Ratings & Models postmortem

[slugboy]

This is back. It’s not explosive plays, or turnovers, or anything other than how good one team is at keeping ahead of the chains vs the other.

So, what’s it say?

It says South Carolina beat VT off of big plays—a TD punt return, two interceptions, a 64 yard pass. Sometimes fans say “if it hadn’t been for a couple of fluke plays, we would have won”. VT and Syracuse would’ve still lost, but it would have been closer.

Texas was a couple of plays away from winning—while Clemson got beat on a down by down basis.

This chart is the graphical version of “if Georgia Tech doesn’t give away three turnovers, that game would have been a beat down”.

(But, yeah, Ohio State did really beat Texas)

It’s from a guy named Parker Fleming

 

[Sarrick]

This is back. It’s not explosive plays, or turnovers, or anything other than how good one team is at keeping ahead of the chains vs the other.

So, what’s it say?

It says South Carolina beat VT off of big plays—a TD punt return, two interceptions, a 64 yard pass. Sometimes fans say “if it hadn’t been for a couple of fluke plays, we would have won”. VT and Syracuse would’ve still lost, but it would have been closer.

Texas was a couple of plays away from winning—while Clemson got beat on a down by down basis.

This chart is the graphical version of “if Georgia Tech doesn’t give away three turnovers, that game would have been a beat down”.

(But, yeah, Ohio State did really beat Texas)

It’s from a guy named Parker Fleming

View attachment 19197

Love this. We had the third most dominating win of the p4 vs p4 games (cough chapel bill falling apart already and Nico was awful for UCLA cough)

But it makes sense if you watched the game. Take away penalties and turnovers we had little to no negative plays and a high amount of like 5-7 yard gains. They could not stop us

 

[slugboy]

This is back. It’s not explosive plays, or turnovers, or anything other than how good one team is at keeping ahead of the chains vs the other.

So, what’s it say?

It says South Carolina beat VT off of big plays—a TD punt return, two interceptions, a 64 yard pass. Sometimes fans say “if it hadn’t been for a couple of fluke plays, we would have won”. VT and Syracuse would’ve still lost, but it would have been closer.

Texas was a couple of plays away from winning—while Clemson got beat on a down by down basis.

This chart is the graphical version of “if Georgia Tech doesn’t give away three turnovers, that game would have been a beat down”.

(But, yeah, Ohio State did really beat Texas)

It’s from a guy named Parker Fleming

View attachment 19197

UNC and UCLA are the most massive P4 beat downs. They’re almost equivalent.

Everyone saw UNC, but I doubt many fans are aware that UCLA was as thoroughly thrashed

 

[slugboy]

I just meant this thread for the stats for the Colorado game, but it’s gotten a life of its own

Anyway, the game on paper team profiles are out. We played a good team, so that should affect the numbers (they aren’t opponent-adjusted as far as I know)

Georgia Tech | 2025 | Game on Paper

Advanced stats for Georgia Tech during the 2025 season

gameonpaper.com

Is just one game, so I don’t know how much this is “Colorado has a good offense or a bad defense”. I would hope the GW game distorts this in a flattering manner for us.

No big passing plays for us. I shudder to think about the teams that had more explosive running plays than us—maybe Utah?

 

[jgtengineer]

Im not sure i agree with the passing percentage of explosives.
We only had 13 completions on 20 total passes whiel allowing 1 sack (which was a draw but eh it happens) so of 21 total drobacks we had 6.8 yards per attempt and 11 YPC.

However
baily had a catch fo 23 yards,
Haynes had one of 27
Hosley had one of 24
half of our passing yards came on three receptions. Pretty much our ownly success in the passing game happened on "explosive" 20+ yard plays.

 

[roadkill]

Im not sure i agree with the passing percentage of explosives.
We only had 13 completions on 20 total passes whiel allowing 1 sack (which was a draw but eh it happens) so of 21 total drobacks we had 6.8 yards per attempt and 11 YPC.

However
baily had a catch fo 23 yards,
Haynes had one of 27
Hosley had one of 24
half of our passing yards came on three receptions. Pretty much our ownly success in the passing game happened on "explosive" 20+ yard plays.

I had the same thought, so I looked at Game on Paper to find their definition of "explosive". For pass plays, they use EPA >2.4. Our highest EPA pass was Hosley's 24-yard catch, which had an EPA of 2.07, so it didn't quite qualify. That's a high bar. On the other hand, we had 7 explosive runs (EPA>1.8).

 

[57jacket2]

Nearly 4 million viewers take in Georgia Tech’s win at Colorado

The Yellow Jackets could play in front of more big audiences this season with a home game against Clemson on the horizon and Clean Old-Fashioned Hate scheduled for Nov. 28.

www.ajc.com

 

[slugboy]

Im not sure i agree with the passing percentage of explosives.
We only had 13 completions on 20 total passes whiel allowing 1 sack (which was a draw but eh it happens) so of 21 total drobacks we had 6.8 yards per attempt and 11 YPC.

However
baily had a catch fo 23 yards,
Haynes had one of 27
Hosley had one of 24
half of our passing yards came on three receptions. Pretty much our ownly success in the passing game happened on "explosive" 20+ yard plays.

It's based on EPA. Rushing plays are explosive when EPA > 1.8. Pass plays have a higher standard and need to be greater than 2.4.

Stockton’s 23 yard reception had an EPA of 1.43.
Jamal Haynes’ 27 yard catch had an EPA of 1.77.
Malachi’s 23 yard catch had an EPA of 2.07–that got us down to the 23 yard line.

=============

Don’t read too much into this—it’s just one game. I think some of the stats may still be off (it’s missing an INT in one set of stats

On offense, we look really good. We don’t have any 40 yard passes or 30 yard passes for TDs, so thats the gap where we’re towards the bottom. Any metric that’s toward the center of the radar is stack ranked low against about 135 FBS teams. If you’re on the outer ring, you’re toward the top in that metric.

We played Colorado. Texas played Ohio State. Wake played KSU, and somebody played UMass, and a bunch of teams played FCS teams. We’re stack ranked against other teams—some played stronger competition, but most played much weaker competition.

On offense, we were comparatively mediocre passing (remember, some other team played Western Carolina and loaded up their stats). We were nearly flexbone efficient rushing the ball—that won’t always happen.

We played pretty dang well against Colorado, except for three turnovers, which are hurting our scores here.

Offensive
Plays	68
Plays/Game	68 T-#65
Total EPA	10.14 #52
EPA/Play	0.15 #50
EPA/Game	10.14 #50
Success Rate	52.9% #19
Starting FP	Own 20 #128

When Passing
Plays	21
Plays/Game	21 T-#120
Total EPA	2.90 #61
EPA/Play	0.14 #57
EPA/Game	2.90 #61
Success Rate	47.6% #49

When Rushing
Plays	47
Plays/Game	47 #16
Total EPA	7.23 #36
EPA/Play	0.15 #47
EPA/Game	7.23 #36
Success Rate	55.3% #16

Defense doesn’t look as good. The tables are easier to read. We gave up 20 points while some other schools were playing Ball State. These are below average stats, but we played an above average team. I think this is a “wait and see” situation.

Defensive
Plays	59
Plays/Game	59 T-#45
Total EPA	1.47 #94
EPA/Play	0.02 #94
EPA/Game	1.47 #94
Success Rate	37.3% #67
Starting FP	Own 32 #121

Against the Pass
Plays	30
Plays/Game	30 T-#57
Total EPA	-1.33 #79
EPA/Play	-0.04 #78
EPA/Game	-1.33 #78
Success Rate	36.7% T-#64

Against the Run
Plays	29
Plays/Game	29 T-#54
Total EPA	2.80 #99
EPA/Play	0.10 #102
EPA/Game	2.80 #100
Success Rate	37.9% #74

 

[roadkill]

I just meant this thread for the stats for the Colorado game, but it’s gotten a life of its own

Anyway, the game on paper team profiles are out. We played a good team, so that should affect the numbers (they aren’t opponent-adjusted as far as I know)

Georgia Tech | 2025 | Game on Paper

Advanced stats for Georgia Tech during the 2025 season

gameonpaper.com

Is just one game, so I don’t know how much this is “Colorado has a good offense or a bad defense”. I would hope the GW game distorts this in a flattering manner for us.

No big passing plays for us. I shudder to think about the teams that had more explosive running plays than us—maybe Utah?

View attachment 19200

Defensive chart doesn't align with my overall impression of our defensive performance in the Colorado game. We forced 5 punts on 10 drives, 305 yards total, and CU only had 3 explosive plays out of 59 total. Yet our percentiles are very low on everything but Pass Expl %, where we allowed only 1.
It almost seems like absolute percentages, rather than percentiles, are being used for some of the stats. For example, our havoc rate was 5% of the total plays, but as shown, it appears to be 5th percentile (or near zero). No way any team could have anywhere close to 100% havoc plays unless we are playing Cumberland.

Edit: If the percentiles are based on just the data for the season so far, as opposed to historical reference data, it's possible they could be skewed due to the small data set.

 

[slugboy]

Defensive chart doesn't align with my overall impression of our defensive performance in the Colorado game. We forced 5 punts on 10 drives, 305 yards total, and CU only had 3 explosive plays out of 67 total. Yet our percentiles are very low on everything but Pass Expl %, where we allowed only 1.
It almost seems like absolute percentages, rather than percentiles, are being used for some of the stats. For example, our havoc rate was 5% of the total plays, but as shown, it appears to be 5th percentile (or near zero). No way any team could have anywhere close to 100% havoc plays unless we are playing Cumberland.

Maybe another email is a good idea.

It does feel a little low on the defensive side to me. Based on the tables, we are bottom quartile in a lot of areas, and that doesn’t feel right

 

[stinger78]

It's based on EPA. Rushing plays are explosive when EPA > 1.8. Pass plays have a higher standard and need to be greater than 2.4.

Stockton’s 23 yard reception had an EPA of 1.43.
Jamal Haynes’ 27 yard catch had an EPA of 1.77.
Malachi’s 23 yard catch had an EPA of 2.07–that got us down to the 23 yard line.

=============

Don’t read too much into this—it’s just one game. I think some of the stats may still be off (it’s missing an INT in one set of stats

View attachment 19212

On offense, we look really good. We don’t have any 40 yard passes or 30 yard passes for TDs, so thats the gap where we’re towards the bottom. Any metric that’s toward the center of the radar is stack ranked low against about 135 FBS teams. If you’re on the outer ring, you’re toward the top in that metric.

We played Colorado. Texas played Ohio State. Wake played KSU, and somebody played UMass, and a bunch of teams played FCS teams. We’re stack ranked against other teams—some played stronger competition, but most played much weaker competition.

On offense, we were comparatively mediocre passing (remember, some other team played Western Carolina and loaded up their stats). We were nearly flexbone efficient rushing the ball—that won’t always happen.

We played pretty dang well against Colorado, except for three turnovers, which are hurting our scores here.

Offensive
Plays	68
Plays/Game	68 T-#65
Total EPA	10.14 #52
EPA/Play	0.15 #50
EPA/Game	10.14 #50
Success Rate	52.9% #19
Starting FP	Own 20 #128

When Passing
Plays	21
Plays/Game	21 T-#120
Total EPA	2.90 #61
EPA/Play	0.14 #57
EPA/Game	2.90 #61
Success Rate	47.6% #49

When Rushing
Plays	47
Plays/Game	47 #16
Total EPA	7.23 #36
EPA/Play	0.15 #47
EPA/Game	7.23 #36
Success Rate	55.3% #16

View attachment 19214

Defense doesn’t look as good. The tables are easier to read. We gave up 20 points while some other schools were playing Ball State. These are below average stats, but we played an above average team. I think this is a “wait and see” situation.

Defensive
Plays	59
Plays/Game	59 T-#45
Total EPA	1.47 #94
EPA/Play	0.02 #94
EPA/Game	1.47 #94
Success Rate	37.3% #67
Starting FP	Own 32 #121

Against the Pass
Plays	30
Plays/Game	30 T-#57
Total EPA	-1.33 #79
EPA/Play	-0.04 #78
EPA/Game	-1.33 #78
Success Rate	36.7% T-#64

Against the Run
Plays	29
Plays/Game	29 T-#54
Total EPA	2.80 #99
EPA/Play	0.10 #102
EPA/Game	2.80 #100
Success Rate	37.9% #74

Yes, when n = 1, and that 1 represents a broad ranging quality from team to team, these stats are simply a starting point.

 

[roadkill]

Maybe another email is a good idea.

It does feel a little low on the defensive side to me. Based on the tables, we are bottom quartile in a lot of areas, and that doesn’t feel right

I don't mind firing off another email, but I would prefer that we have something more concrete to point out, like the Yards/Drive issue. I concur it doesn't "feel" right, but I'd rather not send Akshay down a rabbit hole in search of something nebulous. Especially if the cause is simply a relatively small number of P4 matchups.

 

[HurricaneJacket]

I don't mind firing off another email, but I would prefer that we have something more concrete to point out, like the Yards/Drive issue. I concur it doesn't "feel" right, but I'd rather not send Akshay down a rabbit hole in search of something nebulous. Especially if the cause is simply a relatively small number of P4 matchups.

I would assume N=1 is the issue. There were relatively few P4 vs P4 games or even P4 vs G6 games last week, so talent discrepancies abounded and those will show up in defensive stats (e.g. a team can have a higher havoc rate when their line comes unblocked more plays due to talent differences between FBS and FCS)

 

[RonJohn]

FPI is a joke. It is proprietary and their algorithms are not public. As far as I know not even a high level explanation of their calculations is public. Everything I see tells me that it is geared towards ranking teams high that the SEC wants to be ranked high. I don't doubt that they "adjust" the calculations week to week.

Alabama lost to the team that is currently 40th in the FPI (was 65th before) and is still in the top 15. Texas lost to Ohio State, FPI says less chance of making playoffs that Ohio State, FPI says less chance of making NC game than Ohio State, but is ranked higher than Ohio State and has a better chance of winning the NC than Ohio State. It makes no sense. Especially, how can Texas be calculated to have a lower chance to get to the NC game, but a higher chance of winning the NC game.

Even more indictment against the FPI: GT beat Clemson. Stats were about as even as you could possibly get them. Clemson rose two spots in the FPI, while GT went down two spots. Clemson actually leapt over GT by losing the game. If there are any actual metrics that could cause that to happen, I would be surprised. If there are any useful metrics that could cause that to happen, I would be flabbergasted.

 

[stinger78]

Even more indictment against the FPI: GT beat Clemson. Stats were about as even as you could possibly get them. Clemson rose two spots in the FPI, while GT went down two spots. Clemson actually leapt over GT by losing the game. If there are any actual metrics that could cause that to happen, I would be surprised. If there are any useful metrics that could cause that to happen, I would be flabbergasted.

As one who spends his whole life (just about) in metrics, there are good metrics and there are lousy metrics. Some accurately describe the success/failure of a process, and others don't. I don't know FPI enough to evaluate it, but if this happens then it would be highly suspect in my book.

 

[roadkill]

Even more indictment against the FPI: GT beat Clemson. Stats were about as even as you could possibly get them. Clemson rose two spots in the FPI, while GT went down two spots. Clemson actually leapt over GT by losing the game. If there are any actual metrics that could cause that to happen, I would be surprised. If there are any useful metrics that could cause that to happen, I would be flabbergasted.

I reached the same conclusions (posted in the ranking projection thread). I thought maybe their algorithm was okay, but they used bad data. Game on Paper confirmed that ESPN’s play-by-play dataset was full of errors for our G-W game. However, I cannot think of any logical reason the algorithm would drop us and raise Clemson after yesterday. Game on Paper’s stats for yesterday look reasonable.

 

[Thwg777]

Allow me to put my tinfoil hat on for a sec…

Sports betting is pervasive. These models are intentionally awful as lemmings follow them to place their bets.

 

[slugboy]

If you're unhappy about ESPN's FPI, don't look at SP+. We fell four spots in Bill Connelly's SP+, behind Pitt who fell one spot, and several spots behind Clemson who didn't fall at all

2025 college football SP+ rankings for all 136 FBS teams

The final SP+ rankings, plus strength of schedule and résumé SP+, for the 2025 season.

www.espn.com

 

[billga99]

If you're unhappy about ESPN's FPI, don't look at SP+. We fell four spots in Bill Connelly's SP+, behind Pitt who fell one spot, and several spots behind Clemson who didn't fall at all

2025 college football SP+ rankings for all 136 FBS teams

The final SP+ rankings, plus strength of schedule and résumé SP+, for the 2025 season.

www.espn.com

Didn't number #33 FSU dominate #4 Alabama? What a joke.

 

[roadkill]

We fell four spots in Bill Connelly's SP+, behind Pitt who fell one spot, and several spots behind Clemson who didn't fall at all

From their definition: "It is simply a measure of the most sustainable and predictable aspects of football. If you're lucky or unimpressive in a win, your rating will probably fall. If you're strong and unlucky in a loss, it will probably rise."

I guess we got lucky Saturday.