model projection of Illinois -1.3. This line has bounced back and forth from 2.5 to 3, so I’m confident we’ll see another 3 pop up on the board. I said when I saw this line that Illinois will likely be one of the biggest public sides of the week, and generally when that’s the sentiment, but the data is on the other side, that’s a great bet.
College Football Week 2 Pick: Duke + 2.5 (wait for 3)

Week 2 College Football Model Disagreement

Liberty Flames (-6.5) vs Jacksonville State Gamecocks, O/U 49.5

This is definitely one game I want no part of, based on the projections. Our model variance is 16.5 points, with TSI calling for Liberty -14 but FPI favoring Jacksonville State by 2, with an average projection of Liberty -5. Based on the Week 1 game grades, Jacksonville State was 3.5 points better than Liberty, but the preseason priors (which are still very relevant at this point) indicate that might be more of a one-off than the reality. I’m fascinated to see how this plays out, but I won’t have any action on it.

Toledo Rockets (-7) vs Western Kentucky Hilltoppers, O/U 60.5

Another G5 clash that the models can’t agree on takes us to Toledo, Ohio, where TSI projects an 11-point Rockets win, but SP+ says Western Kentucky -2, with an average model projection of Toledo -5.1 and a variance of 13.3 points. That’s a huge spectrum of potential outcomes here, so this is another wait-and-see game for me to collect another data point on these teams. Last week’s game grades say the Hilltoppers were 7 points better than Toledo (and yes, that’s opponent-adjusted), so I’m curious to see if this is just an overreaction to one week or if the preseason priors are just flat out wrong on these teams.

UNLV Rebels vs UCLA Bruins (-2.5), O/U 54.5

It’s been a mixed bag for UNLV in its first two games with coach Dan Mullen, while UCLA got destroyed by Utah last week in its lone game this season. Our model variance here is 11.4 points, with FPI calling for an 8-point UNLV win, but TSI favoring UCLA by 3.5, with an average model projection of UNLV -1.7. 2025 game grades align with the model average, placing UNLV as 7.5 points better than UCLA. This will be another stay-away for me, and an opportunity to collect data and learn whether in today’s college football era, if preseason priors are becoming less and less relevant, or if overreacting to early-season results is a fool’s errand. 

Keep an eye out for this weekly feature on Thursdays and also my Monday early-week college football bets.