Mid Season Reviews and Rankings

[SuperHuski]

Bye Week with half the season finished. So let's see where the team stands compared to the rest of CFB. Reminder, there are 134 FBS schools.

Computers:

	Massey	Sagarin	ESPN
UConn	93 (+2)	92 (-4)	72 (-5)

Offense Overall:

Stat	Value	Rank
Points/Game	30.4	47
Yards/Game	414.8	46
Points/Play	0.403	54
3D Conversion %	35.14%	95
4D Conversion %	40.00%	105
Redzone Scoring %	88.24%	56

Even with some of our recent shortcomings against Temple, we're still an above average offense.

Defense Overall:

Stat	Value	Rank
Points/Game	22.6	45
Yards/Game	351.6	42
Points/Play	0.328	40
3D Conversion %	24.66%	2
4D Conversion %	37.50%	23
Redzone Scoring %	70.59%	12

We also have an above average defense. Our third down and red zone defense is elite. Not just against bad teams as some of you might say. Duke was 6/18 against us on 3rd downs. Maryland was 6/16 on 3rd downs.

Offensive Rushing:

Stat	Value	Rank
Rush Play %	58.36%	28
Yards/Rush	4.7	46
Rushes/Game	44	12
Rush Yards/Game	208.8	17

You don't need stats to know that our rushing attack has been great.

Offensive Passing:

Stat	Value	Rank
Pass Play %	41.64%	107
Completion %	53.59%	120
Yards/Pass	6.7	92
Passes/Game	30.6	69
Passing Yards/Game	206	84
Int %	3.27%	91
QB Sacked %	2.55%	18

However, our passing offense has been subpar by a number of different metrics. Shoutout to the elite o-line though! My take: we do have better skill players, but a combination of QB play and passing schemes has been lacking.

Misc:

Stat	Value	Rank
TO Margin/Game	-0.6	103
Fumbles/Game	1.8	109
INTs/Game	1.2	82
INT Throw %	3.27%	91
Penalties/Play	0.04	42
Penalties/Game	5.8	51
Opp Penalties/Play	0.05	47
Opp Penalties/Game	7.2	39

We don't cause many fumbles or INTs per game but we seem to be an above average disciplined team.

Overall, we're an above average team. Somewhere in the 40-60 range of all FBS schools. That would put us in the middle of the pack with the B12 and ACC. And we're doing this without anywhere near the funding that they have. The computer rankings are derived from several years of data, so it will take time for those to reflect the current state of things.

Player Stats:
Joe Fagnano is #26 in the country with 11 TDs.
Durell Robinson is #7 with 7.7 yards/rush. #42 in total yard at 421.
Skyler Bell is 18th in total receiving yards at 508.

 

[BlueandOG]

Love it! Thank you for putting this together!

 

[businesslawyer]

Bye Week with half the season finished. So let's see where the team stands compared to the rest of CFB. Reminder, there are 134 FBS schools.

Computers:

	Massey	Sagarin	ESPN
UConn	93 (+2)	92 (-4)	72 (-5)

Offense Overall:

Stat	Value	Rank
Points/Game	30.4	47
Yards/Game	414.8	46
Points/Play	0.403	54
3D Conversion %	35.14%	95
4D Conversion %	40.00%	105
Redzone Scoring %	88.24%	56

Even with some of our recent shortcomings against Temple, we're still an above average offense.

Defense Overall:

Stat	Value	Rank
Points/Game	22.6	45
Yards/Game	351.6	42
Points/Play	0.328	40
3D Conversion %	24.66%	2
4D Conversion %	37.50%	23
Redzone Scoring %	70.59%	12

We also have an above average defense. Our third down and red zone defense is elite. Not just against bad teams as some of you might say. Duke was 6/18 against us on 3rd downs. Maryland was 6/16 on 3rd downs.

Offensive Rushing:

Stat	Value	Rank
Rush Play %	58.36%	28
Yards/Rush	4.7	46
Rushes/Game	44	12
Rush Yards/Game	208.8	17

You don't need stats to know that our rushing attack has been great.

Offensive Passing:

Stat	Value	Rank
Pass Play %	41.64%	107
Completion %	53.59%	120
Yards/Pass	6.7	92
Passes/Game	30.6	69
Passing Yards/Game	206	84
Int %	3.27%	91
QB Sacked %	2.55%	18

However, our passing offense has been subpar by a number of different metrics. Shoutout to the elite o-line though! My take: we do have better skill players, but a combination of QB play and passing schemes has been lacking.

Misc:

Stat	Value	Rank
TO Margin/Game	-0.6	103
Fumbles/Game	1.8	109
INTs/Game	1.2	82
INT Throw %	3.27%	91
Penalties/Play	0.04	42
Penalties/Game	5.8	51
Opp Penalties/Play	0.05	47
Opp Penalties/Game	7.2	39

We don't cause many fumbles or INTs per game but we seem to be an above average disciplined team.

Overall, we're an above average team. Somewhere in the 40-60 range of all FBS schools. That would put us in the middle of the pack with the B12 and ACC. And we're doing this without anywhere near the funding that they have. The computer rankings are derived from several years of data, so it will take time for those to reflect the current state of things.

Player Stats:
Joe Fagnano is #26 in the country with 11 TDs.
Durell Robinson is #7 with 7.7 yards/rush. #42 in total yard at 421.
Skyler Bell is 18th in total receiving yards at 508.

Huh? You showed three computer rankings that put us at an average of about 85. And then jumped to a conclusion that we are somewhere between 40 and 60. I can tell you where that delta comes from (I think). You're using stats that aren't adjusted for strength of opponents, but our rankings absolutely are based on results against our particular strength of opponents.

Having said that, I do appreciate the work laying everything out like that.

 

[imonaboat]

Huh? You showed three computer rankings that put us at an average of about 85. And then jumped to a conclusion that we are somewhere between 40 and 60. I can tell you where that delta comes from (I think). You're using stats that aren't adjusted for strength of opponents, but our rankings absolutely are based on results against our particular strength of opponents.

Having said that, I do appreciate the work laying everything out like that.

"The computer rankings are derived from several years of data, so it will take time for those to reflect the current state of things."

 

[businesslawyer]

"The computer rankings are derived from several years of data, so it will take time for those to reflect the current state of things."

Would you mind sharing what you are quoting from. I thought, at least as to Sagarin and Massey, that they start the season using prior year’s stats but that by now they are ranked just on this year?

 

[AllHusky]

Thanks

 

[Patman]

Would you mind sharing what you are quoting from. I thought, at least as to Sagarin and Massey, that they start the season using prior year’s stats but that by now they are ranked just on this year?

might depend... but my understanding is roughly yours. I'm running a model right now that's very wonky based on a similar idea to massey but worse.... but mostly because I wanted an idea on why Alabama was so high despite losing to vandy. I still don't have an answer other than the idea they may be leaning heavy on last year's data but it may be that I'm using a crappy input... which I am.

I think your criticism is fair for the record. He didn't provide any links.
---
Personally hoping if Saragin is retiring he makes his method public.

 

[Patman]

Found that somebody figured out Ken Massey's method... well, I wish I knew about this 18 years ago.

Massey's "Game Outcome Function"

Kenneth Massey, whose computer poll is one of the six used by the BCS, has a unique "game outcome function" that estimates the odds of "Team A" winning a rematch against "Team B" under identical conditions, given only the score of the game. (the poll he gives the BCS doesn't use this powerful tool,…

omfgpwn3d.livejournal.com

Based on that I fit the model on Division 1 data only for this year only with a collective home field calculation. UConn sits 57. Between Cal and Baylor. Ahead of Duke.

I suspect that if Massey's function is still current that its most likely using previous year's data as a non-trivial input. Keep in mind he ranks all of NCAA and NAIA on the same sheet so he will NEED to use some amount of previous data as a lot of D2/D3/NAIA stay with insular schedules until the end of the year. I have no real idea of what's under the hood.

Now, I don't believe Massey's method is justifiable under statistical methods but these become a means to an end... he may have messed with it an incredible number of ways since 2005.

edit: My last statement is true for virtually all computer ranking systems. I have an idea of the model I'd want to use but I haven't figured out how to write it but I have some "on the way" models I'd like to try. Bayesian nonparametrics and copulas.

edit #2: One reason not to use the new season's data immediately. Navy 15, Army 13 as described the model fit that put uconn at 57... current Massey has them 53 and 57 respectively. Again, he may have also changed his outcome function but I highly doubt he's fundamentally changed his method.

 

[KryHavok]

Found that somebody figured out Ken Massey's method... well, I wish I knew about this 18 years ago.

edit: My last statement is true for virtually all computer ranking systems. I have an idea of the model I'd want to use but I haven't figured out how to write it but I have some "on the way" models I'd like to try. Bayesian nonparametrics and copulas.

Thanks for posting that link. Found myself going down a rabbit hole (trying to setup a matrix to solve) instead of doing what I was supposed to be doing.

 

[KryHavok]

Okay, I went back and looked at that link that @Patman posted above. The Game Outcome Function shown in the table can be easily calculated in Excel. If Team 1 has 30 points and Team 2 has 29, the arithmetic average is 29.5, the standard deviation is (50*(30+29))^0.25 = 7.369797 (this is NOT the arithmetic standard deviation, which would be 0.707107). Never mind using the GOF equation shown in the link, you can easily determine the GOF in Excel using the norm.dist function in which you input the reference value, mean, standard deviation, and distribution type. The GOF for Team 1 would be given by: norm.dist(30, 29.5, 7.369797, TRUE) = 0.527045 ~ 52.7% of Team 1 winning again, where the TRUE indicates a cumulative distribution with a max value of 1.

 

[Patman]

R code

pnorm(A,(A+B)/2,(50)^.25*(A+B)^.25)

 

[KryHavok]

R code

pnorm(A,(A+B)/2,(50)^.25*(A+B)^.25)

I love R and use it for my modeling and simulation work. I found a reference that used matrices to determine Massey's apparent ranks, and this should be straightforward to script.

 

[businesslawyer]

I learned something from this thread. I learned that, unless folks are just using gibberish words thinking no one will know the difference, I’m apparently not as smart about statistics as I thought I was.

 

[FfldCntyFan]

I learned something from this thread. I learned that, unless folks are just using gibberish words thinking no one will know the difference, I’m apparently not as smart about statistics as I thought I was.

 

[huskymedic]

 

[Patman]

I love R and use it for my modeling and simulation work. I found a reference that used matrices to determine Massey's apparent ranks, and this should be straightforward to script.

I just assume he's using logistic regression. If I'm wrong I'm wrong and not in the mood to redo anything.

 

[KryHavok]

I just assume he's using logistic regression. If I'm wrong I'm wrong and not in the mood to redo anything.

I skimmed through his undergrad thesis and came across an article that tried to simplify his approach.
https://maherou.github.io/Teaching/files/CS317/masseyMethod.pdf
When I go through the article using the examples of the 5 Div III schools and try to solve the matrix as shown on page 5, I get a solution in R with parameter estimates of -8, 5.6, 3, 20.6, and -21.2 for Dickinson, Franklin & Marshall, Gettysburg, Johns Hopkins, and McDaniel, respectively. These values can likely be interpreted as the relative ratings of each team with Johns Hopkins having the highest and McDaniel having the lowest. This probably overly simplistic and obsolete. What he does after this is still beyond what I'll attempt to solve, but at least it's now mostly a black box rather than totally a black box.

 

[Patman]

I skimmed through his undergrad thesis and came across an article that tried to simplify his approach.
https://maherou.github.io/Teaching/files/CS317/masseyMethod.pdf
When I go through the article using the examples of the 5 Div III schools and try to solve the matrix as shown on page 5, I get a solution in R with parameter estimates of -8, 5.6, 3, 20.6, and -21.2 for Dickinson, Franklin & Marshall, Gettysburg, Johns Hopkins, and McDaniel, respectively. These values can likely be interpreted as the relative ratings of each team with Johns Hopkins having the highest and McDaniel having the lowest. This probably overly simplistic and obsolete. What he does after this is still beyond what I'll attempt to solve, but at least it's now mostly a black box rather than totally a black box.

this paper just reinvents linear regression matrix theory

 

[Palatine]

It works something like this.

 

[TheDrifter75]

I skimmed through his undergrad thesis and came across an article that tried to simplify his approach.
https://maherou.github.io/Teaching/files/CS317/masseyMethod.pdf
When I go through the article using the examples of the 5 Div III schools and try to solve the matrix as shown on page 5, I get a solution in R with parameter estimates of -8, 5.6, 3, 20.6, and -21.2 for Dickinson, Franklin & Marshall, Gettysburg, Johns Hopkins, and McDaniel, respectively. These values can likely be interpreted as the relative ratings of each team with Johns Hopkins having the highest and McDaniel having the lowest. This probably overly simplistic and obsolete. What he does after this is still beyond what I'll attempt to solve, but at least it's now mostly a black box rather than totally a black box.

AI throughs up some crazy stuff sometimes

 

[Patman]

AI throughs up some crazy stuff sometimes

GLARE

----

Low amount of data, loosely connected schedules. You're going to get extreme fits.

 

[Patman]

Speaking of extreme fits... Northwestern destroying Maryland cost UConn 15 slots in the calculation I have.... all the more reason for the benefits of using prior year's data or formulating a model in that way...