CFB Nerds' Model Predictions for Notable Games in Week 8 and Beyond
Ahoy!
Here's their website. Winners in bold.
Notable Week 8 Games
- Michigan 39.5, Michigan State 15.0
- Ohio State 17.4(!!), Penn State 26.9
- Florida State 23.6, Duke 28.9
- Alabama 21.7, Tennessee 22.9
- Utah 22.7, USC 37.2
- Iowa 19.2, Minny 18.9
Michigan Games After Week 8
- Michigan 32, Ohio State 32.1
- Michigan 29.4, Penn State 32.5
- Michigan 42.7, Maryland 28.9
- Iowa 19.2, Minnesota 18.9
Thoughts
- The Michigan State score looks right, though that doesn't take into account Michigan's (politely denied) wrath. I'm getting ready for a bunch of MSU penalties and Tuttle 2-point conversions late in the fourth quarter.
- The CFB Nerds' model last week had the Game at Michigan 28.1 and Ohio State 31.3, so it seems like the model likes Michigan more against the Buckeyes. It also had UM losing 24-32 to PSU, so this has also tightened up. I'm not sure why. Michigan destroyed Indiana (by a wider margin than Ohio State, though that was early in the season), but Penn State had a similar performance against an overmatched opponent.
- This is a home game for FSU, and it's basically their entire season. Their toughest remaining contests are rivalry games against struggling Miami and Florida. If Duke was playing at home and if FSU had more to worry about on their plate, I think Duke would win, but I think FSU will squeak by, as much as I want to see them lose for Michigan's sake.
- I think Alabama's early games, where new QB Milroe is still figuring things out, drags down their predicted performance in the model. However, I haven't watched Tennessee this year.
- I'd usually favor Utah in a game against USC, especially this USC team, but Utah is really banged up.
- I might watch some of the Iowa game. 1/3 of their touchdown receptions is on crutches with a knee injury (Erick All). That's probably a bridge too far, even for Iowa. Still, it's spooky season, so who knows?
October 18th, 2023 at 1:21 PM ^
Iowa outscoring Ohio State (in different games, of course) ---- that would be something.
October 18th, 2023 at 1:25 PM ^
I presume Iowa will block the tying tenth of a field goal to preserve the epic 19.2-18.9 victory.
October 18th, 2023 at 1:35 PM ^
It's also possible this model is predicting Iowa outgains Minnesota, 19.2 yards to 18.9 yards. Iowa wins by 11 inches!
October 18th, 2023 at 2:16 PM ^
My wife (1991 UM GRAD) always wins with 11 inches...if you know what I'm saying!
(sorry, couldn't help myself)
(That's what she said.)
#humorforone
October 18th, 2023 at 2:49 PM ^
I doubt whether that's what she said. Did you give her an accurate ruler? Or just numbered it whatever you wanted it to be? :D
October 18th, 2023 at 3:01 PM ^
Big Ten refs doing the measurements.
October 18th, 2023 at 3:41 PM ^
Chances are they would call it short for a UM Grad! JT on the other hand...
hmmm. never mind. :D
October 18th, 2023 at 4:40 PM ^
The trick is to convince her that centimeters are inches
October 18th, 2023 at 3:16 PM ^
She said you can't help yourself? Very accommodating of her.
October 19th, 2023 at 12:28 AM ^
You sure they aren’t using metric units. 11cm seems more plausible! /jk
October 18th, 2023 at 2:21 PM ^
So did I.
October 18th, 2023 at 8:59 PM ^
Pass the bleach if so.
October 18th, 2023 at 1:28 PM ^
Hahahaha...I mean, who in their right might would talk to NERDS about predicting sports scores. I mean, nerds don't know anything about sports, especially football! Look at those score predictions!! Hey nerds, YOU CAN'T SCORE PARTIAL POINTS IN FOOTBALL! 32.1...lmao. Nerds gonna nerd I guess.
/s
October 18th, 2023 at 1:54 PM ^
"NERDS!!!!" --Sincerely, Fred "The Ogre" Palowakski
October 18th, 2023 at 2:00 PM ^
Ogre, You Asshole.
October 18th, 2023 at 2:21 PM ^
Nerds are good at sex because it's all they think about while jocks care about sports. Oh, and Betty is like a goat.
October 18th, 2023 at 3:18 PM ^
Though all the numbers are listed with a decimal place of course, that "Iowa 19.2" tickles me, with its hint that if the offensively offense-challenged Iowa could, it would score less than a full point.
October 18th, 2023 at 1:32 PM ^
Maryland, OSU and PSU aren't averaging 30 points a game against us.
October 18th, 2023 at 3:28 PM ^
After reading the 1st section I was thinking 'so MSU is going to score the most points on us of anyone this season? Doubt it, but alright'
After reading the 2nd section, it's hard to take these guys analyses seriously when they're predicting us giving up 30 ppg against PSU, Maryland, and OSU. I know our schedule hasn't been great, but when your model is spitting out stuff like that...you need to look at your model again
October 18th, 2023 at 8:32 PM ^
Clearly they added a "toughness" parameter they are still working the kinks out on
October 18th, 2023 at 1:35 PM ^
I don't think these models take into account that James Franklin is the head coach of Penn State.
October 18th, 2023 at 3:42 PM ^
Yeah, they seem unreasonably high on PSU.
October 18th, 2023 at 4:45 PM ^
Not defending it per se, but I bet that model absolutely LOVES the fact that PSU hung 31 on Iowa. That performance against an elite defense will (not entirely unreasonably) make offenses look really good.
October 19th, 2023 at 11:41 AM ^
Penn State had 94 plays against Iowa, that a lot even without the clock rules.
iowa’s defense had so little rest due to ineptitude of the offense that it was destined for them to allow 31 points at least.
October 18th, 2023 at 4:50 PM ^
Their model is likely being unduly influenced by Penn State playing that juggernaut UMass last week.
October 18th, 2023 at 1:38 PM ^
I think their model just fundamentally doesn't work for some teams, and this year it feels like Michigan is one of those teams. Honestly, the past couple of years their system assumed OSU would beat UM and (I think at least one season) PSU would beat UM and it didn't happen. I assume it really likes MOV and PSU's defense being elite and giving short fields to the offense supersedes all the warning flags like "they can't run the ball consistently against UMass" and "they don't throw the ball more than 10 yards down the field unless they have to".
October 18th, 2023 at 2:01 PM ^
My guess about this model is that it takes Blakes and Donovan's low rushing totals. Add to that JJ's less then a Pennix or Mc Cord pass yards. Maybe it's Michigan's total offensive yard average.
The model is flawed.
October 18th, 2023 at 3:42 PM ^
Yeah, I suspect that it overrates turnovers by the QBs, as McCord and Allar have thrown 1 pick combined while McCarthy has 3 of them. Now, someone who's watched the games would recognize that McCarthy's 3 picks were sort of one-offs while the other two have tried desperately to throw more of them but the defenses haven't obliged.
October 18th, 2023 at 2:17 PM ^
Every model has it's quirks, and yeah this model for whatever reason just doesn't understand or predict Michigan that well. I've seen this issue in years past and it made sense then, since Harbaugh's style of low scoring slow motion blowouts with little passing never made the models happy, even if it was effective. But this year's team is winning games the way you'd think that models would love (balanced offense, no turnover/penalties, racing out to big margins, etc.) and yet Penn State (a team with no explosive plays, a mid level run game, and an ultra conservative QB) and Ohio State (iffy run game, sputtery offense, lack of aggression/pressure on defense) are being looked upon more favorably than us. It's mind boggling at this point, when we have 7 games of data to look at
October 18th, 2023 at 3:45 PM ^
I remember Kenpom saying that his model used to love Wisconsin (in basketball) more than it probably should, and I believe SP+ also used to love Wisconsin (in football) way more than it should as well. I think there are just teams that "crack" these modeling systems and PSU, for whatever reason, is doing that this season.
October 18th, 2023 at 3:12 PM ^
This model doesn’t work for Michigan because no other team pulls their starters midway through the 3rd quarter. I’m guessing their model assumes previous performance is close to the ceiling for each team, so when Michigan only scores 31 and allows a garbage time score, it doesn’t account for the fact that if they put the wood to their opponent they’d likely have won every game by 40+.
October 18th, 2023 at 3:20 PM ^
Boost Michigan's numbers by 1/3 and it should work out.
October 18th, 2023 at 4:56 PM ^
Good for them that they can make a living with a model that is much less accurate than Connelly's SP+
October 18th, 2023 at 1:39 PM ^
I thought we agreed that these guys were morons with a crappy model.
October 18th, 2023 at 1:44 PM ^
Same. Stop giving these idiots free bandwidth.
October 18th, 2023 at 2:43 PM ^
Agreed.
They're not the only site that effectively gets advertised here.
October 18th, 2023 at 2:13 PM ^
Did they copy the BCS formula?
October 18th, 2023 at 2:16 PM ^
I think it's at least an interesting perspective. As the model gets more data on the teams, I think it will sharpen up. We're seeing the score predictions with OSU and PSU get closer and closer to Michigan, for example.
October 18th, 2023 at 2:48 PM ^
The problem I have with them is that they constantly reference "our model" but they never go into what their model is. They never really explain which metrics they're using, how they're weighing them, or how they are anchoring their model to reality.
For example, I know exactly which metrics SP+ uses because Bill Connelly's intentions and methodology are fairly transparent (from Ed Feng):
Connelly’s numbers inform his writing as he ranks college football teams based on four factors.
- Explosiveness – Measured by equivalent points per play, a metric similar to the expected points added used by ESPN’s FPI.
- Efficiency – Measured by success rate, or 50% of the necessary yards on 1st down, 70% on 2nd down, and 100% on 3rd and 4th down.
- Finishing drives – Measured by points per trip inside the opponent’s 40 yard line.
- Field Position – Measured by average starting field position, a number affected by special teams.
His methods takes each of these factors and adjusts for strength of schedule. These four factors are combined to make the final rankings. Connelly provides a sense for the importance of each factor in his original article on football’s five factors.
Meanwhile, I can "use" CFB Nerds model, but I only have the vaguest notions of what they're trying to account for and there is very little information on their website. You get little pieces like "Teams playing very weak schedules are over-valued" – which myself and others would disagree with (it's about how you win, in addition to who you beat), but from afar it seems like they have issues tethering their model to the eye test (to contrast, SP+ incorporates preseason polls through week 8 to make up for the small sample size).
It's just hard for me to take them seriously as a predictive model when they constantly reference their model and rarely - if ever - explain what it is.
October 18th, 2023 at 3:22 PM ^
I'm sorry, m9tt, but you make a fundamental error here, of seeking facts and explanations.
To be fair, that silly approach pops up occasionally on MGoBlog. Still, that's no excuse.
October 18th, 2023 at 3:34 PM ^
I publicly acknowledge my mistake and look forward to my upcoming trip to Bolivia
October 18th, 2023 at 3:49 PM ^
Yeah, I understand not wanting to give away the actual calculations and the weights for them but I look at their "test our system" page and it assumes that Georgia vs. PSU would end with the Nittany Lions winning by 11, which feels a bit weird. And in general any team I put in with PSU framed the Nittany Lions as the likely winner or a virtual tie. That feels like a weird skewing, but as you noted not knowing what they're relying on makes it hard to guess what might be the root cause.
October 18th, 2023 at 4:09 PM ^
Right. They don't have to give away the family recipe, but they should tell me if the mystery meat is chicken or beef or something more exotic, because that's going to make a huge difference in how seriously I'm going to consider it.
October 19th, 2023 at 12:34 AM ^
Because 11 is the same as 10 for all practical purposes
October 18th, 2023 at 2:18 PM ^
I've never bothered to see how accurate it was in predicting other games, but they certainly get Michigan wrong a fair bit.
October 18th, 2023 at 3:10 PM ^
Last week's results:
- Michigan 49, Indiana 10.5
- Actual 52 - 7 pretty damn close
- Michigan 28.1, Ohio State 31.3
- Oregon 35.5, Washington 29.8
- Washington 36 - 33 missed winner, score close
- Notre Dame 43.9, USC 40.5
- Notre Dame 48 -20 not terrible
- Miami (YTM) 31.4, North Carolina 23.8
- North Carolina 41 -31 missed winner
- UCLA 18.1, Oregon State 17.8
- Oregon State 36 - 24 missed winner
All told of the games featured by the OP they went 2-3 outright which is pretty bad IMO. So yeah, their model sucks.
October 18th, 2023 at 5:15 PM ^
Anything that doesn’t clearly state as a baseline that it is opponent adjusted, tempo adjusted and removes garbage time is almost certainly complete horseshit. You might as well tune in to WTKA and listen to Steve Clarke talk about counting stats for half an hour like it’s 1995.
October 18th, 2023 at 1:43 PM ^
39.5 to 15? Now I know where Brian gets his weird score predictions from.
October 18th, 2023 at 1:45 PM ^
Do these guys post a review of their predictions? Cannot find anything about it on google. It seems like as much of a joke as ESPN's FPI.
October 18th, 2023 at 2:17 PM ^
They do matchup preview videos, which might be handy.
Here's the Ohio State/Penn State one: https://www.youtube.com/watch?v=2gTp55smdlE
That also gives you an idea on how they use the model to make their own predictions.