Today in Analytics...

[Skeptic]

Sagarin has us in at #68 in the Predictor (the score that actually works), and Clemson in at #5, after last night’s game. Which means that last night’s game was a heckuva anomaly for them, but not enough to really dent the model:

College Football Ratings Page

View attachment 11209

FEI hasn’t updated yet, but we’re going to be in about the same place. Priors have a lot to do with SP+, FEI, and other ratings, and it’s gonna take a heckuva streak to change the perceptions of how good we are, even among our own fans.

I am filling formal protest about dirty talk on this board.

 

[leatherneckjacket]

Ooof!

 

[presjacket]

Ooof!

View attachment 13325

So we aren’t the worst offense!

 

[leatherneckjacket]

So we aren’t the worst offense!

Thank God for Rutgers and VPISU.

 

[CEB]

Ooof!

View attachment 13325

Maybe the point spread with UVA is 3 because they don’t expect anyone to score...

 

[slugboy]

Hmmm

 

[ibeattetris]

Hmmm

From BCF Toys

We are doing well when teams have a long field, but we are still giving up too many points in medium and short drives. He is right that the total points isn't great, but we need to do better on the medium drives.

This is also heavily influenced by earlier games this year, and hopefully we see the same fire the defense has had the last few weeks.

 

[slugboy]

View attachment 13329
From BCF Toys

We are doing well when teams have a long field, but we are still giving up too many points in medium and short drives. He is right that the total points isn't great, but we need to do better on the medium drives.

This is also heavily influenced by earlier games this year, and hopefully we see the same fire the defense has had the last few weeks.

I wonder what it would be like with just Ole Miss removed.

 

[slugboy]

The composite is back. At this point, I think all preseason information is out of the various different frameworks. We're still behind UVA in most of them--some rate our team higher than others. We're still digging out of a big hole of underperformance in our first four games, vs two games of better-than-expected performance lately.
If I was Clemson, USC, and TCU, I'd be sweating about my ability to get into the college football playoff. Going undefeated is going to be job 1--but they need to start putting a beatdown on other teams.

Here's how the various ratings factor in (we're about 3/5ths down the page, right over Ga Southern)

 

[slugboy]

Also, we're slightly more likely to pass than to run, based on how effective we are at passing. Given our passing game, we're almost exactly at the right run/pass mix.
(Looks like Kansas should actually throw it more?)

 

[85Escape]

We're still digging out of a big hole of underperformance in our first four games, vs two games of better-than-expected performance lately.
If I was Clemson, USC, and TCU, I'd be sweating about my ability to get into the college football playoff. Going undefeated is going to be job 1--but they need to start putting a beatdown on other teams.

When I look at our three losses, they were all three to teams in the 10 - 20 band (and on the upper side of it generally.) When I look at UVa's losses, two were to teams in the lower part of the 10 - 20 range, and the other two were Duke and Louisville, both in the 0 - 10 band. I'm not sure how these statistics work, but something seems off to me that UVa is still above us.

 

[slugboy]

When I look at our three losses, they were all three to teams in the 10 - 20 band (and on the upper side of it generally.) When I look at UVa's losses, two were to teams in the lower part of the 10 - 20 range, and the other two were Duke and Louisville, both in the 0 - 10 band. I'm not sure how these statistics work, but something seems off to me that UVa is still above us.

Clemson, UCF, and Ole Miss all won by larger margins than predicted, so they lowered our ratings. The Duke game was really tight, and wouldn't have moved our margin. The Pitt game where we won despite being a double-digit underdog is the one outlier that has improved the models.
If we'd won the Duke game 23-6 instead of 23-20, (i.e. not had three terrible punt unit plays in a row, and not gone 3-and-out playing prevent offense) we'd have moved a lot more.

 

[iceeater1969]

Clemson, UCF, and Ole Miss all won by larger margins than predicted, so they lowered our ratings. The Duke game was really tight, and wouldn't have moved our margin. The Pitt game where we won despite being a double-digit underdog is the one outlier that has improved the models.
If we'd won the Duke game 23-6 instead of 23-20, (i.e. not had three terrible punt unit plays in a row, and not gone 3-and-out playing prevent offense) we'd have moved a lot more.

Could the model tell we had 3 dl guys protecting the punter when they went for big returns. Coaches can fix that whoops wrong size player on field issue. Or does model just look at yards? If just yards we can make a big jump in st by being competent

 

[85Escape]

Clemson, UCF, and Ole Miss all won by larger margins than predicted, so they lowered our ratings. The Duke game was really tight, and wouldn't have moved our margin. The Pitt game where we won despite being a double-digit underdog is the one outlier that has improved the models.
If we'd won the Duke game 23-6 instead of 23-20, (i.e. not had three terrible punt unit plays in a row, and not gone 3-and-out playing prevent offense) we'd have moved a lot more.

Seems like the right way to include the score in these models would be to look at the time-weighted difference in the score throughout the game. Looking at the final score can be so deceptive since most coaches would rather be assured a win by 1 than take a chance on losing to win by 14. If the model looked at how 'tight' the game is across all 60 minutes, I think they would match better what we see with our eyes. Something like SUM0-60(deltaScore/60). That way if a team jumps out to a big lead and trades points for clock it doesn't look closer than it is. (I'm not a data scientist never played one on TV and didn't sleep in a Holiday Inn Express last night.)

 

[slugboy]

I seem to be finding a lot more interesting views on data lately.

In this, we're on the lower left. Tennessee is in the upper right. You really, really want to have a lot of red zone trips, but you also want touchdowns. We're not scoring much in the red zone, and when we do it's field goals way too often.

 

[leatherneckjacket]

I seem to be finding a lot more interesting views on data lately.

In this, we're on the lower left. Tennessee is in the upper right. You really, really want to have a lot of red zone trips, but you also want touchdowns. We're not scoring much in the red zone, and when we do it's field goals way too often.

I think the two axises are too closely related. Instead, they should have % of drives that make the red zone vs. points per red zone as the two axises.

 

[g0lftime]

I seem to be finding a lot more interesting views on data lately.

In this, we're on the lower left. Tennessee is in the upper right. You really, really want to have a lot of red zone trips, but you also want touchdowns. We're not scoring much in the red zone, and when we do it's field goals way too often.

Maybe a little more creative play calling down there. The failure to score on the goal line against Pitt was about as unimaginative as one can get. Been a problem since PJ retired.

 

[85Escape]

It would be interesting to see how things would be different if we'd been NCAAFB-average at red-zone scoring this year. Would we have won the UCF game? Would that mean that tonight's game would still include the Juice Squad? Inquiring minds want to know!

 

[ibeattetris]

I think the two axises are too closely related. Instead, they should have % of drives that make the red zone vs. points per red zone as the two axises.

That would be completely different.

I also don’t really see how we can come to the conclusion that it is too closely related when the results range from around 2 to around 7. Even the main bill of teams range from 4.2 to 5.8. 1.6 points per drive difference is pretty massive.

It would be interesting to see how things would be different if we'd been NCAAFB-average at red-zone scoring this year. Would we have won the UCF game? Would that mean that tonight's game would still include the Juice Squad? Inquiring minds want to know!

I’d like to see this chart from last year. Pretty sure we are even better this year than last.

Last year we averaged 2.35 points per drive is drives starting inside 40 and this year we are at 3.29. I don’t have ability right now to calculate the redzone drives sadly, but even this increase is substantial.

 

[leatherneckjacket]

That would be completely different.

I also don’t really see how we can come to the conclusion that it is too closely related when the results range from around 2 to around 7. Even the main bill of teams range from 4.2 to 5.8. 1.6 points per drive difference is pretty massive.


I’d like to see this chart from last year. Pretty sure we are even better this year than last.

Last year we averaged 2.35 points per drive is drives starting inside 40 and this year we are at 3.29. I don’t have ability right now to calculate the redzone drives sadly, but even this increase is substantial.

From a data distribution, they are closely related since the data points are almost all perfectly normal distributed around that line and that line has a slope of about 1. They are almost no outliers (Wyoming?).

Intuitively, the higher the percentage of times you score in the red zone, the more points per red zone trip you will have and vice versa. They both are a measure of red zone effectiveness. (Note: It has nothing to do with the fact that there is a range of data points along the X axis from 2 to 7.)

I think you would see more of a scatter plot if the it had % of drives that make red zone and points per red zone. Then you would have plenty of teams in all four quadrants, as one axis would measure drive effectiveness and the other axis would measure red zone effectiveness.