Tuesday, December 23, 2008

Remaking the BCS Formula

Part I of the Troubleshooting the BCS Series

After another season of high-wire acts, the temptation of course is to blow the BCS to smithereens and start over. But that's neither practicable nor desirable.

It's not practicable because the current BCS contract does not run out until after the 2013 season, so any wholesale change toward a playoff now or in the foreseeable future is just not reality based. It's not desirable because without thoroughly considering all the issues - logistics and otherwise - any move made in haste would only invite more disaster.

There is something that could be done - immediately - about the BCS within the framework of the current structure. It's possible to implement changes even before the next season, as the BCS has proved to be quite nimble when it comes to making up stuff as we go - witness this past season's sudden decision to allow three conference teams to be included in BCS bowl games.

When the commissioners gather for their annual meeting in April, instead of patting each other on the back and passing out vapid congratulations, they would be better off re-examining the BCS. And the best place to start would be the formula that creates the BCS standings.

The formula in its current form has survived since 2004, with the only change being that the Harris Poll replaced the Associated Press poll before the 2005 season. The current formula, created in the aftermath of the USC-LSU split championship in 2003, places overwhelming emphasis on the human polls. This is troubling because the most unstable and subjective part of the formula, naturally, is the human polls. And even more troubling is the utter lack of transparency of these two polls.

That, and a few other things, may be fixed rather quickly and easily. So here's how:

1. Bring transparency into the polls - Neither the coaches poll nor the Harris poll reveals its weekly voting results until the final week of the regular season, when the final BCS standings are released. Any system that's worth its salt demands accountability. Imagine if your congressman never had to reveal how he voted until the last bill of his term. That just won't do.

It's easy enough for the BCS to demand the Harris poll make its ballots public every week. The BCS commissioned the Harris poll, therefore owns it. If any voter objects to this new guideline, then see you later. There will be plenty of people willing and able to vote in the Harris poll.

The coaches poll is the more thorny issue. For years the coaches have resisted making their ballots open to the public and it took a near act of congress to get them to reveal the final poll. And since the BCS champion ultimately is merely the coaches poll champion, there's not a whole lot the BCS can do about it, right?

Not really. I believe there is a creative solution. The BCS commissioners can simply issue an ultimatum to their coaches to open the ballots, or else risking to have their votes become even more irrelevant. I'm sure the Black Coaches & Administrators will be more than happy to step in with their own poll. The BCA is not happy with college football's hiring practices - and rightly so - but this will be a way to increase its visibility. For the BCS, this will be a public relations coup. So if the coaches don't play ball, simply have their poll replaced by a new BCA poll.

2. Eliminate tiebreaker madness - What happened in the Big 12 Conference this year is simply unacceptable. The BCS formula was never conceived to determine anything but the top two teams for the BCS title game. The easy thing to do here is for the commissioners to police their own conferences to make sure the BCS formula will never come close to being any sort of a tiebreaker ever again.

3. Allow margin of victory back in the computers - Following the 2001 season, in a bit of a knee-jerk reaction, MoV was purged from all computer rankings ostensibly to discourage teams from running up the score. The problem is, the human voters didn't get the memo. The first fallout from this decision took place in 2003, when Oklahoma, after gotten blown out by four touchdowns in the Big 12 championship game, still stayed No. 1 in the BCS standings thanks to the computers now unable to distinguish a one-point loss from a 28-point rout.

Computer programmers should write their own formulas. The BCS has no business telling what these guys should do. It should demand to audit the results throughout the season, but leave the programming to the professionals. After all, all computer programs had built in a safeguard against excessive scoring, with either 21 points or 28 points as the threshold.

Computer rankings become much more pure and accurate when margin of victory is taken into consideration. Just think about this for a moment: If you were betting your house on a game, would you dare to look at only a team's won-loss records but not how much they won and lost by?

4. Re-calibrate the computer ratings - One of the dumbest things about the BCS formula is it mixes percentages with ordinal numbers. For the human polls, it counts actual votes for teams instead of rankings; but for computers, it uses the rankings only. So when a team falls from 2nd to 3rd in the computers, the consequences are disastrous; but when a team falls from 2nd to 3rd in the polls, it could be just about a dead-heat, as far as the BCS standings are concerned.

The BCS should keep the human part of the formula the way it is but normalize the computer rankings so they become compatible with one another numerically. An easy way to do this is to fix the top team at 1.000 and derive percentages for other teams based on their respective computer scores.

I checked all six BCS computer systems and found that their scores are fairly uniform. Teams ranked 10th score about .86 to .92 points and teams ranked at No. 25 score about .74-.81. What this does is that it stabilizes the system and reduces the wild swings toward the end of the season when a games not even involving the principals holds such sway as the Cincinnati-Hawaii game did at the end of the 2008 regular season.

5. Re-balance the formula - If steps 1-4 are taken, then we can move on to this, which should bring objectivity back to the formula and take some pressure off the voters. I mean, if we're just gonna let the voters decide everything, why even bother with a formula - we can just use the AP poll instead. After all, since the adoption of the current formula, the AP has produced the same top two teams every season anyway.

By going 50-50, the new standings also give the non-BCS teams a fighting chance. The human polls are always stacked against non-BCS teams, but the computers give them a fairer shake. And if we ever move toward at least a plus-one system involving four teams, then a team like Utah in 2008 might have a shot at qualifying.

Did I say Plus-One? Ah, you'll have to come back for Part II.

8 comments:

Jams said...

Very good stuff, as usual. I appreciate the realism involved in your ideas, since it's something sorely lacking amongst college football's observers.

I wonder if there's any effective way to lobby BCS commissioners, et al to affect some of these changes. They're by no means radical, so there's got to be some way to present it to them.

Ben Prather said...

All these ideas are excellent.

#1 is fulfilled for the last poll, the only one that really matters. BCS polls released before then function to market the system and get people talking, and function as an early indicator on the side.

#2 is the conference's issue, not the BCS's

#3 While I strongly agree than MOV greatly increases a computers performance, I like the ideal the standard presents. Besides, the pollsters go crazy with MOV so why not have the balance?

What is needed more is a systematic review of the computers used. If Billingsley is going to consistently be an outlier and be a poor indicator of who will win a bowl game maybe another computer would fair better and should replace it.

#4 Use each computer's raw data. Use a linear scale to place the #1 team at 1.00 and the #25 team to 0.04.

#5 If step #3 is taken the computers would not favor the non-AQ conferences as much as they do.

David said...

1. If I'm reading what you're saying correctly, you want the BCA to playing PR blackmail with the BCS. If so, love it! :) This ballot secrecy is a complete joke. Your comparison to a secret congressional vote is on the mark.

2. I agree in spirit with what you're saying. I suppose my only disagreement is that each conference should have the right to set any tiebreaker criteria they want. If the Big XII wants to base it completely on what gives it the best chance to have the highest BCS ranking so be it. From a fair play standpoint, they should think of something that measures an on-field performance to be their tiebreaker.

3. Again, I see your intellectual quarrel with the BCS telling math geeks what formulas to use. But these computer geeks are working at the behest of the BCS. The BCS is including computer rankings a, b and c and not d and e. So if the BCS it picking what computers to select, it seems like the BCS having a say-so in what they measure seems fair.

4. Again, I see your intellectual quarrel with the BCS. It's an apple-and-orange mixed bag when you count one system by ranking and one by percentage. The only argument I'd make for the current system is that, be definition, aren't you already mixing apples and oranges with polls and computer rankings, anyway? Calibrating what one vote/point means in a poll is easy to measure and compare. Coming up with a percent table that can be applied to all the computer rankings seems to be an impossible task. If I bought into your line of logic (and I think I agree with you on a grand level), then let's make it all ordinal -- polls and computers. In your 1.000 proposal, that's still problematic. I don't know if .0003 off in one computer means in relation to .0003 of in another computer.

5. I don't mind giving voters the big advantage here. In my mind, the computers are only around to settle tied or really close scores/resumes.

Phil said...

David, it's very easy to come up with the percentage.

All of the computer rankings are based on a rating #. Each team's rating divided by the #1 team's rating and you get your number. It's not a direct equivalent to the human polls, because in those someone doesn't always get a unanimous #1 and a perfect score for that poll.

This would also solve another problem that Guru didn't mention, and let all of the computer rankings matter. The fact that the BCS looks at a computer ranking of 26 and a computer ranking of 50 the same is dumb. If you do the percentage, every team can get 6 computer scores.

Amos said...

#1, completely agree, and on that note, I think the computer guys need to explain their system a bit, Colley is the only one who does this in full now, and for transparency I think we need to see how the computer rankings are being determined.

#3, completely agree, some programmers think MoV is important, some don't, let them decide, just like you're letting pollsters decide if they want to let MoV factor in.

#4. I think you may be looking at the math here a bit incorrectly. The way the computer rankings are being tallied is as if each computer poll is a human pollster. The reason you get so much swing out of it, is in reality you're using 4 computer pollsters (after throwing out high and low), unlike the Coaches poll which uses 63 pollsters. Allowing the computers to use their percentages (something I do actually agree with), would be making it different than the human polls, not the same.

Say a single computer decided that Florida and Oklahoma were equally good, they could get similar numbers from the computer. But say a pollster thought Florida and Oklahoma were equally good, the pollster still has to rank one of them a full ranking ahead, and hope that enough people think they're equally good and rank them possibly reversed, in order for the polls to come out showing equality.

Anonymous said...

Why not a honest straight up play off system , thats the fair way.
The BCS is not fair.

RA said...

What's most ridiculous about the whole system is that the weekly polls are not based on 10-14 weeks of real judgement, but rather about slotting teams against pre-season expectations. The Harris Poll, even if not designed to mimic the AP poll, has to virtually mirror it in order to have any "credibility" (a loosely used word here).

That Utah beat a former #1 team in Alabama is only relevant inasmuch as the weekly slotting once leaned Alabama's way. Judging the teams on their play THIS season (without prerankings or expectations), Alabama is a "top 10" team ... but certainly never would have a "number 1" team.

Anyway, yeah, Utah probably has earned the title.

Anonymous said...

I am wondering about what you think of the idea of a "relegation" game in the BCS?

In many Soccer leagues and other sports, they have "relegation", where the bottom team or teams from the Premiere league play against the top team or teams from the lower leagues for the right to stay in the premiere league.

What do you think about having something like that for the BCS? What if the lowest ranked AQ conference champ in a given year played the highest ranked non-AQ conference champ, and the winner's conference would be an AQ conference the next year?

So, as an example ...
In 2004, the 6 AQ Conferences were (in order):
PAC, BigXII, SEC, ACC, BigTen, BigEast, and the highest ranked non-AQ was the MWC. So you would have Utah play Pitt in one of the BCS bowls. In this case we actually had the game, and Utah won.

In 2005 then, the 6 AQ conferences were (in order)
PAC, BigXII, BigTen, SEC, ACC, MWC, and highest ranked non-AQ would have been the Big East. So West Virgina would have played TCU. TCU had a great running Defense, but West Virgina had White, so it would have been a great game. Just to make it easier for this example, lets assume that higher ranked West Virgina (11) beat TCU (14).

In 2006, then the 6 AQ Conferences would then have been (in order):
BigTen, SEC, PAC, BigEast, BigXII, ACC, and the highest ranked non-AQ would have been the WAC with Boise State. So Boise State would have played Virgina Tech. Again, I'm going to assume that the higher ranked team won, and Boise State beat Virgina Tech.

In 2007, the 6 AQ Conferences would then have been (in order):
BigTen, SEC, BigXII, PAC, BigEast, WAC, and the ACC would have been the highest non-AQ. So Virginia Tech would have the chance to win back the AQ by beating Hawaii.

So this year, the 6 AQ Conferences would have been:
BigXII, SEC, PAC, BigTen, BigEast, ACC, and the highest ranked non-AQ would have been the MWC, so Utah would have played Virginia Tech, and the winner would be a AQ conference next year.

-----------

For these examples, I didn't have any upsets. It actually gets really interesting if there are some upsets.
For example, if TCU were to have upset Pat White and West Virgina in 2005, then in 2006, the relegation game would have been BYU vs Louisville, and Boise State would have been left out.

Anyway, I wonder what you think of this idea?
One thing I like about it is that it makes the "worst" BCS game really interesting for all the fans of two conferences. It has been reported that this year's Orange bowl was one of the lowest rated TV broadcasts, probably in part because it had little interest to other fans besides Cincy and VTech. On the other hand, a game for a BCS slot next year between VTech and and Utah would have been of great interest to all of the ACC and all the MWC.

------------

If I were the Tzar of College Football ....
I would have 4 BCS bowls, with #1 vs #4 in the home bowl of the #1 team, and #2 vs #3 in the home bowl of the #2 team, a "relegation" game, one other BCS bowl, and a rotating championship about 7 to 10 days after New Years.

3 out 4 games "mean something" and let the winners move on, which I think would add to the Television Ratings and probably the butts in the seats.

Google