Today in Analytics...

85Escape

Helluva Engineer
Messages
1,450
SP+ is updated, and we're still down at #62. UNC is 18 and Pitt makes it in at 25.

It's gonna take a lot of winning to change our "rating", so hopefully, we make it look bad for a while. It's a "known design consideration":


In ESPN's "FPI" (which I almost never look at), we're up 23 points to #33. It's somewhat resume-based, so the last couple of games help:


OMG, that thing is useless. "Over half is based on preseason projections." Whungh? So what a bunch of people with journalism certificates think before a single snap is made makes up half your model five weeks into the year? That's patently stupid.
 

slugboy

Moderator
Staff member
Messages
10,749
Sagarin updating this morning. We moved up to #49. We were #79 after the narrow loss to Clemson. The model favors Pitt by about half a point in our stadium.

Virginia is down at #66, and is one of about half our remaining schedule that we’d be favored over in a neutral field. UNC is #36. Western Michigan is rated way ahead of us right now.


3C9CB473-E455-4EE1-B720-1B410F958012.png
 

ugacdawg

Georgia Tech Fan
Messages
29
OMG, that thing is useless. "Over half is based on preseason projections." Whungh? So what a bunch of people with journalism certificates think before a single snap is made makes up half your model five weeks into the year? That's patently stupid.

I think when Bill Connelly says 'preseason projections' he's talking about his own SP+ formula's projections which does use historical results, but not the AP/Coaches poll, etc. The reason there is a delayed effect is so the system doesn't yoyo for crazy upsets.

Think of it as if these two teams happened to play again this week, Vegas isn't going to put a GT -23 line on that game. Instead they'll adjust the line from something like UNC -13 to UNC -10. Because there's SO much other evidence UNC is the better team.

However, over time, GT could change that perspective that teaches the computer that last week's win was not the anomaly. At least that's how I understand it. SP+ is intended to project outcomes.

Probably a better indicator of actual performance THIS YEAR exclusively is going to be the FEI efficiency numbers which are opponent-adjusted and considers 'non-garbage possessions in FBS vs. FBS games' in actual 2021 games.

 
Last edited:

slugboy

Moderator
Staff member
Messages
10,749
Even if SP+ wasn’t using a pre-season weight based on last year, we’d have 1 loss to NIU, a win against KSU (where KSU was fighting pretty hard despite the score), an 8 point loss to a 2-loss Clemson, and one good win against a 2-loss UNC.
It’ll take multiple wins to move our ratings and predictions up, which makes sense because right now we’re trying to get bowl-eligible.

The model is just now starting to figure out who played who—there’s not enough information this season to compare a Washington State against a Georgia Tech.

2-2 should have us in the middle of an unbiased but not strongly connected model, right?
 

rfjeff9

Jolly Good Fellow
Messages
401
Even if SP+ wasn’t using a pre-season weight based on last year, we’d have 1 loss to NIU, a win against KSU (where KSU was fighting pretty hard despite the score), an 8 point loss to a 2-loss Clemson, and one good win against a 2-loss UNC.
It’ll take multiple wins to move our ratings and predictions up, which makes sense because right now we’re trying to get bowl-eligible.

The model is just now starting to figure out who played who—there’s not enough information this season to compare a Washington State against a Georgia Tech.

2-2 should have us in the middle of an unbiased but not strongly connected model, right?
If we go to a bowl year 3 then we are ahead of where I thought we would be At this point.
 

85Escape

Helluva Engineer
Messages
1,450
Even if SP+ wasn’t using a pre-season weight based on last year, we’d have 1 loss to NIU, a win against KSU (where KSU was fighting pretty hard despite the score), an 8 point loss to a 2-loss Clemson, and one good win against a 2-loss UNC.
It’ll take multiple wins to move our ratings and predictions up, which makes sense because right now we’re trying to get bowl-eligible.

The model is just now starting to figure out who played who—there’s not enough information this season to compare a Washington State against a Georgia Tech.

2-2 should have us in the middle of an unbiased but not strongly connected model, right?

Fair enough. But...there is a lot of bias in the model that considers one's initial starting position, but doesn't use the position of the team you played when you played them but after you played them. So it gets very difficult to 'move up' as it weights starting conditions except where those starting conditions contradict movement from them. That make sense?
 

slugboy

Moderator
Staff member
Messages
10,749
Fair enough. But...there is a lot of bias in the model that considers one's initial starting position, but doesn't use the position of the team you played when you played them but after you played them. So it gets very difficult to 'move up' as it weights starting conditions except where those starting conditions contradict movement from them. That make sense?
He could do without that preseason weighting, but then the predictions aren’t as good for the first few weeks. And his predictions are based on the previous season plus recruiting, so it’s reasonable
 

85Escape

Helluva Engineer
Messages
1,450
He could do without that preseason weighting, but then the predictions aren’t as good for the first few weeks. And his predictions are based on the previous season plus recruiting, so it’s reasonable

And that, I think, depends on if he wants the model to reflect current sentiment (predict what people feel is right) or be willing to be wildly variable early in the year as that is what the numbers are actually saying (but lose 'believability'...which is his main marketing 'reason to believe'.) I get why he does it that way, but it's weighting being believable over reflecting a slightly more objective reality. Or, in another way of saying it, is the model more reflective or predictive?

Anyway, like I said I'm not an analytics nerd so my opinion on this is about as valuable as a uGA grad's opinion on the sum of 2 and 3.
 

gville_jacket

Ramblin' Wreck
Messages
743
As the season moves along, the preseason rankings in SP+ will be weighted less and less. It's only 50% at this point in the season, not always. It's just better than having nothing at the beginning of the season. Some teams play a hard non-conference game early and others play a couple of cupcakes, is it really fair to judge just based on record when we all know the team with the challenging schedule is probably better? Also, there's always a crazy upset or two early so you don't want crazy ups and downs. Give it another 2-3 weeks and models like FPI and SP+ will level out and we'll have a better idea of where everyone is at.
 

Augusta_Jacket

Moderator
Staff member
Messages
7,875
Location
Augusta, Georgia
Or, in another way of saying it, is the model more reflective or predictive

The two aren't mutually exclusive. I would argue that it is a little of both. As with all models, it has some inherent flaws, but for prognostication purposes, it is far more accurate than most. Where it struggles is with teams like GT this year coming off a 3 win season but looking to be ahead of schedule in a rebuild or teams like Clemson that were playoff contenders last year but have an unforeseen regression. As long as you know the limitations of the model, it can be a very useful tool.
 

slugboy

Moderator
Staff member
Messages
10,749
Kenny Pickett leads the season total ESPN QBR as we approach the Pitt game:

RKNAMEQBRPAAPLAYSEPAPASSRUNSACKPENRAW
1Kenny Pickett (PITT)87.739.417238.331.67.3-4.43.891.5
2Dennis Grosel (BC)82.816.89317.68.310.2-1.50.686.8
3Sam Hartman (WAKE)80.627.215125233.1-1.60.386.7
4Devin Leary (NCST)77.913.918316.720.6-0.3-5.42.268.8
5Sam Howell (UNC)73.815.618518.416.110.2-9.31.370.6
6Brennan Armstrong (UVA)7224.423126.132.2-3.8-5.3375
7Malik Cunningham (LOU)64.417.520121.414.59.9-2.7-0.471.2
8Gunnar Holmberg (DUKE)63.815.718019.722.51-3.3-0.671.3
9Jordan Yates (GT)591.91336.98.93.8-60.353.7
10Braxton Burmeister (VT)48.44.5157109.94.5-5.81.557.4
11D'Eriq King (MIA)47.8-5.91772.110.6-3-6.40.941.4
12D.J. Uiagalelei (CLEM)46.2-13.7164-5.2-0.90.7-6.31.329.6
13McKenzie Milton (FSU)16.4-16.1123-8.60-4.6-3.6-0.520.5

Sims totally blew up on Saturday, and was the weekly leader last week:
RKNAMEWEEKRESULTQBRPAAPLAYSEPAPASSRUNSACKPENRAW
1Jeff Sims (GT)4W45-22vs. (UNC)98.69.8238.12.25.90098.8
2Kenny Pickett (PITT)4W77-7vs. UNH97.615.1337.66.41.6-0.3099.1
3Sam Hartman (WAKE)4W37-17at (UVA)87.47.5387.97.7-0.200.488.7
4Dennis Grosel (BC)4W41-34vs. MIZ86.67.6399.54.95.2-0.80.388.5
5Devin Leary (NCST)4W27-21vs. (CLEM)85.84616.910-1-3.41.866.6
6Gunnar Holmberg (DUKE)4W52-33vs. KU75.17.4498.462.70-0.382.9
7Malik Cunningham (LOU)4W31-23at (FSU)70.64.9556.78-0.9-0.3-0.171.5
8Brennan Armstrong (UVA)4L37-17vs. (WAKE)59.40.4813.15.40.2-2.60.251.4
9Sam Howell (UNC)4L45-22at (GT)53.2-1.7580.84.70.4-4.2-0.242.6
10D.J. Uiagalelei (CLEM)4L27-21at (NCST)46.6-339-1.1-1.81.5-1.10.331.2
11Garrett Shrader (SYR)4W24-21vs. LIB34.1-0.9340.70.30.9-0.80.242.9
12Braxton Burmeister (VT)4W21-10vs. RICH32.81.2402.82.51.3-1.30.357.8
13McKenzie Milton (FSU)4L31-23vs. (LOU)15.7-8.553-5.3-0.8-1.4-3.1015.8
 

slugboy

Moderator
Staff member
Messages
10,749
If you prefer passer rating:

RKNAMEPOSCMPATTCMP%YDSAVGLNGTDINTSACKRTG
1Kenny Pickett (PITT)QB9813274.21,34210.2671517195.6
4Sam Howell (UNC)QB7712163.61,1739.77511417168.5
3Brennan Armstrong (UVA)QB12018066.71,7059.56513310166.7
2Sam Hartman (WAKE)QB7411266.19618.664915162.9
7Gunnar Holmberg (DUKE)QB9312972.11,1408.853325150.9
5Devin Leary (NCST)QB10214868.91,0327511028147.1
6Jordan Yates (GT)QB528164.26137.6535110145.7
9Dennis Grosel (BC)QB416464.14777.549422141
8Braxton Burmeister (VT)QB6210161.47467.4475110137.8
10Malik Cunningham (LOU)QB85137629987.393525132.4
11D'Eriq King (MIA)QB8112266.47676.336349120.8
13McKenzie Milton (FSU)QB589362.45485.923247110.4
12D.J. Uiagalelei (CLEM)QB6311256.35865.244339103.7
 

slugboy

Moderator
Staff member
Messages
10,749
Lol 😂 welcome to “too much analytics”. It’s too early and the models aren’t settled yet.

FEI (Brian Fremeau) just updated. If it’s accurate, we will be looking at a tight game against Pitt

We’re 60, they’re 48, but we’re at home. It doesn’t like their offense much, but it really likes their defense ¯\_(ツ)_/¯

C0F2B205-D22D-4A2C-B580-D57D93C2F0FA.jpeg
 

slugboy

Moderator
Staff member
Messages
10,749
Here’s the Fremeau (FEI/F+). We’re up after the Duke game, and we’re on the positive side of FEI (we don’t have a negative number, and we’re better than average).

We have a 0.10 F+ (#66), #64 on offense, #59 on defense.


UVA is ahead of us and would be favored on a neutral field.

general, vs SP+ (surprised they’re tightly correlated? Kind of hard for them not to be)

 

slugboy

Moderator
Staff member
Messages
10,749
According to ESPN’s “Football Power Index”, we’re projected to have between 5-7 wins now, with the emphasis on “5”. We have slightly less than a 1 in 3 shot at hitting 6 or more wins and bowl eligibility.

 

bobongo

Helluva Engineer
Messages
7,043
According to ESPN’s “Football Power Index”, we’re projected to have between 5-7 wins now, with the emphasis on “5”. We have slightly less than a 1 in 3 shot at hitting 6 or more wins and bowl eligibility.

Beat Virginia and our chance of winning six goes to 1 in 2. Lose and it looks bleak. We gotta beat Virginia.
 

Skeptic

Helluva Engineer
Messages
6,372
Thanks--didn't catch the update date, and I should have. I guess we don't factor in torrential downpours, but they should affect both teams.
As Stonewall Jackson was reputed to have said when one of his generals whined that the raid was wetting the ammunition in one of his Valley battles, "It's raining on their side of the line, too."
 
Top