Started By
Message
re: Top 25 teams since 1960 (power ratings)
Posted on 7/13/26 at 2:27 pm to MilkJug
Posted on 7/13/26 at 2:27 pm to MilkJug
quote:
It uses the team's entire body of work. Again, where a team "deserves" to be placed isn't considered. If 2019 LSU and 2019 OSU were to play tomorrow, this model would favor OSU. If 2019 LSU and 2019 Clemson were to play tomorrow, this model would favor Clemson (very slightly)
I might see, based on their season, how a formula MIGHT favor Ohio State (I would likely disagree, but that's not the point); however, I'm not sure how SOS is weighed so little to give the nod to Clemson. Historical data is not retained as much as a I like, but ESPN does have FPI data from 2019.
Their SOS rankings:
LSU #3
Ohio State #8
Clemson #49
That is POST playoff data; their regular season SOS (without Ohio State - a one score win - and LSU - a 3 score loss) is much, much lower.
quote:
2011 LSU and Bama had all-time great defenses. They did not have great offenses, relative to other eras. They were just out of the top 25. 2011 LSU - 31.4, 2011 Bama - 30.4. And actually, the 2016 Bama defense had one of the best defenses on this list (and much better than 2011 Bama), can you guess why?
If I were to guess, you'd say we faced a tougher SOS while finishing 1st in the country in scoring defense.
quote:
I'm not "weighting certain factors". It's machine-learned. The model optimizes itself to be as accurate as it can (with its given data) at determining outcomes.
I work in IT; machine learning still starts with a premise. The premise, in this case, is SOS (or lack thereof).
1
Posted on 7/13/26 at 2:36 pm to skrayper
quote:
If I were to guess, you'd say we faced a tougher SOS while finishing 1st in the country in scoring defense.
It was scoring defense. Well done. Bama's defense scored 6 pts/game that year! The model tries to determine outcomes - that's it. And a defense that scores points and stops opponents from scoring is better than a defense that just stops opponents from scoring.
quote:
I work in IT; machine learning still starts with a premise. The premise, in this case, is SOS (or lack thereof).
SOS is baked in. But results on the field still matter. LSU had a tougher schedule than Clemson, but LSU also allowed A LOT more points than Clemson did that year. This is how the model compares LSU to Clemson that year:
LSU: 33.3
Off: 23.7
Def: 9.6
Clemson: 33.5
Off: 15.8
Def: 17.7
See the difference? LSU was 7.9 points better on offense, but Clemson was 8.1 points better on defense.
This post was edited on 7/13/26 at 2:38 pm
Posted on 7/13/26 at 2:52 pm to MilkJug
quote:
Look at the rest of OSU's schedule that year. You're putting too much weight into one game.
No, I'm really not.
I don't know how better to explain this to you: your model doesn't do what its supposed to do.
A power model predicts or should predict outcomes. Given discrete and measurable data points within one of the few data ranges within which outcomes can be actually compared and measured your model is consistently failing to predict actual real life outcomes and it isn't even close.
You can argue that YOU think OSU and Clemson would have a different outcome, but that's your opinion. The actual experiment was run and a different outcome was achieved.
Therefore your model doesn't just seem strange or wrong to most observers, its not predicting actual outcomes because you've failed to properly train an algorithm which is capable of doing so.
If you cant even accurately predict or reflect outcomes that are observable why would anyone think it could accurately predict outcomes that aren't at a rate worth paying attention to?
Posted on 7/13/26 at 3:04 pm to tide06
quote:
A power model predicts or should predict outcomes.
which is exactly what my model's goal is.
quote:
Given discrete and measurable data points within one of the few data ranges within which outcomes can be actually compared and measured your model is consistently failing to predict actual real life outcomes and it isn't even close.
what "isn't even close" and how did you determine this? Or this your opinion?
quote:
You can argue that YOU think OSU and Clemson would have a different outcome, but that's your opinion.
Kind of like the one you just gave?
quote:
The actual experiment was run and a different outcome was achieved.
Again, Alabama and Georgia in 2021 - according to your logic, why did we even play the NC game that year? Alabama was clearly better, right?
Alabama beat Georgia 41-24 (at this point "the experiment was run")
then Georgia beat Alabama 33-18
My model predicts, if they played again, Georgia wins 29-24. It looks at the entire body of work for both teams - not just one outlier (41-24)
You are making zero sense and conflicting your own argument.
This post was edited on 7/13/26 at 3:06 pm
Posted on 7/13/26 at 3:10 pm to MilkJug
This is the dumbest ranking I've seen in my entire life
Posted on 7/13/26 at 3:16 pm to MilkJug
Oh also:
The fact that you don’t have over 100 downvotes tells me that it’s early July and no one is paying attention.
Your formula is absolutely fricking retarded and you should feel bad for posting this.
Next time maybe look at the results and ask yourself: is my model spitting out something absolutely fricking retarded?
If the answer is yes maybe just don’t post the results and tweak your model until it spits out something realistic.
The fact that you don’t have over 100 downvotes tells me that it’s early July and no one is paying attention.
Your formula is absolutely fricking retarded and you should feel bad for posting this.
Next time maybe look at the results and ask yourself: is my model spitting out something absolutely fricking retarded?
If the answer is yes maybe just don’t post the results and tweak your model until it spits out something realistic.
This post was edited on 7/13/26 at 3:19 pm
Posted on 7/13/26 at 3:26 pm to Tarpon08
quote:
Oh also:
The fact that you don’t have over 100 downvotes tells me that it’s early July and no one is paying attention.
Your formula is absolutely fricking retarded and you should feel bad for posting this.
Next time maybe look at the results and ask yourself: is my model spitting out something absolutely fricking retarded?
If the answer is yes maybe just don’t post the results and tweak your model until it spits out something realistic.
quote:
It doesn't care about who "deserves" what, it doesn't care about records, reputation, polls, or your feelings.
This post was edited on 7/13/26 at 3:26 pm
Posted on 7/13/26 at 3:35 pm to MilkJug
Look, I know you're trying very hard to defend the system from critique, but it's here and that's what it is.
If your model comes back and says a 2-loss Houston team that only played 3 ranked teams (losing 2) back in 1989 is one of the 25 best teams of the past 65 years, then I would pull the model back and figure out where that went wrong. I cannot think of any analytical model that would put that in the top 100.
Yes, predictive models can be flawed, and yes, they can also be upended by chaos on the field.
Sometimes, though, they are just plain wrong; when that happens, you can either double down and insist it isn't, or you can review the parameters and see if there is something skewing things.
Even the least biased model has some inherent bias because it's built upon things listed/created by humans, and we all have our own biases. We cannot even agree on the best team within our own lifetimes; we sure as heck won't be able to agree on which teams were more dominant to their own era.
If your model comes back and says a 2-loss Houston team that only played 3 ranked teams (losing 2) back in 1989 is one of the 25 best teams of the past 65 years, then I would pull the model back and figure out where that went wrong. I cannot think of any analytical model that would put that in the top 100.
Yes, predictive models can be flawed, and yes, they can also be upended by chaos on the field.
Sometimes, though, they are just plain wrong; when that happens, you can either double down and insist it isn't, or you can review the parameters and see if there is something skewing things.
Even the least biased model has some inherent bias because it's built upon things listed/created by humans, and we all have our own biases. We cannot even agree on the best team within our own lifetimes; we sure as heck won't be able to agree on which teams were more dominant to their own era.
Posted on 7/13/26 at 3:41 pm to skrayper
quote:
If your model comes back and says a 2-loss Houston team that only played 3 ranked teams (losing 2) back in 1989 is one of the 25 best teams of the past 65 years, then I would pull the model back and figure out where that went wrong. I cannot think of any analytical model that would put that in the top 100.
Oof. That’s not good
Posted on 7/13/26 at 3:42 pm to MilkJug
Sometimes, in an effort to solve things via mathematics, we lose sight of the obvious.
Anybody who has watched football knows that 2019 LSU, 2001 Miami, 2020 Alabama (Covid jokes aside), 1995 Nebraska, and probably 2004 USC were next level great. You can sprinkle in some other squads as well.
If this is the top 25 according to your metrics, fine. That just tells me the formula is dogshit and isn’t worthy of clicks. Discussing great teams using this formula is essentially a waste of time.
Anybody who has watched football knows that 2019 LSU, 2001 Miami, 2020 Alabama (Covid jokes aside), 1995 Nebraska, and probably 2004 USC were next level great. You can sprinkle in some other squads as well.
If this is the top 25 according to your metrics, fine. That just tells me the formula is dogshit and isn’t worthy of clicks. Discussing great teams using this formula is essentially a waste of time.
Posted on 7/13/26 at 4:01 pm to skrayper
quote:
Sometimes, though, they are just plain wrong; when that happens, you can either double down and insist it isn't, or you can review the parameters and see if there is something skewing things.
I haven't said I think my model is right and everyone else is wrong. I've provided the reasons why it came to the numbers it came to. Why are you so sure Houston being at #25 is so wrong? They had one of the best offenses we've ever seen. You can say "but they didn't play anyone", well.. prove it. My model considers who they played. That's why "one of the best offenses we've ever seen" is at #25 instead of top 10.
quote:
Even the least biased model has some inherent bias because it's built upon things listed/created by humans, and we all have our own biases.
That's not how these models work. The number one goal is predict the outcome of games. So, given the data it has (final scores, quarter scores, efficiencies, etc.), it iterates through the data and determines how things get weighted until it reaches the lowest possible MAE (mean average error - lower number is better) by testing itself against games it hasn't learned on. The MAE of Vegas spreads is usually around 12.5 points. My model is around 12.7 points. Any bias or data noise gets negatively reflected in the MAE and then rejected by the model.
quote:
We cannot even agree on the best team within our own lifetimes; we sure as heck won't be able to agree on which teams were more dominant to their own era.
I agree with this - so why are you saying my model is "wrong"? Its like you're stating that as a fact.. Again, I'm not saying my model is right. My model reflects what you would see from SP+, Sagarin, and ESPN FPI.
Posted on 7/13/26 at 4:29 pm to MilkJug
Ohio State might have pulled off a 3 peat had the Zach Smith foolishness not derailed Urban. The administration in Columbus, Indianapolis, and Bristol squawled like old women but crickets when the scandal in An Arbor came out.
Posted on 7/13/26 at 4:42 pm to MilkJug
There is nowhere on God's green earth that 2019 Ohio State - who lost a game - is the 3rd best team since 1960.
Posted on 7/13/26 at 5:12 pm to MilkJug
quote:
Do you suggest I not do a graphic anymore?
Could you just not post anymore? That would be better.
Posted on 7/13/26 at 5:52 pm to TigerScorpion
quote:
Could you just not post anymore? That would be better.
You're right. What was I thinking coming to this place with mathematics.
Posted on 7/13/26 at 5:53 pm to MilkJug
So, even though your model supposedly accounts for differences between eras, not a single team from the entire decade of the 60's cracked the top 25? That seems, um, strange.
Posted on 7/13/26 at 6:08 pm to TheTideMustRoll
quote:
So, even though your model supposedly accounts for differences between eras, not a single team from the entire decade of the 60's cracked the top 25? That seems, um, strange.
'66 Notre Dame was just outside the top 25. The next two teams from the 60s was 1969 Texas and 1968 Houston, but they were way outside the top 25 (like 80 something). Teams back then didn't put up big numbers like we've seen in the last ~30 years. Offenses have evolved.
Page 2 of 2
Back to top
Follow SECRant for SEC Football News