This highlights that SOR is actually a flawed metric.
UCLA ends up solidly #1 despite playing only 2 top-50 quality teams so far per FEI (Wisconsin and Oregon).
-Georgia is 9th at 6-1, having played 4 top-50 teams.
-Notre Dame is 12th at 7-1, with a loss to a better team than anyone UCLA played (BYU) and beating multiple common opponents they have with UCLA by more.
-Utah is 4th despite not having played a top-25 quality team. 2 top-50 opps. Best win was Houston by 8.
- Meanwhile Alabama is 15th at 6-2, with the #1 SOS in FEI, having played *7* top-50 teams and 3 top-15 teams. They are 1-2 vs top-15, 5-0 against teams in the quality range Houston has played (all double-digit wins).
Despite how it's promoted, the way the math works out SOR ends up biased far more to W-L records for top teams than actually reflecting wins vs SOS.
(not a shot on Dave for posting btw)
UCLA ends up solidly #1 despite playing only 2 top-50 quality teams so far per FEI (Wisconsin and Oregon).
-Georgia is 9th at 6-1, having played 4 top-50 teams.
-Notre Dame is 12th at 7-1, with a loss to a better team than anyone UCLA played (BYU) and beating multiple common opponents they have with UCLA by more.
-Utah is 4th despite not having played a top-25 quality team. 2 top-50 opps. Best win was Houston by 8.
- Meanwhile Alabama is 15th at 6-2, with the #1 SOS in FEI, having played *7* top-50 teams and 3 top-15 teams. They are 1-2 vs top-15, 5-0 against teams in the quality range Houston has played (all double-digit wins).
Despite how it's promoted, the way the math works out SOR ends up biased far more to W-L records for top teams than actually reflecting wins vs SOS.
(not a shot on Dave for posting btw)
Vegas Matrix Strength of Schedule Rankings top 25 thru week 9
20
3
119
14,380
The ultimate point, regardless of the math, is that the above SOR rankings don't track with what you'd want them to say if you look at the wins and losses of each team.
I try to explain why the math ends up calculating the wrong thing below:
x.com/CFBNerds/statu…
I try to explain why the math ends up calculating the wrong thing below:
x.com/CFBNerds/statu…
PSA: Strength of Record as a metric is a win-loss centric way of telling you where a given team should be ranked.
SOR of opponents is NOT a useful metric when talking about who a given team has beaten.
It's bad that we value wins off rankings and not ratings, but this is worse.
0
0
7
783
People freaking out about Day going for 2 are failing to understand a logic test very similar to the “Monty Hall” problem.
If you can’t understand this, you can’t understand Ryan Day’s decision:
-You stand in front of three closed doors.
-Behind one door is a brand-new car; behind the other two doors are goats.
-You pick one door (for example, Door 1).
-The host, Monty Hall—who knows what is behind every door—opens one of the remaining doors (say, Door 3) to reveal a goat.
-Monty then asks you a question: Do you want to switch your choice to the remaining unopened door (Door 2)?
You ALWAYS switch. Sticking with door 1 gives you 1/3 chance of winning, switching to door 2 gives a 2/3 chance.
The above is due to information. The host ruled out a set of possibilities for you, and it altered the odds for the set you didn’t pick.
People really struggle with this altered information game theory problem, even though it’s *absolutely* true as math.
Ryan Day going for 2 early was getting information on what they needed to do. That gives clarity on his path (2 more possessions to win vs 1 more) that all things equal improve his odds.
The exception is really only if you think going for 2 later is better than earlier due to “momentum.” If it isn’t, waiting it out loses the game when you miss it, but without time to correct. If momentum helps, then they aren’t equal odds and maybe it’s better.
Keep in mind even having an easier 2nd scoring drive before going for 2 doesn’t really matter - if you miss going for 2 but leave 15 seconds left you’re losing. The whole game state hinged on the 2-pt try.
From a vibes perspective, most people struggle with this because they feel like the longer you’re “in the game” the better off you are. That really isn’t true, for the same reason that having a 4 minute drive down 15 was the wrong call even if it likely scores, as it was better to take lower percentage shots to try and score in 2 minutes. Ohio State had “hope” to some, but to people like me the clock situation told me they’d already lost the game.
In view of the above, I’d probably do what Day did. The opposite is fine too - onside kicks are rarely converted due to rule changes over the last decade, and there was almost no time - but that’s based on some very fuzzy guesses on PAT conversion. Neither is a stupid approach.
But if you can’t understand the Monty Hall problem, you aren’t going to truly understand what Day was deciding, either.
If you can’t understand this, you can’t understand Ryan Day’s decision:
-You stand in front of three closed doors.
-Behind one door is a brand-new car; behind the other two doors are goats.
-You pick one door (for example, Door 1).
-The host, Monty Hall—who knows what is behind every door—opens one of the remaining doors (say, Door 3) to reveal a goat.
-Monty then asks you a question: Do you want to switch your choice to the remaining unopened door (Door 2)?
You ALWAYS switch. Sticking with door 1 gives you 1/3 chance of winning, switching to door 2 gives a 2/3 chance.
The above is due to information. The host ruled out a set of possibilities for you, and it altered the odds for the set you didn’t pick.
People really struggle with this altered information game theory problem, even though it’s *absolutely* true as math.
Ryan Day going for 2 early was getting information on what they needed to do. That gives clarity on his path (2 more possessions to win vs 1 more) that all things equal improve his odds.
The exception is really only if you think going for 2 later is better than earlier due to “momentum.” If it isn’t, waiting it out loses the game when you miss it, but without time to correct. If momentum helps, then they aren’t equal odds and maybe it’s better.
Keep in mind even having an easier 2nd scoring drive before going for 2 doesn’t really matter - if you miss going for 2 but leave 15 seconds left you’re losing. The whole game state hinged on the 2-pt try.
From a vibes perspective, most people struggle with this because they feel like the longer you’re “in the game” the better off you are. That really isn’t true, for the same reason that having a 4 minute drive down 15 was the wrong call even if it likely scores, as it was better to take lower percentage shots to try and score in 2 minutes. Ohio State had “hope” to some, but to people like me the clock situation told me they’d already lost the game.
In view of the above, I’d probably do what Day did. The opposite is fine too - onside kicks are rarely converted due to rule changes over the last decade, and there was almost no time - but that’s based on some very fuzzy guesses on PAT conversion. Neither is a stupid approach.
But if you can’t understand the Monty Hall problem, you aren’t going to truly understand what Day was deciding, either.
118
9
225
68,350
Most of the adamant, angry responses I’ve read do some version of:
1) guaranteeing in their mind an early 2 pt try fails, while
2) assuming a later 2 pt try might succeed.
You can’t mix what we know now, and what we don’t know because it didn’t happen, like that.
You either have to assume a later try fails, or justify why it’s demonstrably more likely to succeed. Else you’re being unfair to what Day knew at the time (he had to assume early or late tries were equally likely to fail).
Using their logic, playing to a 2 pt play to end the game is playing to a guaranteed loss (because we know Ohio State fails).
1) guaranteeing in their mind an early 2 pt try fails, while
2) assuming a later 2 pt try might succeed.
You can’t mix what we know now, and what we don’t know because it didn’t happen, like that.
You either have to assume a later try fails, or justify why it’s demonstrably more likely to succeed. Else you’re being unfair to what Day knew at the time (he had to assume early or late tries were equally likely to fail).
Using their logic, playing to a 2 pt play to end the game is playing to a guaranteed loss (because we know Ohio State fails).
3
0
21
3,631
Ohio State going for 2 there down 27-18 is *not* a situation I’d complain about as bad analytics.
Ohio State *must* go for 2 at least once. Doing it on the first try lets them know if they’re playing to get 1 more possession or 2.
Ohio State *must* go for 2 at least once. Doing it on the first try lets them know if they’re playing to get 1 more possession or 2.
139
8
302
49,744
The number of angry, stupid replies is insane.
Yes, you probably lose the game if you don’t convert. That happens either way. But trying earlier lets you know to save enough clock to at least *possibly* win if you fail.
The thing you’re really angry about is that Ohio State was almost certain to lose and Ryan Day down 15 only has bad options.
Yes, you probably lose the game if you don’t convert. That happens either way. But trying earlier lets you know to save enough clock to at least *possibly* win if you fail.
The thing you’re really angry about is that Ohio State was almost certain to lose and Ryan Day down 15 only has bad options.
17
1
72
8,432
The stupid thing the staff did on that drive was 100% going too slowly to work the field and build “momentum” to score, which didn’t leave enough clock.
Ironically, it’s stupid because it made the fans feel better about “staying in it” when in reality they were better off in an offense that scored quickly or not at all.
Ironically, it’s stupid because it made the fans feel better about “staying in it” when in reality they were better off in an offense that scored quickly or not at all.
5
1
42
5,880
Is Florida cursed with a dumb football team?
Or is Florida dumb with a cursed football team?
Or is Florida dumb with a cursed football team?
12
4
145
8,535
I’ve railed on the whole fake “analytics” on going for 2 for years.
The way it works is based on tracking what choice was made in a given game state in the past, and if they won as a result.
But that analysis is DEEPLY flawed. Here’s why:
The analogy would be if you’re stranded with no food and water and there’s a 6 foot ravine you could jump to safety.
People who try to jump the ravine have a much better chance of survival than those who don’t.
The *problem* is that people who doubt they make the jump usually won’t try.
The calculated risk aspect of those historical decisions is simply impossible to control for in data.
This is especially true where the decision maker tends to be conservative. People are scared to jump, coaches are scared to go for 2.
That means the ones who try usually have some good reason to expect success that won’t be reflected in your dataset (e.g., having a great play in the bag, having worn down their OL, being on a hot streak, etc.) and vice-versa (e.g., your RBs having a bad day or your QB having a bum ankle).
Simply put, all the “go for it on 4th or “go for 2” data sets out there are inherently biased to overestimate its helpfulness, and the economics dudebros that push the analysis don’t seem to understand it.
The way it works is based on tracking what choice was made in a given game state in the past, and if they won as a result.
But that analysis is DEEPLY flawed. Here’s why:
The analogy would be if you’re stranded with no food and water and there’s a 6 foot ravine you could jump to safety.
People who try to jump the ravine have a much better chance of survival than those who don’t.
The *problem* is that people who doubt they make the jump usually won’t try.
The calculated risk aspect of those historical decisions is simply impossible to control for in data.
This is especially true where the decision maker tends to be conservative. People are scared to jump, coaches are scared to go for 2.
That means the ones who try usually have some good reason to expect success that won’t be reflected in your dataset (e.g., having a great play in the bag, having worn down their OL, being on a hot streak, etc.) and vice-versa (e.g., your RBs having a bad day or your QB having a bum ankle).
Simply put, all the “go for it on 4th or “go for 2” data sets out there are inherently biased to overestimate its helpfulness, and the economics dudebros that push the analysis don’t seem to understand it.
20
5
113
12,430
I've long argued that 2-pt conversion win rates aren't actually reliable, because the historical data is skewed by coaches being conservative and only going for it when it's likely to 1) work, and 2) win them the game.
This is similar. Coaches set up passing to have higher EPA.
This is similar. Coaches set up passing to have higher EPA.
I firmly believe that the reason we got into that "throw every down because EPA go brrrrr" era a while back is because we didn't stop to consider that the reason why passing EPA went brrrr in the first place is because the run game set it up to go brrrr.
6
2
65
9,081
There is a circular logic to the statistics most don't seem to appreciate (understandable with abstract math).
The math says coaches who go for 2 are more successful.
But it may be that coaches who are *being* successful go for 2 more.
It's almost impossible to untangle.
The math says coaches who go for 2 are more successful.
But it may be that coaches who are *being* successful go for 2 more.
It's almost impossible to untangle.
2
1
17
1,944
Model update for this week.
LSU is still #2 after handling Auburn 38-17. The AP has them #15.
The biggest story this time of year stems from what the model can't see - like injuries. 🧵
LSU is still #2 after handling Auburn 38-17. The AP has them #15.
The biggest story this time of year stems from what the model can't see - like injuries. 🧵
15
7
223
30,571
The model and the AP are converging. 22 of the AP's top 25 are now in our top 25, up from 18 two weeks ago.
Biggest remaining gaps:
Model lower: Florida (#17 AP, #37 model), Boise State (#20, #40), Tennessee (#16, #28)
Model higher: Arizona (#22, #8), LSU (#15, #2), Wake Forest (#14, #6)
Biggest remaining gaps:
Model lower: Florida (#17 AP, #37 model), Boise State (#20, #40), Tennessee (#16, #28)
Model higher: Arizona (#22, #8), LSU (#15, #2), Wake Forest (#14, #6)
2
0
9
2,282
We aren't going to know much about the Big 12 until the postseason.
Every non-conference game they've played: 6-7 against Power 4 teams and Notre Dame, 34-1 against everyone else.
That's a statistical silo. It's like comparing a photo of the tallest mountain in the Alps to the Without knowing how the ranges compare, it's easy to pick the wrong one.
Every non-conference game they've played: 6-7 against Power 4 teams and Notre Dame, 34-1 against everyone else.
That's a statistical silo. It's like comparing a photo of the tallest mountain in the Alps to the Without knowing how the ranges compare, it's easy to pick the wrong one.
1
0
8
1,985