College Football Nerds Model + Model Predictions for Some Games

Submitted by Caesar on October 11th, 2023 at 1:28 PM

Ahoy!


There's a free online prediction model offered by College Football Nerds (link). I believe it's the same thing they use in their previews. I think it's pretty handy, though you need to start an account to do more than two matchups. Here's a picture of the stuff they show on the site:



Below, I've put what the model predicts for some noteworthy games, both this week and further into the future. Winner in bold. 

  • Michigan 49, Indiana 10.5
  • Michigan 28.1, Ohio State 31.3 
  • Oregon 35.5, Washington 29.8
  • Notre Dame 43.9, USC 40.5
  • Miami (YTM) 31.4, North Carolina 23.8
  • UCLA 18.1, Oregon State 17.8

 

Some thoughts:

  • I think Ohio State running up the score and Michigan doing the opposite impacts this prediction for The Game, but I don't know enough about the model to say. However, they mention that the model favors teams with weaker schedules, and I'm thinking running up the score has the same impact as a weak schedule. 
  • Not sure this takes into account home vs. away, or if that even makes a difference in outcomes. That would make me rethink USC/ND, though I think USC is essentially held together by a series of Caleb Williams miracles. 
  • I've seen both UCLA/Oregon State. That looks about right to me. UCLA looked really good, and Oregon State wasn't very consistent. 
MGoBoard
Football
college football nerds
football prediction model
USC
Notre Dame
UCLA
oregon state
Michael Barrett
Michigan
The Game
Ohio State
UNC
Miami (YTM)

hammermw

October 11th, 2023 at 3:16 PM ^

In past years yes, but not this year. I think he even admitted himself that he's struggling a bit with how to handle the transfer data and his model holds onto preseason data longer than most other models. It will be interesting to see how S&P+ performs now that the preseason data is falling off. My guess is it will be back to beating Vegas consistently.

Joined: 08/22/2010

MGoPoints: 519

ca_prophet

October 11th, 2023 at 8:32 PM ^

The best model for predicting winners and final scores is not necessarily the best model for making money.  Bookies want to set lines and wagers so equal amounts of money are on each side, so that they win as much as they pay out - and keep the vig on all of it.  When they can do that, no matter what happens they make money.

Bookies only lose money when they set the lines so that one side outweighs the other - and that side pays off.  Like many people when it comes to money, they prefer to avoid risk.

So even if they use the best model going, the lines they offer won't be same as the model's, so the model can "win" against the Vegas line.

 

Joined: 09/07/2010

MGoPoints: 7087

MichaelCarras

October 12th, 2023 at 2:59 PM ^

That's not true at all. A model can be useful and predictive and not be that useful for betting.  Overcoming the vig is difficult. Sagarin is a good example. Sagarin is infinitely more accurate than human polls like AP and the Coaches Poll or pundit analysis on sports shows or what MGoBlog readers think.  But you won't be able to make a living using Sagarin because they roughly approximate the the betting lines. It is usefulness is getting a good approximation of teams when you don't have a line readily available.

Joined: 01/08/2023

MGoPoints: 507

FauxMo

October 11th, 2023 at 1:36 PM ^

Thanks for putting that screenshot near the top, as I only had to go through one paragraph of text before seeing that their model was flawed and invalid. 

Joined: 06/17/2010

MGoPoints: 137578

Underhill's Gold

October 11th, 2023 at 3:18 PM ^

This.

Given how the model shows UM superior to OSU in 3 of 4 phases, that's clearly true, particularly when our running advantage is huge, while OSU's passing advantage is not quite so large. 

I wondered if this was due to the imbalance between passing and rushing in typical college games. That might explain things given that UM's superiority is in the phase of the game that typically provides less yards.  But the (very simplistic) numbers didn't add up. 

Assume a typical team passes for 238 yards/game and rushes for 156 yards/game. 
57% better rush offense than 156 gives UM an extra ~90 rush yards. 
40% better pass defense than 238 gives OSU an extra ~95 yards(takes away 95 yards from UM)
If we then did the math for UM's slight advantages in pass offense and rush defense, it would show UM with a clear yardage advantage in our simplistic model.

In sum - I think we should assume the model is fairly more sophisticated than an Associated Press game summary. 

(I didn't bother to dig for high quality stats. These are from ncaa.com's 65th ranked FBS pass offense and rushing offense)

Joined: 12/19/2011

MGoPoints: 534

Blau

October 11th, 2023 at 1:42 PM ^

This is cool but I don't see the merit in using the unit items or prediction for games more than 1 or 2 weeks away. Team strategies change week to week depending on the opponent and even 2nd half offensive strategies could throw this thing out of wack real quick. 

To me, this is about as useful as looking up the team stats so far and making an observation rather than a prediction. 

Joined: 01/23/2014

MGoPoints: 35254

JHumich

October 11th, 2023 at 1:45 PM ^

All you have to do is look at how much better the model thinks the OSU passing game would be than ours to know that it's not a good model.

And the team that managed 1.9 YPC against Maryland isn't suddenly going to double that against Michigan...

Joined: 02/18/2017

MGoPoints: 52966

TeslaRedVictorBlue

October 11th, 2023 at 1:54 PM ^

i dont get the simulation. Our defense last year was very good, but not elite elite. and we gave up 23 to a 1st round qb who is doing well in the NFL, with elite WRs everywhere. Their OL doesnt appear much better, they cant run well, and though they have elite WRs, they now have a QB who hasnt proven anything...and our defense appears more stocked at each level... so we're gonna give up 31? At home? 

F the simulation. F Franklin. F Penn State. F the Tough Guy from Toughington, Toughsville USA ...

Joined: 09/02/2021

MGoPoints: 35052