Saturday, August 12, 2006

Versions of the BCS - 5+ and counting...

Versions / 1998 / 1999 / 2000 / 2001 / 2002 / 2003 / 2004 / 2005 / 2006 / 2007 / Conclusions

Since the inception of the BCS, the committee who runs it has made changes to it after almost every season. This tweaking has been done to make it a more accurate system (in theory), but more than anything the changes have been a reflex to controversy that ensued the previous year because the BCS didn't work. This section will look at these changes and at the different versions of the BCS throughout the past 9 years, comparing what each version's final rankings would have been for each of the past nine seasons. In the end, I hope to be able to provide an analysis of how effective each version would have been at avoiding controversy (impossible some years) and getting the top 2 teams into the National Championship game.

(A quick note of thanks: Finding all this data turned out to be much harder than I originally anticipated. I'd like to thank the following people who provided me with numbers and data, tips about where to look, and general support: CFR at the College Football Resource, Kenneth Massey, Herman Matthews, Peter Wolfe, David Wilson, George Vecsey, Robert Dunkel, Todd Beck, and others. Your time is greatly appreciated.)

First, a rundown of all the different versions and when they were in use. (The component descriptions were taken straight from the BCS Standings Explanation.)

Version A

Officially used in 1998
4 Components, all worth equal amounts.
Lowest total wins.

1st Component - Poll Average. Average of the USA Today/ESPN Coaches and AP Media Polls.

2nd Component - Computer Average. Average of the Anderson & Hester/Seattle Times, Jeff Sagarin's USA Today and New York Times rankings. In order to prevent unusual differences in individual formulas, a maximum adjusted deviation of no greater than 50 percent of the average of the two lowest computer rankings is utilized. In cases of adjusted deviation, the high score will become no lower than the middle score. (Note: These computer rankings were allowed to take Margin of Victory into account.)

3rd Component - Quartile Rank. Rank of schedule strength compared to other Division 1-A teams of actual games played divided by 25. This component is calculated by determining the cumulative won/loss records of the team's opponent (66.6 percent) and the cumulative won.loss record of the team's opponents' opponents (33.3 percent).

4th Component - Losses. One point for each loss during the season.

Notes: Pretty straightforward. One component is dedicated to people's views, one is computer rankings, one is SoS (determined by a computer formula), and one is outcome-based. Version A is the only version that includes the "Standard Deviation" rule that I will refer to. Basically, if a team is ranked #1, #2, and #5, they take the two highest rankings (1 & 2), average them (1.5), cut the average in half (.75) and add the average plus its half (1.5 + .75 = 2.25). This result then replaces the highest ranking, in this case #5. So instead of a computer average of 1+2+5=2.66, we get a computer average of 1+2+2.25=1.75, a difference of almost a full point.

Version B

Officially used in 1999 & 2000
4 Components, all worth equal amounts.
Lowest total wins.

1st Component - Poll Average. Same as Version A.

2nd Component - Computer Average. Average of Richard Billingsley, Dunkel Index, Kenneth Massey, New York Times, David Rothman, Jeff Sagarin's USA Today, Matthews/Scripps-Howard, and the Anderson & Hester/Seattle Times rankings. The computer component is determined by averaging the seven highest computer rankings. The lowest (worst) computer ranking is disregarded. (Again, MoV is allowed.)

3rd Component - Schedule Rank. Same as Version A.

4th Component - Losses. Same as Version A.

Notes: One minor change from Version A - adding more computer rankings to the mix, probably in order to form more of a true computer average. As with Version A, one component is dedicated to people's views, one is computer rankings, one is Sos (determined by a computer formula), and one is outcome-based.

Version C

Officially used in 2001
5 Components, all worth equal amounts.
Lowest total wins.

1st Component - Poll Average. Same as Versions A & B.

2nd Component - Computer Average. Average of Anderson & Hester, Atlanta Journal-Constitution Colley Matrix, Richard Billingsley, Kenneth Massey, David Rothman, Jeff Sagarin's USA Today, Matthews/Scripps-Howard, and the Peter Wolfe rankings. The computer component will be determined by averaging six rankings. The highest and lowest will be disregarded.

3rd Component - Schedule Rank. Same as Versions A & B.

4th Component - Losses. Same as Versions A & B.

5th Component - Quality Wins. The quality win component will reward to varying degrees teams that defeat opponents ranked among the top 15 in the weekly standings. The bonus point scale will range from a high of 1.5 points for a win over the top ranked team to a lost of .1 for a victory over the the 15th-ranked BCS team. The BCS Standings at the end of the regular season will determine final quality win points. If a team registers a victory over a team more than once during the regular season, quality points will be awarded just once. Quality win points are based in the standings determined by the subtotal. The final standings are reconfigured to reflect the quality win point deduction.

Notes: Two major changes from Version B. First, some computer rankings were added and some were dropped. This is minor, but the stipulation that the computer rankings not take Margin of Victory into account was major. It is the reason the Dunkel Index decided not to participate in the BCS after the 2000 season. Second, the addition of the Quality Win component was necessitated by the fact that in 2000 Florida State was invited to the National Championship game over Miami (FL) who had beaten them earlier in the season. This was the first of the major BCS controversies, and the first post-emptive reaction by the Committee to right what they saw as a wrong.

Version D

Officially used in 2002 & 2003
5 Components, all worth equal amounts.
Lowest total wins.

1st Component - Poll Average. Same as Versions A, B, & C.

2nd Component - Computer Average. Average of Anderson & Hester, Richard Billingsley, Colley Matrix, Kenneth Massey, New York Times, Jeff Sagarin's USA Today, and the Peter Wolfe rankings. The computer component will be determined by averaging six of the seven rankings. The lowest (worst) computer ranking will be disregarded.

3rd Component - Schedule Rank. Same as Versions A, B, & C.

4th Component - Losses. Same as Versions A, B, & C.

5th Component - Quality Wins. The quality win component will reward to varying degrees teams that defeat opponents ranked among the top 10 in the weekly standings. The bonus point scale will range from a high of 1.0 points for a win over the top ranked team to a lost of .1 for a victory over the the 10th-ranked BCS team. The BCS Standings at the end of the regular season will determine final quality win points. If a team registers a victory over a team more than once during the regular season, quality points will be awarded just once. Quality win points are based in the standings determined by the subtotal. The final standings are reconfigured to reflect the quality win point deduction.

Notes: Two minor changes from Version C, which is unusual considering what happened in 2001... The first change is that the computer rankings were modified yet again. Instead of eight rankings, the Rothman and the Matthews/Scripps-Howard rankings are out, and the NY Times are back in after a year away. MoV is still out though, so this is only a minor change. The second change is to award quality win points only for victories against the BCS Top 10 instead of Top 15. Again, a seemingly minor change. It's interesting that these were the only modifications after 2001 though, which was the year of the "Nebraska didn't even make it's conference championship game but made it to the NC" debacle.

Version E

Officially used in 2004
3 Components, all worth equal amounts.
Highest average percentage wins.

1st & 2nd Components - Poll Averages. Team percentages are derived by dividing a team's actual voting points by a maximum of 1625 possible points in the AP Poll and 1525 possible points in the USA Today/ESPN Coaches Poll.

3rd Component - Computer Average. Six computer rankings calculated in inverse points order (25 for #1, 24 for #2, etc.) are used to determine the overall computer component. The best and worst raking for each team is dropped, and the remaining four are added and divided by 100 (the maximum possible points) to produce a Computer Rankings Percentage. The six computer rankings providers are Anderson & Hester, Richard Billingsley, Colley Matrix, Kenneth Massey, Jeff Sagarin, and Peter Wolfe. Each computer ranking accounts for schedule strength in its formula.

Notes: Between Versions D and E is when the biggest changes occurred. First, lowest point total wins was changed to highest percentage wins. This first change personified the hard turn the rankings were taking. Even though the way the components were averaged and added didn't really change much, the aesthetics of the system were flipped upside down, giving the appearance of a completely new ranking system. This could be a good thing (if people said that some necessary improvements were finally being made) or a bad thing (if people said that it was just another reactionary gesture aimed at fixing problems of past years).

Because USC had been ranked #1 by both polls in 2003 and still didn't make it to the championship game, the 2003 season ended with a split title, the very thing the BCS was designed to avoid. So what did the BCS do? Make the human polls worth 2/3 of the rankings, a huge change since before this the human polls combined had only been worth 1/5 of the rankings. In all fairness, the change to using actual voting points and percentages from the human polls instead of straight poll rankings was a great change that should've been made much sooner.

Changing the way the computer rankings were displayed to fit the new percentage system was necessary, but it made them even more visually complex, adding to the overall (and overreaching) beatdown the computers received this year. Possibly to alleviate some of this newly visible complexity, the rankings dropped the Quartile Rank (SoS) component, the Loss component, as well as the Quality Win component, figuring that these things were already taken into account in the polls and computers. So all in all, a major shift for the BCS rankings. Did they work? Well...

Version E-2

Officially used in 2005, 2006, & 2007
3 Components, all worth equal amounts.
Highest average percentage wins.

1st & 2nd Components - Poll Averages. Team percentages are derived by dividing a team's actual voting points by a maximum of 2825 possible points in the Harris Interactive and 1550 possible points in the USA Today/ESPN Coaches Poll.

3rd Component - Computer Average. Six computer rankings calculated in inverse points order (25 for #1, 24 for #2, etc.) are used to determine the overall computer component. The best and worst raking for each team is dropped, and the remaining four are added and divided by 100 (the maximum possible points) to produce a Computer Rankings Percentage. The six computer rankings providers are Anderson & Hester, Richard Billingsley, Colley Matrix, Kenneth Massey, Jeff Sagarin, and Peter Wolfe. Each computer ranking accounts for schedule strength in its formula.

Notes: The minor change before 2005 was that the Harris Interactive poll replaced the AP poll. After splitting the title in 2003 between USC and LSU, the AP finally had enough after the USC-Okalhoma-Auburn debacle of 2004 in which there were three undefeated teams at the end of the regular season, one of which had to be left out of the National Championship game. The AP Decided to remove itself from the BCS rankings so that there wasn't a conflict of interest, since they had been reporting on news that they were creating. So the Harris Poll was invented and used, replacing the AP Poll in the official rankings. That's why this version is labeled E-2 - it's basically the same as Version E, especially considering that in 2005 & 2006 the AP and Harris polls produced exactly the same Top 10 at the end of the regular season (with the exception of USC & Oklahoma switching at #7 & #8 in 2006).

_________________________________________________

So there they are, the (roughly) 5 versions the BCS Rankings have used. Next we'll look at the results each version would have produced in each of the 9 BCS years.

A slight disclaimer first. I still need some key pieces of data to officially complete these results:
1) Peter Wolfe's final regular season rankings from 1998, 1999, and 2000
2) The Dunkel Index's final regular season rankings from 2004
3) The New York Times' final regular season rankings from 2004, 2005, and 2006.

For the first two, if anyone has or knows where I can find this information, please let me know. The last one is going to be tricky, mainly because the NY Times stopped conducting their own computer rankings after the 2003 season. But I'm working on that one.

Because some of the computer polls are incomplete, I've had to make some projections based on mathematical possibilities, a task which actually went easier than I expected in most cases. I'm confident of my projections, even if they're not mathematically certain, and my reasoning for each one is included.

But enough of this - I'll explain more as we go along...

Top > 1998

2 comments:

Anonymous said...

Yay

pay per head service said...

I don't think that they think enough how to fix BCS. They should hire people that they are quite well known in that kind of thing to help them to fix it.