August 17, 2008

Who's who amongst BCS conferences.

It is apparent that there is an elephant in the blog and we need to address the issue before the season starts. Maybe I just like swinging baseball bats at hornets nets.

For this thread the h policy will be temporarily suspended. This immunity shall be revoked for anyone questioning the h policy in any manner what-so-ever.

Once the season starts, for all comments related to this thread and subject to the h policy, the h policy shall be modified to include a link to this page. Indeed, all comments related to this thread with no disclaimer in the tread being commented on will generally be subject to the h policy.

Which conferences should/should not be BCS conferences and why?
Which teams deserve a promotion to a BCS conference and why?

Shortly a thread will be created to determine which benchmarks best measure the performance of a conference. A thread will be created after the BCS polls are first released and after the season to examine the results of this season relative to the established benchmarks. These threads will be the only ones to reopen this discussion. If you have anything to say on this topic now is the time to say it.

All statements should be backed up with metrics and numbers based on the metrics used. Be ready to have your methodology shot down. Everyone remain calm. Very different views from our own will be expressed. At the end of the day we will all agree to disagree.

To start the conversation I will express two measures of conference strength.

BCS Computers:
Using the 4 BCS computers that provide a conference comparison it is possible to generate a composite ranking of the conferences. A poll like ranking loses much of the information available. Here the raw data is converted to a z-sore for each computer ranking using only the conferences listed. For each conference the values below represent the average z-score of the four programs.

SEC: 1.08
PAC 10: 0.98
BIG 12: 0.87
Big East: 0.70
ACC: 0.58
BIG 10: 0.54
MWC: -0.11
C-USA: -0.93
WAC: -1.04
MAC: -1.29
Sun Belt: -1.38

I have been running similar results in the past. Historically the Big 10 is usually closer to 0.90. The ACC is the most variable, ranging from 0.3 to 0.90. The Big East ranges from 0.3 to 0.7. Non BCS conferences occasionally break 0.3. C-USA, WAC and MAC usually end up higher than this. The gap between BCS and non BCS was unusually large last year.

This result strongly supports the statement that the BCS conferences are properly identified.

This afternoon I revisited a methodology I first thought of at the end of last year using only the BCS standings. If two teams differ by 1 spot across the board this equates to a difference in the BCS standings of 0.04. Teams closer than this are close in the BCS standings. Replacing each team with the conference they represent and using a baseball like games behind to measure the difference in schedule allows one to determine.

Hawaii and Florida had a difference of 0.0335 making them close in the standings. 12-0 is three games ahead of 9-3. The SEC is three games ahead of the WAC based on these teams.

For each conference, taking the average games behind of all its close rankings gives a measure of the strength of each conference compared to those it is close to. Like baseball all results are given in terms of games behind the leader.

I fudged this by hand last spring, using the final polls and computer rankings to estimate what a final BCS standing would look like after the Bowls. Here I broke out Excel to actually solve the equations using iteration on the pre bowl standings. The results surprised me.

PAC 10: --
SEC: 0.08
Big 10: 0.62
Big 12: 0.83
Big East: 0.86
ACC: 0.93
C-USA: 0.98
MWC: 1.67
WAC: 2.56
MAC: N/A
Sun Belt: N/A

These are determined pre-bowl and don't reflect Ohio States loss to LSU or the ASU loss to Texas, among other significant Bowl Games. The post bowl approximation moved the PAC 10 to .5 behind the SEC, the Big 10 with the ACC, and C-USA with the MAC and Sun Belt.

Also the Big East, C-USA and the MWC are determined by only two close spots, making them less precise. For C-USA I would argue they are a bit inflated. This data does support the idea that C-USA is closer to making the jump than the MWC.

C-USA playes higher caliber BCS teams OOC than the MWC or WAC and this shows up here, as well as the loss column.

Indeed an expanded MWC, with Boise State and Fresno State, may narrow this gap. When these teams play each other they will accumulate more losses. The likelihood of these going undefeated, or even having a 10 win season, would be traded for a higher result if it happens. The final expected ranking would remain the same unless play on the field is improved. Expansion increases the variability of these teams allowing them to attain a better maximum.

As a final note: I am not fair. I am biased towards the non BCS conferences and teams.

Let the analytical swarms begin!

Comments:

Who's who amongst BCS conferences.
It is apparent that there is an elephant in the blog and we need to address the issue before the season starts. Maybe I just like swinging baseball bats at hornets nets. For this thread the h policy will be temporarily suspended. This immunity shall be revoked for anyone questioning the h policy in any manner what-so-ever. Once the season starts, for all comments related to this thread and subject to the h policy, the h policy shall be modified to include a link to this page. Indeed, all comments related to this thread with no disclaimer in the tread being commented on will generally be subject to the h policy. Which conferences should/should not be BCS conferences and why? Which teams deserve a promotion to a BCS conference and why? Shortly a thread will be created to determine which benchmarks best measure the performance of a conference. A thread will be created after the BCS polls are first released and after the season to examine the results of this season relative to the established benchmarks. These threads will be the only ones to reopen this discussion. If you have anything to say on this topic now is the time to say it. All statements should be backed up with metrics and numbers based on the metrics used. Be ready to have your methodology shot down. Everyone remain calm. Very different views from our own will be expressed. At the end of the day we will all agree to disagree. To start the conversation I will express two measures of conference strength. BCS Computers: Using the 4 BCS computers that provide a conference comparison it is possible to generate a composite ranking of the conferences. A poll like ranking loses much of the information available. Here the raw data is converted to a z-sore for each computer ranking using only the conferences listed. For each conference the values below represent the average z-score of the four programs. SEC: 1.08 PAC 10: 0.98 BIG 12: 0.87 Big East: 0.70 ACC: 0.58 BIG 10: 0.54 MWC: -0.11 C-USA: -0.93 WAC: -1.04 MAC: -1.29 Sun Belt: -1.38 I have been running similar results in the past. Historically the Big 10 is usually closer to 0.90. The ACC is the most variable, ranging from 0.3 to 0.90. The Big East ranges from 0.3 to 0.7. Non BCS conferences occasionally break 0.3. C-USA, WAC and MAC usually end up higher than this. The gap between BCS and non BCS was unusually large last year. This result strongly supports the statement that the BCS conferences are properly identified. This afternoon I revisited a methodology I first thought of at the end of last year using only the BCS standings. If two teams differ by 1 spot across the board this equates to a difference in the BCS standings of 0.04. Teams closer than this are close in the BCS standings. Replacing each team with the conference they represent and using a baseball like games behind to measure the difference in schedule allows one to determine. Hawaii and Florida had a difference of 0.0335 making them close in the standings. 12-0 is three games ahead of 9-3. The SEC is three games ahead of the WAC based on these teams. For each conference, taking the average games behind of all its close rankings gives a measure of the strength of each conference compared to those it is close to. Like baseball all results are given in terms of games behind the leader. I fudged this by hand last spring, using the final polls and computer rankings to estimate what a final BCS standing would look like after the Bowls. Here I broke out Excel to actually solve the equations using iteration on the pre bowl standings. The results surprised me. PAC 10: -- SEC: 0.08 Big 10: 0.62 Big 12: 0.83 Big East: 0.86 ACC: 0.93 C-USA: 0.98 MWC: 1.67 WAC: 2.56 MAC: N/A Sun Belt: N/A These are determined pre-bowl and don't reflect Ohio States loss to LSU or the ASU loss to Texas, among other significant Bowl Games. The post bowl approximation moved the PAC 10 to .5 behind the SEC, the Big 10 with the ACC, and C-USA with the MAC and Sun Belt. Also the Big East, C-USA and the MWC are determined by only two close spots, making them less precise. For C-USA I would argue they are a bit inflated. This data does support the idea that C-USA is closer to making the jump than the MWC. C-USA playes higher caliber BCS teams OOC than the MWC or WAC and this shows up here, as well as the loss column. Indeed an expanded MWC, with Boise State and Fresno State, may narrow this gap. When these teams play each other they will accumulate more losses. The likelihood of these going undefeated, or even having a 10 win season, would be traded for a higher result if it happens. The final expected ranking would remain the same unless play on the field is improved. Expansion increases the variability of these teams allowing them to attain a better maximum. As a final note: I am not fair. I am biased towards the non BCS conferences and teams. Let the analytical swarms begin!