BCS computer question

[Tideski]

Can someone explain why the BCS computers have us ranked #12? With our schedule, does one loss make that big a deal according to the computers? Does any know or understand the equation?

 

[GreatDanish]

Each computer uses different equations.
The loss is the main part right now. As the SOS data is still weak, there is a lot of variability. Each week, the SOS component will begin to take shape.
Right now, there is less of a difference in team records than there will be at the end of the season. That is, at the end of the season, W-L records will make somewhat of a bell curve. Now, it's probably just as close to being a uniform distribution.
What that means is that it's a lot more difficult to differentiate between a 4-2 South Carolina team and a 4-2 Indiana team right now. Translated to human terms, the computers are saying, "We still don't really know exactly how good South Carolina is, or if they have just gotten lucky a few times."
Each week, if we win, our computer rankings should increase across the board.

 

[BamasTide]

Quite simply there are eleven unbeaten teams left.

 

[GreatDanish]

Quite simply there are eleven unbeaten teams left.

Being undefeated won't always equate to a better computer ranking. One-loss Florida was higher than both undefeated TCU and Boise State after the SEC championship last year. As more games are played, the more likely a one-loss team is to move up in the computers.

 

[DrollTide]

I think part of it is that some of our wins looked great at the time (Penn St., Gators, Hogs) but these are being depreciated by events. When we played Florida, it was thought of as #1 against #2, but Florida aren't even ranked now.

 

[Bamathrunthru]

Mama said life was like a box of chocolate you never know what you going to get:conf2::conf3::blush:

 

[popechild]

Can someone explain why the BCS computers have us ranked #12? With our schedule, does one loss make that big a deal according to the computers? Does any know or understand the equation?

We currently range between 4th and 20th in the computer rankings. But simply put, the teams we've beaten haven't turned out to be very good, or at least there's no way to confirm that they're any good yet. We haven't beaten a team that doesn't have at least one other loss, and we lost to a team that has two other losses. Our wins have come against teams with the following records:
1-6
3-3
1-5
4-2
4-3
3-3
total: 16-22

Compare that with Stanford, the only 1-loss team with an average computer ranking higher than Alabama's. Their loss came at 1-loss #4 Oregon. And the records of the teams that they've beaten:
3-3
3-3
2-5
4-3
5-2
total: 17-16

Considering that all the other teams currently ranked ahead of us in the computers are undefeated, that sounds pretty reasonable to me.

 

[Tideski]

Believe it or not, that helps me understand. We really need Auburn to stay undefeated until they roll into town. Looks like the SECCG will be a bust in as far as helping us move up in the polls. The Auburn game could almost take the place of the SECCG.
And I really do want them undefeated so we can take the wind out of their sails.
Assuming of coarse that we win out.

 

[PacadermaTideUs]

Can someone explain why the BCS computers have us ranked #12? With our schedule, does one loss make that big a deal according to the computers? Does any know or understand the equation?

We currently range between 4th and 20th in the computer rankings. But simply put, the teams we've beaten haven't turned out to be very good, or at least there's no way to confirm that they're any good yet. We haven't beaten a team that doesn't have at least one other loss, and we lost to a team that has two other losses. Our wins have come against teams with the following records:
1-6
3-3
1-5
4-2
4-3
3-3
total: 16-22

Compare that with Stanford, the only 1-loss team with an average computer ranking higher than Alabama's. Their loss came 1-loss #4 Oregon. And the records of the teams that they've beaten:
3-3
3-3
2-5
4-3
5-2
total: 17-16

Considering that all the other teams currently ranked ahead of us in the computers are undefeated, that sounds pretty reasonable to me.

popechild pretty much nailed it as to why Bama is 12th in the computer rankings, though Stanford's loss came to undefeated (not 1-loss) Oregon, who is #1 in all three human polls, #8 by the computers, and #2 in the BCS. Though each computer formula functions differently according to the creator's intent, in general, the computers only have us 12th because this early in the season, each team doesn't yet have a large enough sample size of games to be able to accurately factor in strength of schedule, and more importantly, performance vs common opponents. Consequently, simple W-L record is weighted proportionately more heavily, and a 6-win 1-loss team doesn't rank very highly when compared to 10 6-win 0-loss teams. As everybody builds a larger resume, various computer algorithms will get a better picture of the relative value between beating 1 team vs. beating another (or penalty that should be assessed for losing to 1 team vs another). As the season moves along, the cream will rise.


About the BCS in general...Without getting too technical, here's the abbreviated history of the BCS formula, and the changes made over the years as to how it has treated the computer polls:

1998: Formula devised averaging the AP and Coaches polls (Human average) and averaging three computer rankings (Computer average). Human and computer averages were combined with a strength of schedule factor, with a penalty applied for each loss. According to this formula, human judgement accounted for 1/2 of the formula, while computer number crunching accounted for the other 1/2 of the ranking. Each human poll accounted for 1/4 of the ranking and each computer poll accounted for 1/6 of the ranking.

1999: BCS committee correctly recognized that there was a wide range of rankings produced by various computer models. They therefore decided to include 5 more models (total now to 8), drop the lowest ranking and average the remaining 7. According to this formula, human judgement accounted for 1/2 of the formula, while computer number crunching accounted for the other 1/2 of the ranking. Each human poll accounted for 1/4 of the ranking and each computer poll accounted for 1/14 of the ranking.

2001: Concerned that teams were running up the score in order to improve their computer ranking, BCS dropped computer models which were heavily reliant on margin of victory (MoV) as a ranking parameter. 2 computers (NYT and Dunkel) were replaced with 2 others which did not use MoV. 8 total computer formulas still used, dropping the lowest and the highest and averaging the remaining 6. A quality win component was added to reward victories against top 15 teams. According to this formula, human judgement accounted for 1/2 of the formula, while computer number crunching accounted for the other 1/2 of the ranking. Each human poll accounted for 1/4 of the ranking and each computer poll accounted for 1/12 of the ranking.

2002: Still concerned with inflated scores against cupcake opponents, committee outlawed MoV usage at all. Two more formulas were dropped. One (Sagarin) changed his formula so that it didn't include MoV. The New York Times returned with a non MoV formula, and a new non MoV poll was added (total still 8). Highest/lowest dropped, remaining 6 still averaged. Quality win reward modified to only reward wins over top 10 teams. According to this formula, human judgement accounted for 1/2 of the formula, while computer number crunching accounted for the other 1/2 of the ranking. Each human poll accounted for 1/4 of the ranking and each computer poll accounted for 1/12 of the ranking.

2004: Following controversy over 2003 season when AP felt USC was #1 and LSU was #3 but LSU and Oklahoma played for the title due to LSU's strength in the computer rankings, BCS altered the formula to weight humans more heavily than computers. Now, instead of averaging the AP and Coaches and then averaging that figure with the computer average, the two human polls were considered independently. The 8 computers were pared down to 6, with the highest and lowest dropped and averaging the remaining 4. Quality wins were dropped. According to this formula, human judgement accounted for 2/3 of the formula, while computer number crunching accounted for the other 1/3 of the ranking. Each human poll accounted for 1/3 of the ranking and each computer poll accounted for 1/12 of the ranking.

2005-present: AP (media), still throwing a tantrum over their being ignored, withdrew from BCS. They were replaced by Harris Interactive. Otherwise, the formula is unchanged.

I provide that history to point out the fact that nearly every year from 2001 until 2005, the BCS altered its formula and imposed changes on the participating computer algorithms in such ways as to try to either minimize the computer influence, or to force their solutions into agreement with the human factor. In my opinion, the changes they made through 99 - to increase the computer sample size so as not to rely to heavily on any one computer solution - were appropriate. But since 2001, they've eliminated Margin of Victory (so that a 1pt win is worth the same as a 30pt win), and all but negated the weighting of the computer influence in deference to human opinion. The computers are now treated as little more than a modifier of human opinion.

Now certainly, there are bad computer models. Even more so since the BCS has started tinkering with and mandating what participating models can and can't factor into their rankings (MoV, etc). But if the point was to include objectivity into the formula in order to offset human opinion, minimizing that objectivity and trying to force it into agreement with that human opinion has been counterproductive.

 

[GreatDanish]

But since 2001, they've eliminated Margin of Victory (so that a 1pt win is worth the same as a 30pt win), and all but negated the weighting of the computer influence in deference to human opinion. The computers are now treated as little more than a modifier of human opinion.

Just to clarify, these were the reasons the MoV element was excluded.

1. It was only included in about half of the computer rankings at the time.
2. As weird as it sounds, MoV was a difficult variable to determine. Did you account for "garbage points" (for the winning team or the losing team)? A game that is 35-0 tells a different story than a game that was 72-47. The big nail in the coffin for it, though, was how to interpolate MoV with SoS. That is, how would a 47-0 over Utah State compare to a 31-14 win over Florida?
3. The human polls already accounted for MoV. Weak point, but nonetheless, it was one of the points. After all, human polls also account for W-L record and SoS too...
4. The complex web of games considered in the calculations of SoS was sufficient to determine the strength of a team.

The MoV definitely helps the computers be more accurate in their predictions, because it gives another level of data.

 

[PacadermaTideUs]

Just to clarify, these were the reasons the MoV element was excluded.

1. It was only included in about half of the computer rankings at the time.
2. As weird as it sounds, MoV was a difficult variable to determine. Did you account for "garbage points" (for the winning team or the losing team)? A game that is 35-0 tells a different story than a game that was 72-47. The big nail in the coffin for it, though, was how to interpolate MoV with SoS. That is, how would a 47-0 over Utah State compare to a 31-14 win over Florida?
3. The human polls already accounted for MoV. Weak point, but nonetheless, it was one of the points. After all, human polls also account for W-L record and SoS too...
4. The complex web of games considered in the calculations of SoS was sufficient to determine the strength of a team.

The MoV definitely helps the computers be more accurate in their predictions, because it gives another level of data.

And point 2 above is a valid reason for a statistician to be cautious when using MoV. Nonetheless, The problem I have with the BCS's having removed it altogether is that the committee wasn't and isn't composed of statisticians. It would be like someone with only very lay knowledge about tax law but concerned about making sure to do things the right way, hired an accountant to do his taxes. Then when confronted with a final figure he didn't like, he demanded that the CPA exclude certain sources of income that the accountant, in his expertise, believed should be reported.

Either you want an objective analysis or you don't. When you start tinkering with someone else's objective analysis because you don't like what it's telling you, it ceases to be objective. To a T, each computer poll is statistically shown to be more accurate, both predictively and retrodictively, without the BCS's tinkering.

The third reason above, though correct, seems irrelevant. Of course human opinions account for MoV and SoS and W/L and home field advantage and any number of measurables and immeasurables in developing an overall ranking. Human opinion even develops an overall ranking. So by the logic of the third point above, why even consider the computer polls at all? Why not just stick with the AP (or the Coaches, or Harris, or Legends, etc) and be done with it?

For objectivity. Human opinion considers all of these factors subjectively in developing its ranking, which is why its ranking itself is subjective. Is it less accurate? More accurate? Not necessarily. Human judgement can sometimes discern key differences that pure numerical analysis cannot reveal. But in general, the rules that are subjectively applied by humans in evaluating MoV, SoS, etc, are done so unevenly and discriminately, because the human animal is simply not capable of applying a set of rules and interacting variables completely and evenly across a field of 244 football teams (including FCS programs, which must be included since FBS teams play them) which are all interconnected through common opponents. The human brain is simply not capable of unwaivering, accurate and indiscriminate application of extremely complex rule-sets on that large of a scale. Hence, the inclusion of computer algorithms.