Thursday, October 28, 2010

Why Computers Still Won't Matter

There are a number of dissenting comments on my piece - The Meaninglessness of Computers. A couple of posters didn't quite agree with my analysis and a few others simply think that this year will be different. That this year, for the first time since the last BCS formula overhaul in 2004, a team that's either No. 1 or No. 2 in the polls will miss the BCS title game.

My response: Still not going to happen.

The primary reason why a No. 2 team (let along a No. 1) in the polls will not be jumped by another team in the BCS title game is this: The voters have learned over the years to game the system. They know how much power they wield, but more importantly, they know how to use that power.

Because the BCS formula is set up so heavily titled in favor of the human element, the voters essentially have veto power over the computers. That veto power is manifested in the way they've voted in the two teams they want to see in the BCS championship game.

The last time the computers might have made a difference was 2006, when Florida and Michigan were jockeying for the No. 2 spot to face Ohio State in the BCS title game. But the computers finished in a dead heat between these two teams, so the voters' desire to shift to the Gators proved decisive.

It hasn't been close since.

Since 2006, the voters have learned that the way to make sure their 1-2 choices would face each other is to make very clear who's No. 2. In the BCS setup, the differential between each place in each of the three components is .0400. Of course, that margin may expand or shrink based on the actual vote count. But if every voter put Team A at No. 2 and Team B at No. 3, the gap would be .0400 in that poll.

Now, take a look at the voting results between the second and third place teams from 2007-2009, in each of those years a team ranked either first or second in the computers failed to make the BCS title game:

2007: Coaches poll gap - .0386, Harris poll gap - .0580. Total - .0966
2008: Coaches poll gap - .0294, Harris poll gap - .0478. Total - .0772
2009: Coaches poll gap - .0495, Harris poll gap - .0498. Total - .0993

Those kinds of gaps are difficult for the computers to overcome. In essence, the aforementioned Team B would need to be 3 full places ahead of Team A in all computers to overcome the difference. It doesn't happen.

In 2007, Virginia Tech was No. 1 in the computers, but the yawning gap in the polls that LSU owned ensured the Tigers got to play for the BCS title. The same went for computer No. 2s Texas and Cincinnati, respectively against Florida and Texas in '08 and '09.

So of course in 2010 this hypothesis will be applied to Boise State. At the moment, it's easy to deduce that the Broncos, despite being No. 2 in the polls, may fall short in the final standings because of their deficiencies in the computer rankings.

But we're not at the final weekend yet, where the standings will be the only one that matters. The voters are typically vacillating with their ballots right now. And with the protection of anonymity, they're free to make dramatic changes without having to give any reason or being detected by anyone.

On the final weekend of the season, their views will harden and they will make a determination on whom they want in the BCS title game. While the voters don't vote as an entire bloc in either poll, a consensus usually emerges and most everyone toes the line. Besides, some voters have been known to rig their ballots by placing a competitor team way down to diminish its chances.

While the voters seem content to place Boise State at No. 2 right now, don't be so sure that they'll still feel that way if Auburn goes through the rest of the schedule unscathed. Conversely, should Auburn, Michigan State and Missouri all lose a game before the end of the season as expected, then the voters will make a clear determination between Boise State and TCU/Utah, and that will be enough for the Broncos to overcome whatever might be their computer deficit.

The only wild card this year is Alabama, which is the only one-loss team that has the potential to jump Boise State in the polls. All other one-loss teams currently in contention lost to teams that are still ahead of them but behind Boise State, making such a jump extremely unlikely.

But even if the Tide re-emerge as a factor, the computer-ranking differential between Alabama and Boise State will not be as significant as what the voters ultimately decide. If they feel that a one-loss Alabama is more worthy than an undefeated Boise State, then expect them to defect to the Tide at No. 2 en masse, once again rendering the computer rankings meaningless.

So it's back to the polls. This is where truly every vote counts.

9 comments:

tyamdaly said...

Then why have the computer polls in the first place. Since they don't matter. They would matter if they used all six. And those computer are allowed to use their "full" formulas. Even though there are only 2 that we know of that has different ranking in their full formula then what they are "told" to use. Those are the 2 that are pretty drastic in their rankings. Leave the system alone and take the top 8 and have a f***** playoff. You could still have your major bowls and you could add a few more great games in December. The lost NCAA football month. At least until X-Mas. It's all about the $$$$$ going to certain people. Thats it. When they could have more interesting games which would give them more money but then they couldn't control where it goes, could they?

Southern Duck said...

@Guru,

I don't have as much heartburn with this article as your first, mainly because you removed some of your hyperbole. Where I disagree with you is that I believe the computers do have an influence in two scenarios.

First is settling a virtual tie amongst the voters. As you noted, they had the chance to decide things in 2006, but they were also deadlocked. I think it is naive to think that the voters will never be split again on who should be the #2 team, and that they will always arrive at this magical consensus.

The second scenario (and the one I still believe we may see played out this year) is to keep a team with a particularly weak SOS out of the CG. I don't think this was an intentional function of the computers. But at some point the voters run out of power to overcome a weak computer ranking.

Let's look at Boise State's four undefeated seasons. Assume for those years that Boise State actually got 100% of the #1 votes in both polls, and that the teams above them all lose .04 points from each voting poll to reflect BSU jumping ahead of them. In 2004 and 2008, the voter's would not have had enough power to vote BSU in even with all #1's. In 2006 and 2009 they would have had just barely enough, with BSU needing a .985 average from the voting polls in 2006 and .987 in 2009. My point is that there are limitations on how weak of a computer ranking the voters can actually overcome.

Historically, an undefeated BSU has been ranked behind all other undefeateds as well as almost all one loss AQ teams, finishing between .73 and .78. Their final computer rankings this year will be dependent on how many no-loss and one-loss majors remain, but to state that "the computer-ranking differential between Alabama and Boise State will not be as significant as what the voters ultimately decide," ignores history. The computer advantage of a one-loss SEC champ over an undefeated BSU has ranged between .12 to .16. In BSU's favor is that the WAC as a whole is having its best season since the MWC split off in the late 90's. The WAC looks to post their best OOC winning pct since 1994, and that should help BSU. The voters will have very limited power if BSU finishes with a .73 computer score like they did in 2008. But, if they can post something in the mid .8s, then power shifts back to the voters.

Bottom-line, you still haven't convinced me that the computers won't matter, especially for Boise State.

Southern Duck said...

@Guru,

I don't have as much heartburn with this article as your first, mainly because you removed some of your hyperbole. Where I disagree with you is that I believe the computers do have an influence in two scenarios.

First is settling a virtual tie amongst the voters. As you noted, they had the chance to decide things in 2006, but they were also deadlocked. I think it is naive to think that the voters will never be split again on who should be the #2 team, and that they will always arrive at this magical consensus.

The second scenario (and the one I still believe we may see played out this year) is to keep a team with a particularly weak SOS out of the CG. I don't think this was an intentional function of the computers. But at some point the voters run out of power to overcome a weak computer ranking.

Let's look at Boise State's four undefeated seasons. Assume for those years that Boise State actually got 100% of the #1 votes in both polls, and that the teams above them all lose .04 points from each voting poll to reflect BSU jumping ahead of them. In 2004 and 2008, the voter's would not have had enough power to vote BSU in even with all #1's. In 2006 and 2009 they would have had just barely enough, with BSU needing a .985 average from the voting polls in 2006 and .987 in 2009. My point is that there are limitations on how weak of a computer ranking the voters can actually overcome.

Historically, an undefeated BSU has computer rankings behind all other undefeateds as well as almost all one loss AQ teams, finishing between .73 and .78 in the computers. Their final computer rankings this year will be dependent on how many no-loss and one-loss majors remain, but to state that "the computer-ranking differential between Alabama and Boise State will not be as significant as what the voters ultimately decide," ignores history. The computer advantage of a one-loss SEC champ over an undefeated BSU has ranged between .12 to .16. In BSU's favor is that the WAC as a whole is having its best season since the MWC split off in the late 90's. The WAC looks to post their best OOC winning pct since 1994, and that should help BSU. The voters will have very limited power if BSU finishes with a .73 computer score like they did in 2008. But, if they can post something in the mid .8s, then power shifts back to the voters.

Bottom-line, you still haven't convinced me that the computers won't matter, especially for Boise State.

Southern Duck said...

Not sure why the double post happened. Sorry bout that.

Anonymous said...

"All other one-loss teams currently in contention lost to teams that are still ahead of them but behind Boise State, making such a jump extremely unlikely."

Stanford lost to Oregon, which is ahead of Boise. But I don't think Stanford will pass Boise.

Anonymous said...

You are wrong -- the computers do matter. If Boise State is #2 in the human polls at season end the computers will still send a #3 ranked Auburn. Am I wrong...or are you?

The Guru said...

No, that's exactly the point. The voters' massive defection from Boise is precisely the 1 always plays the 2 since 2004. They manipulate their ballots so the computers won't matter.

It's more psychology than mathematics.

tyamdaly said...

Oh how wrong you are. The computers matter now more than ever. Here is my projections as of 11/6/2010 at 11:10 edt.

1. Oregon
2. TCU
3. Auburn
4. Boise St.
5. LSU
6. Wisconsin
7. Stanford
8. Nebraska
9. Ohio St.
10 - 25. Does it matter now?

Anonymous said...

Man your fuckin crazy.......Your website sucks bad.... Everything you say is insane...

Google