Florida or Alabama in the BCS championship game? Predictable. TCU as their opponent? More likely than you think.
Here are the latest BCS rankings. The press has been covering the changes by saying that USC had moved up to #5, while TCU passed Boise State. Basically, all that happened is that Boise State and Cincinnati fell three spots each, from #4 and #5 to #7 and #8. Iowa, USC, and TCU slid up from 6, 7, and 8 to 4, 5, and 6. But in terms of game-play, #3 Texas was the only impressive team in the top 5, beating Missouri on the road 41-7. #1 Florida struggled at Mississippi State, 29-19; #2 Alabama barely beat Tennessee at home, 12-10; #4 Iowa just edged Michigan State 15-13, and #5 USC got into a shootout at home against Oregon State, prevailing 42-36. Yet all four either held their spot or moved up, thanks to the BCS computers which aren't allowed to use margin of victory.
#7 Boise State and #8 Cincinnati both won big—the Broncos 54-9 at Hawaii and the Bearcats 41-10 over Louisville—but the wins over weak teams didn't get them many points in the computer rankings. TCU was the big winner on Saturday, gaining cred with both voters and computers for their 38-7 trashing of BYU.
Now we have to ask whether USC could get into the national championship ahead of an undefeated BCS-conference team. It's assumed that an undefeated Florida/Alabama and Texas have nothing to worry about. But Iowa and Cincinnati could be passed by the Trojans.
Iowa has a lot of upward potential in the Coaches and Harris polls, however, and is #1 in the computer rankings. Even if they keep winning by the skin of their teeth they'll remain ahead with the computers, and a few solid wins will allow them to move up in the polls. It might be close, but Iowa probably edges USC if a spot opens for the national title game.
Cincinnati has problems, however, being three spots behind the Trojans. Worse, they're behind TCU in the computer rankings and roughly tied in the polls. Next week's Syracuse game won't help them much, and having Illinois on their schedule is now a detriment. They should pass Boise State at some time, but a lot of teams might have to lose to help an undefeated Bearcats team make the national championship game.
Let's look again at each team's remaining schedule and project each game, giving it a percentage-to-win value. Then we'll compute the team's overall % chance to finish the season undefeated. We'll do this by using the X-conference 726-team power rating to give every game a team has played a rating; we then cross-compare all of these ratings with the game ratings of their opponent and see how many of the cross-comparisons are victories. Basically it's a way to look at all the aspects a team has shown so far in the season, from their worst to the best performances, since on a given day any version of a team could "show up." The method is explained in more depth here.
For each team a record is projected: First, simply by counting wins and losses based on whether a team is favored in that game, and second, by adding the percentages to obtain a cumulative projection, or rather the record a team is most likely to finish at. Odds of going undefeated are computed by multiplying the odds of winning each game.
Florida remaining schedule & forecast |
||||||||
| 10/31 | Sat | N | *Georgia | 85.7% | W | |||
| 11/7 | Sat | vs. | *Vanderbilt | 99.9% | W | |||
| 11/14 | Sat | @ | *South Carolina | 92.9% | W | |||
| 11/21 | Sat | vs. | Florida International | 99.9% | W | |||
| 11/28 | Sat | vs. | Florida State | 89.8% | W | |||
| 12/5 | Sat | vs. | Alabama? | 50.0% | ?? | |||
Favored: 12-0-1 Projected: 12-1 |
||||||||
Odds of going 12-0: 71.5%, 13-0: 35.7% (up 0.2%) |
||||||||
Florida Gators (7-0; BCS rank: #1)
The Gators' problems with Mississippi State gave the rest of the teams on their slate a bit more hope—but not much. Georgia and South Carolina increased their odds, but Vanderbilt and FIU still have only a snowball's chance.
But Alabama struggled, too, and that tilted the inevitable confrontation from a slight 'Bama win into a dead heat. As a result, while Florida's chances of going 12-0 fell 3 percentage points, their odds of going 13-0 went up by 0.2%.
The projected number of wins declined slightly from 12.198 to 12.184, basically saying that a 12-1 season is the "expected value" outcome. The game-comparison system doesn't pick a favorite in the SEC title game but the power rating it's based on does: Florida is slightly less than a point ahead of Alabama now, and therefore would be considered the favorite. But it's already assumed that one of these two teams would be in national championship game, and I'll compare both to Texas and any other teams that have a good chance of being their opponent.
Alabama remaining schedule & forecast |
||||||||
| 11/7 | Sat | vs. | *Louisiana State | 89.3% | W | |||
| 11/14 | Sat | @ | *Mississippi State | 93.8% | W | |||
| 11/21 | Sat | vs. | Tennessee-Chattanooga | 99.9% | W | |||
| 11/28 | Sat | @ | *Auburn | 81.3% | W | |||
| 12/5 | Sat | N | Florida? | 50.0% | ?? | |||
Favored: 12-0-1 Projected: 12-1 |
||||||||
Odds of going 12-0: 68.1%, 13-0 34.0% down 1.9% |
||||||||
Alabama Crimson Tide (8-0; BCS rank: #2)
Having Tennessee out of the way really helps Alabama's odds of going 12-0, taking away the uncertainty of what was seen as their toughest test before likely facing Florida. But their poor performance in the game hurt them, increasing the odds that other teams on their slate can beat them.
LSU's odds jumped four percentage points; Mississippi State's by the same; and Auburn's odds by six and a half. Only Chattanooga didn't gain any semblance of hope from the Tide's narrow win.
Overall the Tide slightly regressed, decreasing their odds of going 12-0 from 68.5% to 68.1%. That's pretty remarkable for getting a win that wasn't a foregone conclusion. Normally with each win the odds should take a sizeable jump, but Alabama's performance introduced more uncertainty than the win removed. Their odds of going 13-0 took a hit, too, falling from 35.9% to 34.0%.
Texas remaining schedule & forecast |
||||||||
| 10/31 | Sat | @ | *Oklahoma State | 79.6% | W | |||
| 11/7 | Sat | vs. | Central Florida | 99.9% | W | |||
| 11/14 | Sat | @ | *Baylor | 99.9% | W | |||
| 11/21 | Sat | vs. | *Kansas | 95.9% | W | |||
| 11/26 | Thu | @ | *Texas A&M | 85.7% | W | |||
| 12/5 | Sat | vs. | Nebraska? | 69.4% | W | |||
Favored: 13-0 Projected: 12-1 |
||||||||
Odds of going 12-0: 65.7%: 13-0: 45.4% down 11.6% |
||||||||
Texas Longhorns (7-0; BCS rank: #3)
Texas didn't remove any uncertainty by beating Missouri, whom they were basically given a 100% chance to beat by the game-comparison system. And by winning in such a forceful fashion, they should have further decreased many of their future opponents' meager odds of beating them.
But Oklahoma State's showing against Baylor pushed the Cowboys up in the ratings and created a major re-evaluation of the game-pairs against Texas; now the Cowboys win a few more close contests and from a 5.6% chance last week they are now given a 20.4% chance at home.
UCF and Baylor still have no hope, but Kansas's odds increase slightly, and Texas A&M's take a big leap after their blowout win over Texas Tech. So despite the Longhorns playing perhaps their best game of the year at Missouri, the road to a 12-0 season just got a lot tougher, dropping from 89.2% to 65.4%. In fact, just the game at Oklahoma State is now seen as more dangerous that their entire remaining schedule was just last week. This is likely the result of having more data; last week's odds were inflated, and have corrected more toward reality.
Another huge development for Texas is the uncertainty in the North division. With Nebraska losing it's harder to project an opponent. We could go through a rigorous process of giving a percentage chance for each possible opponent, but that's beyond the scope of these projections (and beyond what I'm willing to calculate). Instead, forecasting game by game, here is the predicted finish in the North:
Team conf all
Nebraska 5-3 8-4
Missouri 4-4 8-4
Iowa State 4-4 7-5
Kansas 3-5 7-5
Kansas State 3-5 5-7
Colorado 1-7 2-10
So Texas is still likely to face Nebraska, in Dallas, for the Big Twelve championship. The Cornhuskers' odds of getting an upset have decreased from 36% to less than 31%. Still, Texas' chances of going 13-0 have declined from last week, from 57% down to 45.4%. The system is saying that it's more likely than not that they suffer a loss somewhere along the way.
Iowa remaining schedule & forecast |
||||||||
| 10/31 | Sat | vs. | *Indiana | 93.8% | W | |||
| 11/7 | Sat | vs. | *Northwestern | 95.3% | W | |||
| 11/14 | Sat | @ | *Ohio State | 43.8% | L | |||
| 11/21 | Sat | vs. | *Minnesota |
85.9% | W | |||
Favored: 11-1 Projected: 11-1 |
||||||||
Odds of going undefeated: 33.6% up 10.2% |
||||||||
Iowa Hawkeyes (8-0; BCS rank: #4)
Iowa got rid of a big chunk of uncertainty by beating Michigan State, last second TD pass or not. But the odds for their remaining games—including a projected loss at Ohio State—remain roughly the same. So while the Hawkeyes odds of going undefeated take a big jump, they don't surpass the critical 50% threshold.Home games against Indiana and Northwestern are still seen as safe, and the road trip to Ohio State is still a likely loss. Even if they get past that, Minnesota offers somewhat of a challenge. Things are looking better than last week, when Iowa had less that a 1 in 4 chance of going undefeated; now they have about a 1 in 3 chance. Still not great, but better.
Not only that, but as mentioned above their chances for going to the national title game should they go undefeated are looking pretty good.
TCU remaining schedule & forecast |
||||||||
| 10/31 | Sat | vs. | *UNLV | 99.9% | W | |||
| 11/7 | Sat | @ | *San Diego State | 95.9% | W | |||
| 11/14 | Sat | vs. | *Utah | 91.8% | W | |||
| 11/21 | Sat | @ | *Wyoming | 99.9% | W | |||
| 11/28 | Sat | vs. | *New Mexico | 99.9% | W | |||
Favored: 12-0 Projected: 12-0 |
||||||||
Odds of going undefeated: 88.0% up 32% |
||||||||
TCU Horned Frogs (7-0; BCS rank: #6)
More than any other team, by far, TCU improved both their odds of going undefeated and their chances of going to a BCS bowl game. They even considerably improved their chances of playing for the national championship.
The strong win over BYU bumped them up in the voter polls, improved their power rating, and took away the majority of the uncertainty from the rest of their season. BYU had a 40% chance of upsetting the Horned Frogs, contributing most of TCU's 44% chance to lose a game.
But with more game data becoming available, a few of their opponents' chances increase. San Diego State goes from hopeless to a 4% chance of upset. Utah makes a slight gain, too. But UNLV, Wyoming, and New Mexico are judged foregone conclusions. All told, TCU now has an 88% chance of going undefeated, up from last week's 56%.
But like last week, what that will mean is unknown. Now that we're projecting Texas to lose a game, this creates an opening, one that Iowa can't fill due to their projected loss. TCU is next in line among the undefeated teams, but is behind USC. And while Boise State probably won't catch them, Cincinnati could gain support from voters and even the computers since TCU's remaining schedule is soft overall.
The USC question is relevant: will the Trojans win out? The answer, after looking at just their next game, is no. Probably not.
USC (6-1, BCS#4) odds
@Oregon: 40.8% L
@Arizona St: 83.7% W
Stanford: 73.2% W
UCLA: 83.7% W
Arizona: 71.4% W
Favored: 10-2 Cume: 10-2 (9.53 wins)
Odds of winning out: 14.9%
Oregon is a 59% favorite by the game-comparison system. And several of their other games could present a challenge, too. Overall USC is expected to barely manage to go 4-1. Beating Oregon would help a lot, but even then they'd be expected to drop one of their last four games. So for the purposes of this projection USC can be discounted for now.
The upshot is that TCU is close to being a favorite for the national title game. It will come down to the particulars of the BCS system, but right now Texas, Iowa, and USC all seem to be facing a loss. Of the three, only the Longhorns' loss is a "cumulative" loss, i.e. there is no specific team that is pegged to beat them. They are favored to win every game. So seeing Texas lose is the most important thing for TCU fans.
Boise State remaining schedule & forecast |
||||||||
| 10/31 | Sat | vs. | *San Jose State | 99.9% | W | |||
| 11/6 | Fri | @ | *Louisiana Tech | 93.9% | W | |||
| 11/14 | Sat | vs. | *Idaho | 96.4% | W | |||
| 11/20 | Fri | @ | *Utah State | 93.9% | W | |||
| 11/27 | Fri | vs. | *Nevada | 87.8% | W | |||
| 12/05 | Sat | vs. | *New Mexico State | 99.9% | W | |||
Favored: 13-0 Projected: 13-0 |
||||||||
Odds of going undefeated: 74.6% up 1.1% |
||||||||
Boise State Broncos (7-0; BCS rank: #7)
Boise State shredded Hawaii 54-9 last weekend with a particularly strong defensive performance in a road game. The win itself didn't do much for their odds, as they were a 97.2% favorite, but the winning margin helped their odds in the upcoming games against Louisiana Tech and Idaho. But Nevada, who beat Idaho badly, added a bit to their odds for the upset. San Jose State and New Mexico State remained near-certainties.
When all is said and done, Boise State's odds of going undefeated increase about a bit less than you'd expect after knocking off a foe like Hawaii. From 73.5% last week, the Broncos now have a 74.6% chance of winning all six of their remaining games. With almost a 3 in 4 chance of being 13-0, you'd think they'd be sitting pretty, but a few bad developments hit in the BCS rankings. Mainly, TCU passing them—coupled with the Horned Frogs' now-overwhelming odds of going undefeated—mean that Boise State may be left out of a BCS bowl at 13-0.
Broncos fans could hope for a TCU loss, but this year there's actually a decent chance that two non-BCS teams could be in BCS bowls, especially if TCU makes it to the national title game. So maybe Boise State fans should root for possible BCS bowl teams like Notre Dame to lose, which is more likely.
Cincinnati remaining schedule & forecast |
||||||||
| 10/31 | Sat | @ | *Syracuse | 99.9% | W | |||
| 11/7 | Sat | vs. | *Connecticut | 98.0% | W | |||
| 11/13 | Fri | vs. | *West Virginia | 98.0% | W | |||
| 11/28 | Sat | vs. | Illinois | 99.9% | W | |||
| 12/5 | Sat | @ |
*Pittsburgh | 64.3% | W | |||
Favored: 12-0 Projected: 12-0 |
||||||||
Odds of going undefeated: 61.7% down 2.2% |
||||||||
Cincinnati Bearcats (7-0; BCS rank: #8)
Cincinnati has four games it should definitely win, and then a tough road game. Even if they win out, they might be left out after dropping to #8 in the BCS, but anything can happen.
The Syracuse and Illinois games are all but certain. The UConn and West Virginia home game odds took a big jump this week and are at near-certain status, too. West Virginia and UConn are both good teams who have played consistently above average football, but neither has had many great games. The upshot is that almost all of their performances rate below the Bearcats' performances, which have been stellar and very consistent, too.
The one game whose odds decline is the all-important showdown with Pitt. The Panthers blasted USF Saturday and the game is looking daunting. With the increased odds for the two other games, Cincy almost holds its percentage even, but it declines slightly from 63.9 to 61.7%. Cincinnati has to keep winning impressively during the next four-game stretch so it can be in position to stake a claim in the title game should lots of other teams fall.
How Many Teams Will Go Undefeated?
Last week the odds said that three (and almost four) teams would be expected to remain undefeated before the bowl games, and tallied individually, five teams had a greater than 50% chance of doing so. Now, even though every team won last week, only four can make that claim. But that matches the new expected number of undefeated teams. The odds of six teams going undefeated took a big jump, from 1.2% to 4.3%, and the odds of all seven teams suffering a loss fell from 0.4% to just 0.13%.
Who will the undefeated teams be?
- TCU: 88.0%
- Boise State 74.6%
- Florida/Alabama: 69.7%
- Cincinnati: 61.7%
- Texas: 45.4%
- Iowa: 33.6%
If this happens, we'll see the Florida/Alabama winner against TCU or Cincinnati. Both have their drawbacks: TCU is a non-BCS school and Cincinnati's schedule is soft. But there aren't many viable 1-loss teams to replace them.
We've ruled out USC above, and the next four in the current standings are LSU, Oregon, Georgia Tech, and Penn State. LSU is expected to lose to Alabama; Oregon could win out and still be behind Boise State, the team that beat them; likewise a 1-loss Penn State would be logically behind a 1-loss Iowa. Georgia Tech's remaining schedule is kind—perhaps too kind to get them to pass Cincy or TCU, even with an ACC title game win. In short, only USC or LSU seems in position to make a run at the title game, and both have a rough road to get there.
With the Florida/Alabama game a 50/50 proposition (Florida by 1 point by score margin, though) and TCU-Cincinnati an unknown (though we suggest TCU is in better position), we'll look at all four matchups, from most expected to least expected:
Florida 61.2% over TCU
Alabama 62.5% over TCU
Florida 59.2% over Cincinnati
Alabama 62.5% over Cincinnati
So the winner of the SEC title game is expected to win the national title. No big shocker there. Both of them lose to Texas, however, so they better hope that the Longhorns suffer that unspecified upset! If Texas beats Oklahoma State next week, they'll probably again be over 50% and then favored to win it all. Right now only cumulative odds are preventing us from putting Texas in the title game, and cumulative odds become less meaningful the fewer games a team has left.
LA Times article ignorant about computer ratings
Here we go again. The annual article showing a complete lack of understanding about computer rating systems. It's not so much the ignorance that bothers me, but that the authors never even attempt to try to gain an understanding. They just push forth a few examples that make computer rankings look stupid or irrational, and then say "wow, can't possibly make heads or tails of that! Why even try?"
The examples in the article make complete sense to me, mostly because my two rating systems mirror the examples almost exactly. The author (Chris Dufresne) exhibits dismay that Jeff Sagarin's two systems can evaluate the same team differently. Here are his examples:
He doesn't even understand how Sagarin combines his ratings. The real numbers he should have cited are:
Using my comparable ratings (Strength is similar to Predictor, and Success mirrors ELO somewhat):
So while the differences in my ratings aren't quite as extreme, they're pretty close on the whole. So why the discrepancies between two methods of rating college football teams? There are many reasons.
One is that it's early in the season. With only four or five games to go by, any one game can have a big effect on a score-based system. Likewise, using winning and losing only, five games is a paltry amount to base a rating on. This is why I (and Mr. Sagarin) combine the two systems. When combined, the deficiencies of the two systems are lessened. USC ends up #3, Florida #8, and Northwestern #27 in my Overall ratings. In Sagarin's method, USC is #1, Florida is #7, and Northwestern #28.
Also, early ratings are sometimes handled differently until there are enough "connections" to make valid comparisons. Jeff Sagarin uses some weighing of last year's results until this year has enough data. At this point, according to his site, Predictor (the score-margin system) still is using some weight from last year, while the ELO portion does not. This is naturally going to lead to greater discrepancy between the two.
For the SportsRatings power rating, there is never any use of last year's data. But early ratings are naturally weighed heavier with the results of Strength rather than Success. The Success ratings build throughout the year as teams win games. It's difficult for a Win-Loss-only system to have good results until teams have played about 10 games; until then, the score-based system carries more weight. My motto is "you can't argue with Success" but only at the end of the season. Until then, some results, like Oregon State's upset of USC, can be at least somewhat written off as a fluke.
Dufresne does point out that USC's huge wins over Ohio State and Oregon count for a lot in a margin-based system. From there it doesn't take a lot for the ratings to make perfect sense; why should a 3-1 team by any higher than #16 anyway?
Florida's case is easier to understand. If you only knew that some "Team X" had defeated four teams that were each 2-3, and Team X had lost to another that was 3-3, wouldn't #46 seem like a possible ranking? But knowing that the score of the four wins were 56-10, 26-3, 30-6, and 38-7, and that the loss was by 1 point, you might think it reasonable that they be in the top ten.
Just a little digging could help people understand the computer rankings, and it's an area that journalists have failed. There is so much coverage of every aspect of college football by the press—offenses, injuries, history, human-interest storylines—but the computer rankings are addresses with either pure ridicule and/or ignorance. There is no attempt to educate the fans on exactly what mechanisms they operate by. Why not? Sure, it may make for a longer, more in-depth article, but it has become a crucial area due to the BCS system. But all we get is "oh my God, look at what those computers did!"
Even for the financial meltdown in the stock market has articles that attempt to explain, step-by-step, just what happened in a technical sense. And that "system" is so complex than no one really understands how it works or where it is going since its scope is so huge. College football has a finite amount of games—it would be possible to show exactly what a computer is "thinking" every step of the way!
Sportswriters do a disservice to fans when they use the computer ratings only to take potshots at the BCS rather than to introduce fans to a very interesting nexus of sports and mathematics. The introduction of computer power ratings has clearly influenced sports and college football in particular. Spreads on games hew close to computer models these days. And pollsters, over the last ten years, have started to refer to power ratings when making their choices. In the early days of Sagarin's ratings being featured in USA Today, the difference between his ratings and the polls was huge. Then many of the teams he ranked high started to "upset" other teams. After years of this happening, you started to see these teams enter the top 25 earlier, and move up faster, than they did in the past.
Now, with spreads and the AP poll mirroring the computer rankings so much, any discrepancy is quickly noted. But does the AP poll make that much sense right now? #24 Pitt beat #19 South Florida who beat #16 Kansas; all teams have 1 loss. Michigan State is at #23 at 5-1, but the 4-1 Cal team that beat them isn't rated. There are justifiable reasons for these discrepancies perhaps. But the polls have another advantage: they are the aggregate of 70 different opinions. Combine 70 computer rankings and you wouldn't be able to pick one or two apart for having widely contrasting results.
The inconsistencies and biases that show up in the human polls are one reason I like computer power ratings. It's also why I base the SportsRatings Top 25—a non-computerized, subjective ranking—on a few basic principles so that teams are ranked in a fair manner. First and foremost, I try to explain why I put teams where I do, so that someone who takes issue with them at least knows where I'm coming from.
It's a lot easier just to mock a rating system than to try and understand it. But I think there is fan interest in knowing how these systems work, and journalists would be doing their job to go a step further in this direction.
October 06, 2008 at 10:39 PM in AP/Coaches/Harris poll, commentary, Power Ratings | Permalink | Comments (0) | TrackBack (0)
Digg This | Save to del.icio.us