Massey composite has 4 tiers of conferences. The SEC held onto #1 and the Pac barely held onto #5. (Updated to January 16th-chart shows 69 programs in)

SEC 40.47---------40.64
Big 12 41.32-------41.28
Big 10 42.68-------42.64

ACC 57.89---------58.00

Pac 12 68.12--------68.22
MWC 68.76----------68.85
AAC 69.00-----------68.86
Ind. 74.68-----------74.69

Sun Belt 83.64------83.57
CUSA 90.38---------90.12
MAC 91.78-----------91.81

(01-14-2022 11:34 PM)bullet Wrote:  Massey composite has 4 tiers of conferences. The SEC held onto #1 and the Pac barely held onto #5.

SEC 40.47
Big 12 41.32 - - - - First Tier
Big 10 42.68

ACC 57.89 - - - - - Second Tier

Pac 12 68.12
MWC 68.76
AAC 69.00 - - - - - - Third Tier
Ind. 74.68

Sun Belt 83.64
CUSA 90.38 - - - - - Fourth Tier
MAC 91.78

Actually, the Massey Composite doesn't arrange conference into tiers, but it appears that you or someone else has sorted them into four four tiers based on their Massey Composite conference rankings.

That's ok, but others might choose to arrange them differently.

For example, there are some who would argue that there must be at least two teams per tier in a multi-tiered system.

If there have to be at least two teams per tier, by definition, the data would be forced into a more parsimonious (i.e., simpler) and most likely more reliable three-tier arrangement, such as this:

SEC 40.47
Big 12 41.32 - - - - First Tier (40-50)
Big 10 42.68

ACC 57.89 (closer to the mid-point of the second tier than to first tier)
Pac 12 68.12
MWC 68.76 - - - - - Second Tier (55-75)
AAC 69.00
Ind. 74.68 (closer to the mid-point of the second tier than to third tier)

Sun Belt 83.64
CUSA 90.38 - - - - - Third Tier (80-95)
MAC 91.78

.

Interestingly, using the RPI data (realtimerpi.com), the FBS teams could be arranged into three tiers, made up of the same conferences (although not in exactly the same order):

Big 10 30.54
SEC 30.46 - - - - - - - First Tier (27 - 31)
Big 12 28.90

Ind. 26.07
ACC 25.80
MWC 24.89 - - - - - - Second Tier (23 - 27)
AAC 24.65
Pac 12 23.43

Sun Belt 22.82
CUSA 20.76 - - - - - - Third Tier (19 - 23)
MAC 19.98

.

I haven't called this yet. The MC has had about 94 computers in its mix this year, and only 69 have reported so far.

These numbers, particularly between the SEC/B12 and MW/AAC are so close that new additions could swing things around the other way.

It's also noteworthy that the PAC is barely ahead of both the MW and AAC. That could change too.

(01-15-2022 02:39 AM)Milwaukee Wrote:  

(01-14-2022 11:34 PM)bullet Wrote:  Massey composite has 4 tiers of conferences. The SEC held onto #1 and the Pac barely held onto #5.

SEC 40.47
Big 12 41.32 - - - - First Tier
Big 10 42.68

ACC 57.89 - - - - - Second Tier

Pac 12 68.12
MWC 68.76
AAC 69.00 - - - - - - Third Tier
Ind. 74.68

Sun Belt 83.64
CUSA 90.38 - - - - - Fourth Tier
MAC 91.78

Actually, the Massey Composite doesn't arrange conference into tiers, but it appears that you or someone else has sorted them into four four tiers based on their Massey Composite conference rankings.

That's ok, but others might choose to arrange them differently.

For example, there are some who would argue that there must be at least two teams per tier in a multi-tiered system.

If there have to be at least two teams per tier, by definition, the data would be forced into a more parsimonious (i.e., simpler) and most likely more reliable three-tier arrangement, such as this:

SEC 40.47
Big 12 41.32 - - - - First Tier (40-50)
Big 10 42.68

ACC 57.89 (closer to the mid-point of the second tier than to first tier)
Pac 12 68.12
MWC 68.76 - - - - - Second Tier (55-75)
AAC 69.00
Ind. 74.68 (closer to the mid-point of the second tier than to third tier)

Sun Belt 83.64
CUSA 90.38 - - - - - Third Tier (80-95)
MAC 91.78

.

Interestingly, using the RPI data (realtimerpi.com), the FBS teams could be arranged into three tiers, made up of the same conferences (although not in exactly the same order):

Big 10 30.54
SEC 30.46 - - - - - - - First Tier (27 - 31)
Big 12 28.90

Ind. 26.07
ACC 25.80
MWC 24.89 - - - - - - Second Tier (23 - 27)
AAC 24.65
Pac 12 23.43

Sun Belt 22.82
CUSA 20.76 - - - - - - Third Tier (19 - 23)
MAC 19.98

.

The Sagarin ratings are very similar, as you would expect, with the exception of Independents (where the bottom three of UConn, UMass and NMSU are a real drag).

SEC 78.7
B12 77.8
B1G 77.4

ACC 73.1
PAC 72.8
AAC 67.8
MWC 67.5

IND 61.8*
SBC 61.7
MAC 61.4
USA 59.9

* Notre Dame, BYU, Army and Liberty have a combined rating of 77.2, while UConn, UMass and NMSU are rated at 41.4.

(01-15-2022 05:43 AM)quo vadis Wrote:  I haven't called this yet. The MC has had about 94 computers in its mix this year, and only 69 have reported so far.

These numbers, particularly between the SEC/B12 and MW/AAC are so close that new additions could swing things around the other way.

It's also noteworthy that the PAC is barely ahead of both the MW and AAC. That could change too.

Figured they would all be in by now. Guess not. Its interesting to see how odd some of the results are in individual polls. The very top is fairly consistent, with UGA 1 or 2, Alabama #2 to #5, Michigan #2 to #7, Ohio St. #2 to #9. But some of the others:
Oklahoma St. 1 to 12
Baylor 2 to 17
Oklahoma 5 to 24
Michigan St. 5 to 33
Texas 20 to 76 (#47 overall)

KInd of interesting that TSR, the one poll that doesn't have UGA #1 doesn't have all the Big 12 schools notably higher, really just OSU and Baylor and the bottom two. ISU is lower.
Oklahoma St. 1 vs composite 5
Baylor 4 vs. 7
Oklahoma 8 vs. 9
Iowa St. 29 vs. 26
KSU 22 vs. 31
Texas Tech 38 vs. 45
Texas 43 vs. 47
WVU 48 vs. 55
TCU 57 vs. 72
KU 92 vs. 108

Sagarian is also interesting.

Utah (11) lost to SDSU (53), BYU (33)
Oregon St (44) lost to Utah State (55)
Washington State (45) lost to Utah State (55)
UCLA (29) lost to Fresno State (51)
UCLA (29) lost Arizona State (36)
Boise State (28) lost to SDSU (53)

How do you throw all that into a computer and come up with those rankings? What algorithm would actually produce those results???

(01-16-2022 01:06 PM)Sactowndog Wrote:  Sagarian is also interesting.

Utah (11) lost to SDSU (53), BYU (33)
Oregon St (44) lost to Utah State (55)
Washington State (45) lost to Utah State (55)
UCLA (29) lost to Fresno State (51)
UCLA (29) lost Arizona State (36)
Boise State (28) lost to SDSU (53)

How do you throw all that into a computer and come up with those rankings? What algorithm would actually produce those results???

No algorithm is designed to predict winners 100% of the time. There will always be upsets. If the algorithm is good, the higher rated team should beat the lower one at least 70% of the time. The closer two teams are in rank, the closer to 50% they get. There's really not that much difference between #29 and #36, especially if the game is played at #36. And the earlier in the season it is the greater the likelihood of upsets. Anything before November is early.

(01-16-2022 09:03 PM)ken d Wrote:  

(01-16-2022 01:06 PM)Sactowndog Wrote:  Sagarian is also interesting.

Utah (11) lost to SDSU (53), BYU (33)
Oregon St (44) lost to Utah State (55)
Washington State (45) lost to Utah State (55)
UCLA (29) lost to Fresno State (51)
UCLA (29) lost Arizona State (36)
Boise State (28) lost to SDSU (53)

How do you throw all that into a computer and come up with those rankings? What algorithm would actually produce those results???

No algorithm is designed to predict winners 100% of the time. There will always be upsets. If the algorithm is good, the higher rated team should beat the lower one at least 70% of the time. The closer two teams are in rank, the closer to 50% they get. There's really not that much difference between #29 and #36, especially if the game is played at #36. And the earlier in the season it is the greater the likelihood of upsets. Anything before November is early.

Yes, to my knowledge, no predictive models say one team will win with 100% certainty. Even when it is a big mismatch, the model will say something like "Alabama has a 94% chance to beat Vanderbilt".

But sometimes, 6% chances come through and win. That's sports. So the validity of the model is determined by how good it is at predicting games over a large sample of games.

(01-15-2022 11:29 AM)bullet Wrote:  

(01-15-2022 05:43 AM)quo vadis Wrote:  I haven't called this yet. The MC has had about 94 computers in its mix this year, and only 69 have reported so far.

These numbers, particularly between the SEC/B12 and MW/AAC are so close that new additions could swing things around the other way.

It's also noteworthy that the PAC is barely ahead of both the MW and AAC. That could change too.

Figured they would all be in by now. Guess not. Its interesting to see how odd some of the results are in individual polls. The very top is fairly consistent, with UGA 1 or 2, Alabama #2 to #5, Michigan #2 to #7, Ohio St. #2 to #9. But some of the others:
Oklahoma St. 1 to 12
Baylor 2 to 17
Oklahoma 5 to 24
Michigan St. 5 to 33
Texas 20 to 76 (#47 overall)

Why are the AP, Coaches, Reddit and CFP selection committee included in the Massey Composite? Every other entry ranks all 130 teams. Don't these four distort the rankings of the other 105 teams?

(01-17-2022 11:55 AM)ken d Wrote:  

(01-15-2022 11:29 AM)bullet Wrote:  

(01-15-2022 05:43 AM)quo vadis Wrote:  I haven't called this yet. The MC has had about 94 computers in its mix this year, and only 69 have reported so far.

These numbers, particularly between the SEC/B12 and MW/AAC are so close that new additions could swing things around the other way.

It's also noteworthy that the PAC is barely ahead of both the MW and AAC. That could change too.

Figured they would all be in by now. Guess not. Its interesting to see how odd some of the results are in individual polls. The very top is fairly consistent, with UGA 1 or 2, Alabama #2 to #5, Michigan #2 to #7, Ohio St. #2 to #9. But some of the others:
Oklahoma St. 1 to 12
Baylor 2 to 17
Oklahoma 5 to 24
Michigan St. 5 to 33
Texas 20 to 76 (#47 overall)

Why are the AP, Coaches, Reddit and CFP selection committee included in the Massey Composite? Every other entry ranks all 130 teams. Don't these four distort the rankings of the other 105 teams?

I don't think you get much distortion from 4 out of (usually) 100 or so rankings?

(01-15-2022 11:29 AM)bullet Wrote:  

(01-15-2022 05:43 AM)quo vadis Wrote:  I haven't called this yet. The MC has had about 94 computers in its mix this year, and only 69 have reported so far.

These numbers, particularly between the SEC/B12 and MW/AAC are so close that new additions could swing things around the other way.

It's also noteworthy that the PAC is barely ahead of both the MW and AAC. That could change too.

Figured they would all be in by now. Guess not. Its interesting to see how odd some of the results are in individual polls. The very top is fairly consistent, with UGA 1 or 2, Alabama #2 to #5, Michigan #2 to #7, Ohio St. #2 to #9. But some of the others:
Oklahoma St. 1 to 12
Baylor 2 to 17
Oklahoma 5 to 24
Michigan St. 5 to 33
Texas 20 to 76 (#47 overall)

It's weird that almost a week after the final game was played, we have 69 ranking systems accounted for when all season long it's been about 94. Usually, these systems are updated quickly.

Maybe 69 is all we're going to get in the final MC?

(01-17-2022 10:04 AM)quo vadis Wrote:  

(01-16-2022 09:03 PM)ken d Wrote:  

(01-16-2022 01:06 PM)Sactowndog Wrote:  Sagarian is also interesting.

Utah (11) lost to SDSU (53), BYU (33)
Oregon St (44) lost to Utah State (55)
Washington State (45) lost to Utah State (55)
UCLA (29) lost to Fresno State (51)
UCLA (29) lost Arizona State (36)
Boise State (28) lost to SDSU (53)

How do you throw all that into a computer and come up with those rankings? What algorithm would actually produce those results???

No algorithm is designed to predict winners 100% of the time. There will always be upsets. If the algorithm is good, the higher rated team should beat the lower one at least 70% of the time. The closer two teams are in rank, the closer to 50% they get. There's really not that much difference between #29 and #36, especially if the game is played at #36. And the earlier in the season it is the greater the likelihood of upsets. Anything before November is early.

Yes, to my knowledge, no predictive models say one team will win with 100% certainty. Even when it is a big mismatch, the model will say something like "Alabama has a 94% chance to beat Vanderbilt".

But sometimes, 6% chances come through and win. That's sports. So the validity of the model is determined by how good it is at predicting games over a large sample of games.

Of course. My point was more the MWC and PAC had a large number of cross over games and games versus BYU. Then they played internally. Seems odd you can have that many large gaps. I’m sure it’s possible just curious why that algorithm varies so much from the others. I have heard he switched to an ELO model versus a cross results model but that could be incorrect.