Ratings & Models postmortem

[RonJohn]

Yeah, Yards/Drive doesn't line up with both teams' total yards and total drives, and you can't make the numbers work by tossing out drives either. Something's amiss here. Also, there are other apparent errors in the raw data that call into question the overall integrity of the advanced stats for this game. The ESPN stats don't include King's interception, and the Game on Paper advanced stats that @slugboy reproduced here (which use ESPN data) show only 3 TFLs, yet ESPN counts 6. I thought perhaps ESPN was overcounting TFLs when multiple players participated in the tackle, but that doesn't line up either.

Even more than that, if there are ten drives each, it is impossible to have 56.87 yards/drive mathematically. If the denominator is 10, and the numerator is an even number there will not be anything in the hundreds place. 56.8 is possible, but not anything XX.X7. It can't be penalties either that caused the issue. GT had 8 penalties for 65 yards. That includes both offense and defense. If you attribute ALL of those to the offense it still isn't 568 yards. (463+65=528) Something is wrong with the calculations, or the description of the stat.

 

[GTG768B]

Key will patch up the Center issues. Just some week 1 nonsense, glad we were able to overcome it and get the W.

I can't tell you how awesome it is to have confidence that our coach will fix a problem.

 

[Root4GT]

Key will patch up the Center issues. Just some week 1 nonsense, glad we were able to overcome it and get the W.

It was far more than just the snaps. That is the easy fix. Actually doing a good job pass blocking is far more of a concern.

Good thing we have a FCS Team this week to work on the OL pass protection issues!

 

[slugboy]

How is the Yards/Drive calculated? It says each team had 10 drives, and GT-56.87, CO-40.92 Yards/Drive. However GT had 463 yards, not 568.

I looked at the drive data. It's still different than what ESPN has as total numbers but it matches their drive information (458 yards for GT, 305 for Colorado). So, either ESPN's totals are wrong, or their drive information is wrong, I guess.

GT:

Total Yards	458​
Average Drive Yards	45.8​

Start	Offense	Result	Description	Yards	EPA	Start WP%	End WP%	Score
Q1 15:00	GT	Fumble	2 plays, 11 yards, 0:25	11	-4.85	62.30%	54.70%	0-0
Q1 12:08	GT	Fumble	4 plays, 27 yards, 1:46	27	-3.29	44.60%	40.50%	0-7
Q1 9:48	GT	Interception	6 plays, 38 yards, 2:51	38	2.57	43.80%	38.80%	0-7
Q1 4:55	GT	Field Goal	13 plays, 84 yards, 6:30	84	4.55	42.40%	53.00%	3-7
Q2 10:25	GT	Touchdown	10 plays, 80 yards, 5:57	80	6.33	54.20%	70.80%	10-7
Q2 1:39	GT	Field Goal	9 plays, 49 yards, 1:39	49	2.05	70.40%	70.10%	13-7
Q3 12:54	GT	Punt	6 plays, 14 yards, 2:39	14	-3.42	70.90%	62.90%	13-7
Q3 6:00	GT	Touchdown	11 plays, 75 yards, 5:46	75	8.16	61.20%	84.30%	20-13
Q4 8:04	GT	Punt	4 plays, 19 yards, 2:45	19	-1.56	61.10%	50.20%	20-20
Q4 2:43	GT	Touchdown	5 plays, 61 yards, 1:44	61	4.87	73.50%	97.70%	27-20

Colorado:

Total Yards	305​
Average Drive Yards	30.5​

Start	Offense	Result	Description	Yards	EPA	Start WP%	End WP%	Score
Q1 14:20	COL	Touchdown	5 plays, 36 yards, 2:27	36	3.23	45.30%	54.10%	0-7
Q1 10:18	COL	Punt	3 plays, 0 yards, 0:27	0	-3.24	59.50%	56.20%	0-7
Q1 6:51	COL	Punt	3 plays, 9 yards, 2:03	9	-1.24	57.80%	57.60%	0-7
Q2 13:31	COL	Punt	5 plays, 23 yards, 3:01	23	-2.2	52.50%	45.80%	3-7
Q2 4:33	COL	Field Goal	8 plays, 51 yards, 2:54	51	2.38	31.80%	41.10%	10-10
Q3 15:00	COL	Punt	4 plays, 25 yards, 2:06	25	-1.34	30.20%	29.10%	13-10
Q3 10:02	COL	Field Goal	8 plays, 39 yards, 4:15	39	0.6	37.10%	43.40%	13-13
Q3 0:14	COL	Touchdown	15 plays, 75 yards, 6:49	75	6.81	18.80%	46.60%	20-20
Q4 5:04	COL	Punt	5 plays, 22 yards, 2:28	22	-2.59	49.80%	26.50%	20-20
Q4 1:07	COL	End of Game	6 plays, 25 yards, 1:07	25	-0.72	6.30%	0.00%	27-20

 

[RonJohn]

I looked at the drive data. It's still different than what ESPN has as total numbers but it matches their drive information (458 yards for GT, 305 for Colorado). So, either ESPN's totals are wrong, or their drive information is wrong, I guess.

GT:

Total Yards	458​
Average Drive Yards	45.8​

Start	Offense	Result	Description	Yards	EPA	Start WP%	End WP%	Score
Q1 15:00	GT	Fumble	2 plays, 11 yards, 0:25	11	-4.85	62.30%	54.70%	0-0
Q1 12:08	GT	Fumble	4 plays, 27 yards, 1:46	27	-3.29	44.60%	40.50%	0-7
Q1 9:48	GT	Interception	6 plays, 38 yards, 2:51	38	2.57	43.80%	38.80%	0-7
Q1 4:55	GT	Field Goal	13 plays, 84 yards, 6:30	84	4.55	42.40%	53.00%	3-7
Q2 10:25	GT	Touchdown	10 plays, 80 yards, 5:57	80	6.33	54.20%	70.80%	10-7
Q2 1:39	GT	Field Goal	9 plays, 49 yards, 1:39	49	2.05	70.40%	70.10%	13-7
Q3 12:54	GT	Punt	6 plays, 14 yards, 2:39	14	-3.42	70.90%	62.90%	13-7
Q3 6:00	GT	Touchdown	11 plays, 75 yards, 5:46	75	8.16	61.20%	84.30%	20-13
Q4 8:04	GT	Punt	4 plays, 19 yards, 2:45	19	-1.56	61.10%	50.20%	20-20
Q4 2:43	GT	Touchdown	5 plays, 61 yards, 1:44	61	4.87	73.50%	97.70%	27-20

But the question still remains: From your original post, where does the 56.87 Yards/Drive come from?

 

[slugboy]

But the question still remains: From your original post, where does the 56.87 Yards/Drive come from?

I’m not sure, but it’s almost 458/9. I would guess that somewhere in the data feed the total yards and the number of drives is off

 

[roadkill]

But the question still remains: From your original post, where does the 56.87 Yards/Drive come from?

I’m not sure, but it’s almost 458/9. I would guess that somewhere in the data feed the total yards and the number of drives is off

To see if the stats for the Colorado game were just a data anomaly, I checked a couple of last year’s games. Same issues with yards/drive and plays/drive. You can’t make the numbers work, and to the point about the two decimal places, working backwards never results in an integer for plays or yards.

Game on Paper’s glossary defines less intuitive terms like EPA, but does not provide much insight into their yards/drive or plays/drive numbers, likely because the math should be obvious. Problem, is, their results cannot be obtained by simple arithmetic. It seems that there may be some more complex math used where the results don’t line up with typical assumptions.

I feel like the overall stats are still directionally correct, but it makes one wonder why they can’t get simple things like Yards/Drive right.

 

[slugboy]

To see if the stats for the Colorado game were just a data anomaly, I checked a couple of last year’s games. Same issues with yards/drive and plays/drive. You can’t make the numbers work, and to the point about the two decimal places, working backwards never results in an integer for plays or yards.

Game on Paper’s glossary defines less intuitive terms like EPA, but does not provide much insight into their yards/drive or plays/drive numbers, likely because the math should be obvious. Problem, is, their results cannot be obtained by simple arithmetic. It seems that there may be some more complex math used where the results don’t line up with typical assumptions.

I feel like the overall stats are still directionally correct, but it makes one wonder why they can’t get simple things like Yards/Drive right.

 

[GTRhino24]

To see if the stats for the Colorado game were just a data anomaly, I checked a couple of last year’s games. Same issues with yards/drive and plays/drive. You can’t make the numbers work, and to the point about the two decimal places, working backwards never results in an integer for plays or yards.

Game on Paper’s glossary defines less intuitive terms like EPA, but does not provide much insight into their yards/drive or plays/drive numbers, likely because the math should be obvious. Problem, is, their results cannot be obtained by simple arithmetic. It seems that there may be some more complex math used where the results don’t line up with typical assumptions.

I feel like the overall stats are still directionally correct, but it makes one wonder why they can’t get simple things like Yards/Drive right.

There are tiny fractions of a yard left at the end of every play. We just round those down and deposit them into an account. Kinda like Superman 3. It’s kinda foolproof.

 

[RonJohn]

To see if the stats for the Colorado game were just a data anomaly, I checked a couple of last year’s games. Same issues with yards/drive and plays/drive. You can’t make the numbers work, and to the point about the two decimal places, working backwards never results in an integer for plays or yards.

Game on Paper’s glossary defines less intuitive terms like EPA, but does not provide much insight into their yards/drive or plays/drive numbers, likely because the math should be obvious. Problem, is, their results cannot be obtained by simple arithmetic. It seems that there may be some more complex math used where the results don’t line up with typical assumptions.

I feel like the overall stats are still directionally correct, but it makes one wonder why they can’t get simple things like Yards/Drive right.

That is my concern. If simple calculations are wrong, it calls all of the calculated stats into question, which makes the source unreliable.

 

[roadkill]

That is my concern. If simple calculations are wrong, it calls all of the calculated stats into question, which makes the source unreliable.

I've emailed the Game on Paper site developer with our concerns; if I get a response, I'll share it.

 

[Heisman's Ghost]

I think it actually moves the needle as much as ever since they won 9 last year. It definitely didn’t for most of 2023 since they got waxed by Oregon four games into Deion’s tenure.

That was with a Heisman winner and a NFL quarterback. Winning 9 games in the Big 12 does not mean that much. Colorado got its head handed to them by both Kansas and BYU. We spotted them a 3-0 turnover margin and they still lost. Teams with a negative 3 turnover margin like Tech had win less than 1% of the time.

 

[Heisman's Ghost]

PFF grades are out.

PFF Grades and Snap Counts For Georgia Tech In Their Win Over Colorado

Georgia Tech was tested early and often on the road tonight against Colorado, but they found a way to overcome them and get a crucial road win to start the seas

www.si.com

Buzz, how accurate do you think these grades are? Just curious. Disclaimer, I would not know if they were accurate or not if they hit me upside the head with the grades.

 

[BuzzGT]

 

[slugboy]

SP+ is no longer paywalled.

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

We moved from 43rd to 36th. We are two sports behind Duke, who moved up 11 sports due to their win over Elon. Our offense is to 20. Our defense is ranked 59th, and special teams it's 100th. There is a ton of bias from last year in this model, and it will be there for weeks.

 

[roadkill]

There is a ton of bias from last year in this model, and it will be there for weeks

With the current amount of roster turnover seen across the sport, bias from last year's data makes SP+ an even less reliable predictor in the early season. IIWII.

 

[slugboy]

With the current amount of roster turnover seen across the sport, bias from last year's data makes SP+ an even less reliable predictor in the early season. IIWII.

I'm posting it more so people can watch it move, and know where we started

(it’s kind of like watching the stock market at work. unless you’re day trading, it’s a distraction. Hopefully, it a fun distraction)

 

[bobongo]

There are tiny fractions of a yard left at the end of every play. We just round those down and deposit them into an account. Kinda like Superman 3. It’s kinda foolproof.

I was thinking of Office Space.

 

[stinger78]

SP+ is no longer paywalled.

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

We moved from 43rd to 36th. We are two sports behind Duke, who moved up 11 sports due to their win over Elon. Our offense is to 20. Our defense is ranked 59th, and special teams it's 100th. There is a ton of bias from last year in this model, and it will be there for weeks.

Thank you for pointing this out. Bias is epic in predictive sports statistics, IMPO.

 

[TechPhi97]

Thank you for pointing this out. Bias is epic in predictive sports statistics, IMPO.

It’s the only way ESPN can have “Top 10 Matchups!” when no games have been played.