Writers of Pro Football Prospectus 2008

Most Recent FO Features


» BackCAST 2018

The question is not whether Saquon Barkley is the best running back in this draft class. The question is whether any running back, even one as good as Barkley, warrants a top-five draft selection in the NFL in 2018.

08 Oct 2007

Confessions of a Football Junkie: Shock-Proof

by Russell Levine

Nothing that happens in this college football season should shock us. Not after Appalachian State won at Michigan. Not after Syracuse and its Pop Warner offense lit up Louisville. Not South Florida in the top five. Not a weekend with five of the top-10 teams in the AP poll losing -- four of them to unranked opponents.

Yet Saturday offered up a jaw-dropper big enough to make all those moments take a distant back seat.

Stanford 24, USC 23.

Stanford was 1-11 in 2006. It had lost three previous conference games this season by a combined score of 141-51. It had a backup quarterback, Tavita Pritchard, making his first college start. The game was at the Los Angeles Coliseum, where USC hadn't lost since the first year of George W. Bush's first term. That's USC, the team of the three regular-season losses the past four seasons. The Trojans were favored by anywhere from 38 to 41 points, making this possibly the largest point-spread upset in college football history.

This was Mike Tyson losing to Buster Douglas in Tokyo. This was the Jets beating the Colts in Super Bowl III. If it wasn't quite the equal of the Miracle on Ice, it's only because there were no geo-political implications. Even so, in football terms, the Berlin Wall just fell.

How so? Every one-loss team just regained a legitimate shot at the national title. Heck, if half the top 10 keeps losing each weekend, a two-loss team like Florida could be in play for a spot in the BCS championship. Another surprise: South Florida and Cincinnati probably only need a loss each by Cal and Ohio State to control their own destiny for a trip to New Orleans for the title game. Yes, you read that correctly.

For now, USC's stumble means that LSU is the consensus no. 1 team in the nation after the Tigers rallied to beat Florida Saturday night. Many will caution against assuming that LSU can run the table in the SEC, and if this season has taught us anything it is that assumptions are unwise. Still, the SEC is not what it was a year ago. Only LSU has looked dominant, and the Tigers have but two games remaining against opponents that are currently ranked: contests against Kentucky and Auburn the next two weeks before they presumably would face the East Division champion in the SEC title game. As of today, the favorite to take that spot might well be Tennessee, a team with two losses (including a 39-point stinker at Florida) that nonetheless controls its own destiny to get to Atlanta.

LSU was hardly overwhelming against Florida. The Gators ran the ball over, around, and through LSU's vaunted defense for much of the contest before untimely turnovers and LSU coach Les Miles's riverboat gambling swung the game in favor of the home side. His decision-making, which included going 5-for-5 on fourth downs in the contest, was akin to hitting on and endless series of 16s in blackjack -- without ever busting. He passed up a chip-shot field goal to tie in the final two minutes in favor of yet another fourth-down run. That's the kind of call that, should it go wrong, ends up getting the coach hanged in effigy around campus.

As if just to make things a little more difficult, Miles even opted to risk his final timeout, while trailing with 10 minutes to play, to challenge a call that netted his team a whole 10 yards of field position.

Strategically, LSU hardly benefited from USC's loss. But try telling that to a fan base that is still bitter over having to share a national title with the Trojans in 2003. When the final score of the USC game was announced in Baton Rouge, the crowd celebrated wildly, as did the LSU team on the sidelines, despite the fact they were trailing Florida at the time.

Assuming it can beat Kentucky, which is coming of its first loss of the season, next week, LSU will be atop the BCS standings when they are first issued on October 14. Ohio State, on the other hand, will surely be sending Stanford coach Jim Harbaugh a Christmas card if the Buckeyes are able to make it to New Orleans. Then again, Harbaugh is a Michigan man so salutations from Columbus are unlikely, but Stanford did Ohio State a huge favor nonetheless. In what was supposed to be a rebuilding year, the Buckeyes are 6-0 and should be heavy favorites in each of their remaining games.

Cal, which was idle Saturday, is also in prime position to reach the title game, but the Trojans' loss may have actually hurt the Bears. Cal wasn't going to reach the championship without beating USC November 10, but a win now will be somewhat less impressive to the voters and computers that determine the BCS standings.

Boston College, which has been flying under the radar all season, also now finds itself in position to begin thinking this could be a special year. The Eagles still have considerable hurdles, to wit: four of their final six games are on the road, including a trip to Virginia Tech, and that doesn't include a potential rematch with the Hokies in the ACC title game in Jacksonville.

Despite all the chaos over the season's first six weeks, there is much to be decided. There are still two more undefeated teams (11) then at this time last year. Remaining conference games will trim that number by at least four teams and based on the way this year has gone, odds are that several others will get upset. Not even Miles would hit that particular 16.

John L. Smith Trophy

The commenters in the SDA thread had this one pegged. This week's winner is none other than Florida coach Urban Meyer, for an inexplicable clock-management gaffe in the final minutes of his team's loss to LSU.

Once LSU converted its final fourth-down to gain a 1st-and-goal at the Florida five, Meyer should have begun calling time to preserve a chance to win in regulation should the Tigers either kick a field goal for the tie or score the go-ahead touchdown. The Gators still had all three timeouts, yet Meyer allowed 20-plus seconds to come off the clock before the first-down snap. After LSU ran the ball on second down, Meyer allowed another 29 seconds to drain away before calling timeout anyway.

If the strategy was to try and win the game in regulation right there and not take any chances with leaving time on the clock, fine. I don't agree with it, but once you've made that choice you have to stick with it. Instead, Meyer wasted nearly a full minute, before deciding, hey, maybe we'd like another shot with the ball after all.

Florida began its final possession with 1:09 to play and two timeouts. Do you think the Gators might have preferred to have two minutes-plus and one timeout?

Look, Meyer is a great coach. His results speak for themselves. But I'd be lying if I said I didn't enjoy giving him the award this week. He's smug, and at times over-tinkers in areas that don't need tinkering. And after his failed effort to call a last-second timeout and force Auburn to kick the game-wining field goal twice last week, perhaps there's karma involved in his JLS Trophy this week.

BlogPoll Ballot

This season, I'll again be voting in the BlogPoll, hosted by mgoblog. I'll post my ballot in Junkie each week. Feel free to comment, and I may adjust may rankings based on your suggestions.

Rank Team Delta
1 LSU 1
2 California 1
3 Ohio State 1
4 Boston College 3
5 South Florida --
6 Oregon 3
7 Cincinnati 3
8 West Virginia 4
9 South Carolina 5
10 Oklahoma 8
11 Missouri 11
12 Southern Cal 11
13 Illinois 11
14 Kansas 12
15 Virginia Tech 2
16 Arizona State --
17 Florida State 3
18 Kentucky 12
19 Wisconsin 11
20 Auburn 5
21 Florida 10
22 Hawaii 1
23 Texas A&M 3
24 Indiana 2
25 Virginia 1

Dropped Out: Kansas State (#13), Georgia (#15), Purdue (#19), Michigan State (#23).

Rankings that may require further explanation: Earlier in the season, I felt the top four were a cut above the rest. Now I limit that number to one: LSU. I view teams two-four as basically interchangeable, a shade better than the Big East teams.

West Virginia is behind Cincinnati in part because of Pat White's injury. I wanted to punish USC more for losing to Stanford, but I couldn't honestly say I thought the teams I had lower than the Trojans would beat them. I'm still withholding much judgment on Arizona State until they play somebody. Their rank will move drastically after they face Cal in three weeks.

I don't like having two-loss teams in the poll, and I'm the last guy that would want to over-respect the SEC, but again, I just feel that Florida and Auburn are both pretty good teams that would beat a lot of ranked squads. And at the bottom, the new entrants are a crapshoot. I'll give A&M, Indiana, and Virginia a chance to impress me.

Games I watched at least part of: Kentucky-South Carolina, Eastern Michigan-Michigan, Wisconsin-Illinois, Northwestern-Michigan State, Texas-Oklahoma, South Florida-Florida Atlantic, Cincinnati-Rutgers, Ohio State-Purdue, Florida-LSU, Nebraska-Missouri.

Portions of this article appeared in Monday's New York Sun.

Posted by: Russell Levine on 08 Oct 2007

77 comments, Last at 12 Oct 2007, 11:16am by oljb


by andrew (not verified) :: Mon, 10/08/2007 - 4:55pm

I notice just about everyone had BC if not Oklahoma leapfrog South Florida, as if the though of them in the top four would be ridiculous....

I don't know, they weren't that impressive over FAU, though.

by David (not verified) :: Mon, 10/08/2007 - 4:58pm

Not Western Michigan for the JLS Trophy? I'm still looking for the logic in taking a deliberate safety when up by seven points.

by pawnking (not verified) :: Mon, 10/08/2007 - 5:09pm

Could South Florida be making a Miami/Florida State type leap? Going from a homcoming sisters of the poor team to a national power in this decade?

Right now, I'm rooting for them.

by Dr. Kenneth Noisewater (not verified) :: Mon, 10/08/2007 - 5:15pm

Historically, the BCS computers have been very kind to Pac-10 squads. It's been the human polls that have underrated them. Cal's win over Oregon, and their upcoming wins over Arizona State and USC, should be good enough to fend off an unbeaten Ohio State for the #2 spot.

And if not, well, frankly, us Cal fans will be ecstatic with a Rose Bowl bid. It's been a long time.

by NewsToTom (not verified) :: Mon, 10/08/2007 - 5:15pm

I admit I haven't seen Virginia play lately, but I have a hard time believing the team that I saw look so pathetic against Wyoming could possibly be one of the top 25 teams in the country. I would suggest instead ranking UConn, which is at least undefeated against their admittedly more pathetic foes, or Colorado with two losses against top 25 teams and a top 10 win.

I also find it curious that you feel the need to defend ranking Cincinnati ahead of West Virginia, since the Bearcats are undefeated and the Mountaineers are not.

I feel a Chris Weinke at NCSU c. 1998-style game coming out of Matt Flynn sooner or later. I also don't trust Les Miles-his bad decisions against UF were validated by results, but that probably just means he'll remember the good results more than that they were at least questionable decisions.

by Charles Odell (not verified) :: Mon, 10/08/2007 - 5:29pm

Actually, given the way the season's been running, with high-ranking teams going down in flames against huge underdogs (Mich/App State, USC/Stanford, Oklahoma/Colorado, UCLA/Utah, West Virginia/Syracuse), I think LSU's gotta worry about that Ole Miss game...

by Chris Heinonen (not verified) :: Mon, 10/08/2007 - 5:34pm

Just so you know, USC losing is the largest point spread upset in college history, which I know only because Oregon State held the record, winning at University of Washington as a 35 point underdog in the early 80's (when OSU had a nice 4-40 or so stretch).

by Pat (not verified) :: Mon, 10/08/2007 - 5:35pm

A couple of weeks ago, I wrote a satire post pretending to be the 2007 NCAA Football season, basically saying, "hey, it's been too normal these past few years, I've decided to *&^!$ some *!#^( up."

I had no idea it would come true. I'm just waiting for the Duke-LSU BCS championship game.

Seriously, in a million years, would anyone have picked the best teams in the Big 10 this year to be Ohio State, Illinois, Wisconsin, Indiana?!?

Plus, I mean, South Florida in the top 10. If you had told someone 10 years ago that South Florida would in the top 10 in a decade, they never would've believed you. Guaranteed. Because USF didn't have a football team 10 years ago.

Michigan losing to Appalachian State should be the mascot for the 2007 season. It's too bad NCAA 2007 comes out so early, because someone from Appalachian State should be on it. That was just the entire season throwing up a big flag saying "Beware, college football world! Beyond this point, there be dragons!"

by Pat (not verified) :: Mon, 10/08/2007 - 5:47pm

Oh, and one other thing, given that it's local news here: Hawaii, this season, should show why David Lewin's projection system works, even though it's bizarre to think that someone goes from potential bust to potential star by sticking around another year in college. It's not just sticking around another year. It's performing in that year like you did previously. I'm not sure Brennan, for instance, is still a first/second round pick. He's now racking up an injury history, and his backup is having almost as much success.

by Fnor (not verified) :: Mon, 10/08/2007 - 5:47pm

Real upsets, huh? Maybe I should give college ball another chance....

by bradluen (not verified) :: Mon, 10/08/2007 - 5:50pm

If I'm an LSU fan, I'm very worried about Auburn. If we were ranking on how good the teams look right now, Auburn looks top ten or even top five, and I think this week they'll be the first team to stop Arkansas's running game. They do need Quentin Groves back to have a shot in Baton Rouge, though.

by bradluen (not verified) :: Mon, 10/08/2007 - 6:05pm

And I'll bet you dollars to donuts that Brennan still goes in the first two rounds.

by nath (not verified) :: Mon, 10/08/2007 - 6:08pm

So... Florida loses on the road by four to the #1 team in the country, and drops ten spots?

This is why I don't get college football rankings. They seem to have so little to do with which teams are actually better and everything to do with not losing, regardless of how cupcake your schedule is.

I mean, do you really think Wisconsin would beat Florida? I think they would get annihilated. I guess if you felt you HAD to drop Florida behind Auburn, this makes sense. But it just seems silly that any loss is an automatic 10-spot drop regardless of context.

by James G (not verified) :: Mon, 10/08/2007 - 6:25pm

Texas A&M?!!?!? 34-17 losers to Miami, and barely OT winners over Fresno State. I'd also be more likelly to give a rank to Wyoming, 23-3 winners over Virginia, and only loss is at Boise State, than to Virginia.

by lionsbob (not verified) :: Mon, 10/08/2007 - 6:55pm

Missouri vs. Cincinatti in the National championship game. The only way we can get a playoff system by 2008.

by citizen jason (not verified) :: Mon, 10/08/2007 - 6:56pm

I don't have a problem with witholding judgement on ASU until they beat someone, but the same thing could be said for a lot of teams in the list. Wisconsin, for one, hasn't beat anyone decent. (Although they have lost ...) And who has Cincinatti beaten? Rutgers? (I'm still not really sure how they were ever ranked ...)

Before any Cincy lovers flame me, I'm not complaining--I'm just saying that a lot of the teams on the list either: a) haven't played anyone; or b) beat teams that were ranked at the time but since then have been shown to be less than a quality opponent. (See, for example, anyone who made noise by beating Louisville ...)

by DMP (not verified) :: Mon, 10/08/2007 - 6:58pm

I am thankful for all the craziness going on only because it distracts you, Russell, from having the space or focus to write up some ILL snide remarks about MSU's loss to Northwestern at home. I know the local media already went to full assault (Drew Sharp has become my favorite Sunday morning read), and the fans seem to already be hitting the panic button. I think it's far too early. Although being all the way out in Seattle and not getting that game on TV here, I am glad I didn't have to watch it.

by The Boilermaster (not verified) :: Mon, 10/08/2007 - 7:11pm

17: Which is why colorado, with a win over OU and closer-than-the-score-indicates loses to ASU and FSU, should be ranked.

Close losses to good teams > blowout wins vs terrible teams.

by The Boilermaster (not verified) :: Mon, 10/08/2007 - 7:12pm

edit on 18. I meant to say (16:) not (17:).

by oljb (not verified) :: Mon, 10/08/2007 - 7:23pm

Charles Odell,
Your list of upsets is incorrect. West Virginia just beat Syracuse by 41 points, scoring more points than any opponent ever has in the Carrier Dome. You are probably thinking of Louisville.

But I agree that LSU might also get upset; it would be fitting with everything else that's happened this season, and I didn't see any reason to think that they are leaps and bounds ahead of the rest of the pack during the Florida game.

by bradluen (not verified) :: Mon, 10/08/2007 - 7:23pm

The most proven teams (multiple wins over quality opponents, doesn't take consistency into account):

1. LSU
2. South Florida
3. Auburn
4. Cal
5. South Carolina
6. Florida State
7. Illinois
8. Oklahoma

Unproven teams (at most one quality win, which might've been a fluke, or else the opposition might still turn out to suck): West Virginia, Virginia Tech, Florida, Hawaii, Kentucky, Kansas.

by navin (not verified) :: Mon, 10/08/2007 - 7:41pm

No one has learned any lessons from last year--you can't judge the relative quality of teams from difference conferences by watching them play in conference games. Two great teams will look much worse than they are when they play each other, just like two bad teams will look better than they are if they play each other.

So what does this mean?
1) BC is probably overrated by at least 10 spots in the polls.
2) You can't say the SEC is weaker than they were last year, you need to get more OOC results to determine that.
3) Forget that USF is called South Florida. If their name was West Virginia, they would be #2 right now.

by Narghile (not verified) :: Mon, 10/08/2007 - 7:42pm


Missouri has beaten Illinois and Nebraska...

by bradluen (not verified) :: Mon, 10/08/2007 - 7:50pm

One team that isn't as proven as its ranking: Boston College. Yeah, Matt Ryan is my hero, and they deserve to be ACC favorites, but the wins are Wake Forest (3-2), NC State (1-5), at Georgia Tech (3-3), Army (3-3), UMass (I-AA) and Bowling Green (3-2). If they win at Virginia Tech in three weeks, I'll resume gushing over them.

by bradluen (not verified) :: Mon, 10/08/2007 - 7:57pm

23: Missouri aren't far off my list, but there's a very large question mark hovering over Nebraska at the moment. (To be fair, Missouri are part of the reason that question mark is there.)

by kibbles (not verified) :: Mon, 10/08/2007 - 8:11pm

Out of curiousity, how do you do your rankings? Do you rank teams based on, if the championship were tomorrow, how deserving they would be to appear in it? Or do you rank based on the whole "if they played 10 games on a neutral field, who would win more?" angle? Because if it's the former, your rankings look fine, but if it's the latter, they look all screwy. Do you really think that Cincinnati would go 6-4 in ten games against USC, Florida, or Oklahoma on a neutral field?

by oljb (not verified) :: Mon, 10/08/2007 - 8:42pm

Other than brand name, what reason do you have to think that Cincinnati would not in fact win 60% of its games against USC, Florida and OU?

This "my team lost, but since it is more famous it would probably win most of the time" line is one of the most irritating excuses leveled whenever a team with a less impressive history beats a team that fancies itself a top-tier power.

There is no such thing as a series in football. Teams are not going to play each other 10 times. This is like the Grim Reaper telling Bill and Ted "best 6 out of 11".

by Kevin 11 (not verified) :: Mon, 10/08/2007 - 9:13pm

My Meaningless Top 25:

1. LSU
2. California
3. Ohio State
4. South Florida
5. Missouri
6. Boston College
7. Arizona State
8. South Carolina
9. Oregon
10. West Virginia
11. Virginia Tech
12. Auburn
13. Florida
14. Oklahoma
15. Illinois
16. Southern Cal
17. Cincinnati
18. Kentucky
19. Kansas
20. Wisconsin
21. Florida State
22. Purdue
23. Tennessee
24. Kansas State
25. Michigan

**Missouri makes a quantum leap thanks both to it’s slaughter of Nebraska, and the fact that it’s win vs Illinois is looking better every week.

**Has LSU lost lat night, Cal would be a legitimate #1. I had them #2 last week, and said they had a case for the #1 spot.

13- I agree. If anything, Florida proved how GOOD they were on Saturday night, barely losing on the road to the unanimous #1.

22, Point 3: I'm not sure if they'd deserve to be #2, but if their name was Ohio State or UCLA, no one would be talking about a one loss team possibly getting a shot at the National Championship ahead of USF. Cal has the same problem, albeit to a lesser extent.

22, Point 1: There are really 13 teams you'd rank ahead of undefeated Boston College?

by hooper (not verified) :: Mon, 10/08/2007 - 10:12pm

Re: 3

Understand that I say this with no substantiation, but just a fear about it all. As much as I would also like to see the USF story continue, I'm really worried about finding out later on that they're really the third-best pro team in Florida this year. USF is a good place for a new program to suddenly flourish into a contender, but the deep history of recruiting problems in Florida and a rise-from-nowhere event will likely get some looks. I'm not saying that there actually is anything suspicious, but I do fear the possibility. I hope not.

I'm torn on Virginia. On one hand, they really laid an egg against Wyoming (7220 feet had nothing to do with it). On the other hand, it makes Wyoming look all that much better (my alma). Given that they're 4-1 now, and assuming they keep winning, how many more wins would have to occur before they start showing up in the rankings? Just curious - alma, like I said earlier.

Actually, on Wyoming: next year, they play in Knoxville for UT's homecoming. Given that UW is improving and UT seems a little spotty (and will lose Ainge after this year), that game might be more than UT expected. I'll still be attending school at UT then, so it should be fun times. Maybe I'll make one of those split jersey things - UT creamsicle and UW mustard-n-chocolate. Yummy. :)

by Pat (not verified) :: Mon, 10/08/2007 - 11:18pm

#12: If Brennan has another 5 INT day next week, and misses more time struggling with his ankle, I'll take that bet. Nothing drops a quarterback's stock faster than injury concerns, and Brennan already had the "is he a system QB?" worry before Graunke came out and played pretty much just as good.

by mm (not verified) :: Mon, 10/08/2007 - 11:30pm

I agree with those who say you can't justify having LSU #1 and rank Florida #21. If Florida didn't have LSU on their schedule they'd only have 1 loss right now and they'd be in your top 10. You punish them for losing on the road by one play to your #1????? This is the sort of silliness we expect from the AP poll, not from Football Outsiders. If anything, their play this weekend justifies moving them up, not down.

by Pat F. (not verified) :: Tue, 10/09/2007 - 12:05am

ESPN.com's front page news blurb on the Trojans' QB's injury reads "USC's Booty Hurting." Hee hee!

by Roy (not verified) :: Tue, 10/09/2007 - 12:12am

as others have argued, i see a lot of inconsistency here. i would also add kansas jumping up to 14 after - and presumably for - beating a team that plummets out of the rankings. this is just like the AP poll, which we already have.

by mmm... sacrilicious (not verified) :: Tue, 10/09/2007 - 12:22am

#2: It wasn't a bad move... it took a kickoff return at the end of the game to beat them. I'd say there is a much greater chance of something bad happening punting out of there.

by Chad Gerson (not verified) :: Tue, 10/09/2007 - 1:06am

As a Hoosiers fan, this has been a long time coming. Not BCS-bound, but we'll be in a respectable bowl, and there have been years I thought I might never say that again. We can build on this.

by Chad Gerson (not verified) :: Tue, 10/09/2007 - 1:12am

News headline on ESPN.com: "USC's Booty hurting."

by Pat (not verified) :: Tue, 10/09/2007 - 1:15am

This is the sort of silliness we expect from the AP poll, not from Football Outsiders

You can't do a poll this year with logic. It just won't work. The season doesn't make sense at all.

For what it's worth, the computer rankings typically have Wisconsin above Florida as well, too, so it's not that silly or illogical. Two losses just anchors your position relative to others a lot more than one loss does.

Colley has Florida at 19. Sagarin has Florida at 21. Just considering who-beat-who, Florida in the high teens/low twenties isn't that crazy. When you take into account their losses were close, okay, maybe they should be higher, but it's perfectly within reason to disagree on that point.

by Travis (not verified) :: Tue, 10/09/2007 - 2:08am


Can you explain the mathematics behind Colley's new inclusion of I-AA teams as groupings? It seems to lead to weird results, such as "I-AA Group 2" as the number 27 team in the country, and a higher-than-expected ranking for Michigan.

by Roy (not verified) :: Tue, 10/09/2007 - 2:27am

actually, pat, sagarin has florida #9 and wisconsin #32. you need to look at sagarin's ratings at usatoday to see this. if you are looking at the ratings in some kind of BCS table of computer ratings, you're seeing the BCS manipulation of his rankings. sagarin does a "politically correct" ranking that only counts wins and losses - this is what the BCS uses, and it seems to me obviously something anyone on this site would disregard. he also does a ranking that disregards wins and losses. the final sort at the usa today site is a compromise between the 2. noting the distinction here seems crucial b/c the criticism of this FO ranking for being just like the AP poll is that it's mainly just sorting by wins and losses.

by Fourth (not verified) :: Tue, 10/09/2007 - 3:26am

I hope sticking it to Florida felt nice for you. LSU keeps hitting on 16 and winning...your words...or else Florida would be what? 5th? While we're arbitrarily slamming other schools and coaches (every coach does the timeout before the field goal now, Buffalo just used it against Dallas to no avail), I'm looking forward to when Michigan overpays Miles to be Carr minus the class.

by bradluen (not verified) :: Tue, 10/09/2007 - 4:31am

re Colley: By some fluke five of the season's six I-AA over I-A wins ended up in group 2. This is daft (since group 2 is supposed to be the second best group of I-A teams) but the makeup of each group is pretty unstable, so this effect won't last.

by Pat (not verified) :: Tue, 10/09/2007 - 5:01am

#39: I've explained this here before, but let me summarize quickly. While Sagarin's little "politically correct" dig makes it sound like the big bad BCS forced something that all the smart people didn't want, the truth is that the BCS told the statistical rankings to remove margin-of-victory based on several statisticians' recommendations. The basic reason is this: there's no unbiasable game output function other than the binary "won/loss," and since the other measures are biasable (and, indeed, have conference biases) why bother using them, since you just end up with the same problem the human polls have?

Moreover, Sagarin's own "politically correct" mini-criticism doesn't make any sense. An Elo ranking system isn't "politically correct." It's simple, unbiasable and is used in bajillions of other sports. Sagarin in fact uses Elo plus the points-based in a combination, thus implying that he's not nearly as confident of the "Predictor" ranking as he's claiming. Averaging them, like he does, makes no statistical sense whatsoever (he claims they're 'diametrically opposed', which is completely wrong).

So when I say "Sagarin's rankings" I usually just mean the ELO-CHESS ranking. Why? Because I know what it's doing, and it's at least unbiasable. "Predictor" has obvious biases (unbalanced conferences - with only a few good teams, and a few really bad teams - tend to 'push up' the best teams higher than they should, and 'push down' the worst teams lower than they are) and the combined ranking makes no sense.

So please, don't call the ELO rankings a 'manipulation.' They've been around a long time. They're completely solid statistically - it's just a maximum likelihood technique. The margin-of-victory based techniques rely on using a single game output function across all of Division IA, which makes little sense.

by Pat (not verified) :: Tue, 10/09/2007 - 5:25am

#38: No, honestly, I can't. I was planning on sending an email to him about that. It doesn't make any sense to me at all.

Basically, the idea seems to be rank all of Division IAA. Then chunk that entire division up into large groups of teams to try to find an entire subset that has a decent connection to Division IA. Then use those connections to roughly align the Division IAA groups inside Division IA.

The problem with this is simple: he seems to be ignoring the original ranking of Division IAA teams! So you end up with Group 2, which is 5-1 against Division IA opponents, but presumedly significantly worse (by a statistically larger sample!) against Group 1, which is 0-7 versus Division IA opponents!

This makes no sense. If the season finishes out with the IAA groups in order in the IA rankings, he'll be really lucky. But he needs to alter that algorithm somehow to take into account the relative strength of each IAA group as known from the IAA rankings.

I'm not sure it's easily solvable in the way he's doing the rankings, though, which is strictly solving the linear system of equations. I could do it in a maximum-likelihood formulation relatively easily.

It might settle out, but I'm really not sure. Fundamentally he's flat out ignoring a huge set of data in favor of a much smaller set. That's always dangerous.

by Pat (not verified) :: Tue, 10/09/2007 - 5:37am

I should say, the one simple suggestion I would say is that, at the very least, he needs to change it such that the groupings have played each other. So "Group 1" should have a victory over "Group 2," etc. This ignores the "closeness" of the groups, but it's better than ignoring everything.

by James G (not verified) :: Tue, 10/09/2007 - 9:02am

39- You should note that right now, both Sagarin Predictor and Sagarin overall are still using the original ranking element, and only the ELO ranking is from this year only. I won't even pretend to understand what Pat said, however. What I do know is that a journal article was published last year in Naval Research by Sokol from Georgia Tech suggesting that linear regressed markov chains made better predictions of the NCAA basketball tournament when they used MOV.

by Russell Levine :: Tue, 10/09/2007 - 12:26pm

My general philosophy when it comes to these rankings is to not pay too much attention to where I had them ranked last week. The Florida thing is always tough. Here you have a team that just outplayed the #1 team in the nation for most of the game ... but they lost. To me, that makes me wonder if they were good enough to play like that against LSU, how could they get beat at home by Auburn?

So even though it looks like I'm punishing them big for losing to LSU, I prefer to think of it is I'm upping the punishment I gave them for Auburn.

Playing close against top teams is great, but you have to win. Florida is a two-loss team right now, one with defensive issues and a very limited offensive playbook. I think their ranking is justified.

As for the commenter that thinks I love Miles, I can assure you I do not. The first part of this article was written for the New York Sun, where I try to hide my Michigan bias. If Miles is the next Michigan coach, I'll learn to deal with it. Right now, I'm not crazy about the idea. I think some of decision-making is highly questionable (although not most of the 4th down calls vs. Florida) and he's a little too loose-lipped for my liking.

by Fourth (not verified) :: Tue, 10/09/2007 - 5:27pm

About the bias thing, I think it wasn't your Michigan bias but your Big Ten bias that came through when you attributed Florida's performance in that game to a lackluster LSU defense. You had already decided Florida wasn't a good team before that game, so therefore LSU can't be all that good either. I'm just surprised anyone could miss the quality of the players on the field, and the way they were executing (with a couple exceptions of course). Most other websites and blogs I read, SEC-centric or not, are calling that the game of the year in college football (see Pat Forde's new column, see Brian at MGO salivating over Les Miles for 2,000 words, etc).

Obviously LSU is the best team in the country, even though they aren't perfect. Florida is a young team with many flaws--like a thin playbook, like a light defensive line, like a raw secondary. But that young team showed some serious gut and grit, and flashes of the offensive brilliance that is to come in the next couple of years. I'm obviously very much a homer, but I was damn proud of that performance on Saturday by my alma-mater and I thought the way you characterized it was inconsistant with what most other people saw. Why I care so much about one blogpoll ballot, I'm not really sure, I suppose it's just a combination of my hatred of the current national media voting and the quality I've come to expect from this site. I guess I should start my own poll huh? :)

by bradluen (not verified) :: Tue, 10/09/2007 - 5:28pm

Pat - any reference for why unbalanced conferences biases Sagarin's predictor? It doesn't follow intuitively for me.

Agree that Colley's I-AA method throws out way too much data.

by James G (not verified) :: Tue, 10/09/2007 - 5:44pm

I think Florida probably shouldn't have dropped from a loss to LSU. They should have already been dropped due to very narrowly beating Mississippi followed by losing to Auburn. And a close loss shouldn't send them screaming back to the top.

by James G (not verified) :: Tue, 10/09/2007 - 5:57pm

bradluen - see if you can follow the link in my name to Andrew Dolphin's ratings explanation, which explains why the score rating must be biased. I don't entirely follow, but I think I get the gist of why bias exists.

What I don't understand is why that bias is bad. This, I think, is Pat's contention.

by bradluen (not verified) :: Tue, 10/09/2007 - 6:10pm

Thanks James.

by kibbles (not verified) :: Tue, 10/09/2007 - 6:38pm

Re #27: kibbles-
Other than brand name, what reason do you have to think that Cincinnati would not in fact win 60% of its games against USC, Florida and OU?

This “my team lost, but since it is more famous it would probably win most of the time� line is one of the most irritating excuses leveled whenever a team with a less impressive history beats a team that fancies itself a top-tier power.

Unless Sagarin's rankings contain a "name recognition" variable, I think there are lots of perfectly logical, unbiased reasons to believe that USC, Oklahoma, or Florida would beat the stuffing out of Cincinnati, none of which relate to a team's recent success or national profile.

Re #42: So when I say “Sagarin’s rankings� I usually just mean the ELO-CHESS ranking. Why? Because I know what it’s doing, and it’s at least unbiasable. “Predictor� has obvious biases (unbalanced conferences - with only a few good teams, and a few really bad teams - tend to ‘push up’ the best teams higher than they should, and ‘push down’ the worst teams lower than they are) and the combined ranking makes no sense.

Discussion of bias aside, which set of rankings most accurately predicts future results, the Elo-Chess, the "predictor", or the combined? Bias and consistency are nice and all, but I'm far more concerned with results. My impression was always that "predictor" most accurately predicted future results, which would make it the ranking that I care the most about. I mean, hell, if you want to rank teams without bias, why not go straight win/loss and put UConn at the top of the rankings? That would certainly be a very unbiased way of rating teams, but would it be a GOOD way? I certainly don't think so.

by Pat (not verified) :: Tue, 10/09/2007 - 6:56pm

#45: Margin of victory in basketball is much better defined. Scores in football are typically "3-2" (i.e. 21-14) as opposed to "78-56" (or "39-28" if you want to think of scores as quantized in 2s). The NCAA basketball tournament is also a much more uniform field than "all of NCAA Division IA football."

Margin of victory just doesn't do that much in football, since the cases where it's well defined (blowouts) are the cases where it's least important (you typically already knew that one team was good and another bad).

by James G (not verified) :: Tue, 10/09/2007 - 6:59pm

Pat - I don't understand how MOV can't have meaning in football when pythagorean wins was shown to have the same effect in the NFL as in MLB (i.e. it is a better predictor of wins for the next season than straight W-L).

by bradluen (not verified) :: Tue, 10/09/2007 - 7:20pm

What I get from the Dolphin page linked in my name:

Using scores for the purpose of ranking achievement (as opposed to prediction) is bad for any number of reasons; for one, it's unfair to teams that "may have a style that encourages close wins, but nevertheless win most of their games", e.g. Tressel ballers.

However, using scores to determine strength of schedule is useful, because there are few nonconf games, and ranking teams solely on won-loss means differences between conferences are too big compared to differences within conferences.

I largely agree, but worry that conferences that bother to play D might then be underranked. Of course it's not clear if any such conferences exist this year.

by Pat (not verified) :: Tue, 10/09/2007 - 7:28pm

Pat - any reference for why unbalanced conferences biases Sagarin’s predictor? It doesn’t follow intuitively for me.

A conference that's full only of blowouts and close games contains no anchor points for the "distance" between the 'haves' and the 'have nots.' Since it's easier to explain blowouts by having the blown out team be very bad, and the victorious team be very good, it tends to shove the two ends of the conference apart.

Without margin of victory, the precision of each ranking will be worse, but the bias will also be gone.

Bias and consistency are nice and all, but I’m far more concerned with results.

I don't think you understand what I mean by "bias." Bias, in this case, means that not all the teams in the league are treated equally. Even if you're more accurate at predicting games on average, you can get that by just improving your predictions for a subset of the league significantly, and not at all (or even harming!) for another subset.

Bias is the worst thing that can happen in a sports ranking that's used for a playoff! In this case, you're saying "team A isn't as good as team B, because on average, teams that perform like team A aren't as good as teams that perform like team B" even though there may be a subset of teams like team A that are better (or as good) as team B.

That would certainly be a very unbiased way of rating teams, but would it be a GOOD way? I certainly don’t think so.

Congratulations, welcome to the 1700s (or 1960s) and dealing with developing a chess ranking. You're asking the exact same question they asked. Thankfully, we can use the exact same mechanism they did.

Developing a mechanism for creating a good way to rank teams based only on wins/losses has been studied for a long time. It works. It's fine. It's not amazingly precise in the absence of fractional-win information, but there isn't any unbiasable fractional-win information anyway.

My impression was always that “predictor� most accurately predicted future results

I don't believe the polls should reflect a "prediction of how the teams will perform." They are an evaluation of how the teams have performed. It's easy to say "well, team X is more likely to win in the future than team Y - team Y only won because they recovered 10 out of 10 fumbles, and we know they're random, so they're unlikely to continue that pace".

But in some sense, that's silly. Imagine if team A, ranked 3, beats team B, ranked 2, but the margin of victory wasn't big enough to push team B ahead of team A. What's the ranking saying? "Well, if team A played team B again, team A would likely win more often." That's a crazy thing to say : team B did win - regardless of whether or not it would happen in some hypothetical rematch, it did happen in reality.

The same situation could happen without margin of victory, but in that case, you're making a different statement: you're saying that team B's schedule (and results) were not as impressive as team A's, even taking into account team B's victory over team A. That valuation is much less fraught with craziness. (You can get less direct examples, too, imagining if Texas lost in 2005 and the question of PSU/OSU/Texas).

Using margin of victory is fine if your ranking isn't used to evaluate the season (i.e. not in the BCS). Using it in the season is just flat wrong. The season is not about who "would" win. It's about who "did" win.

by James G (not verified) :: Tue, 10/09/2007 - 7:36pm

Pat, thanks for the explanation in 56. I do agree that if we are using a system to make a playoff (as we are in the BCS) that what did happen (or retrodictive rankings) is a better measure than predictive rankings.

by Pat (not verified) :: Tue, 10/09/2007 - 7:37pm

Pat - I don’t understand how MOV can’t have meaning in football when pythagorean wins was shown to have the same effect in the NFL as in MLB (i.e. it is a better predictor of wins for the next season than straight W-L).

First, there's a difference between the NFL and college football. The NFL is a mostly uniform strength league. Division IA is not.

Second, it's not that it doesn't have meaning. It's the degree of meaning. Between two closely ranked teams, their relative margins of victory aren't going to be separate them significantly enough that it's worth using. It may improve the predictivity, especially on average, but that's not worth the cost in bias and the fundamental problem of switching from determining who played the best and who is the best.

by James G (not verified) :: Tue, 10/09/2007 - 7:38pm

55 - that sounds like a pretty good summary. I think the lower score reason (i.e. conferences that bother to play defense) is the reason I suspected that something like PYTH W-L% was better than straight up MOV. Not to mention, even if we use MOV, it should normalized by the # of drives, since that's the unit that you have to score in.

by bradluen (not verified) :: Tue, 10/09/2007 - 7:51pm

Bias, in this case, means that not all the teams in the league are treated equally.

Under that definition, then technically, any rating system that isn't drawing names out of a hat is biased. Even with a system that only used W-L, 2006 Boise State could've been the greatest team ever, and the computers still wouldn't have agreed with putting them in the national title game.

It seems this kind of bias is mitigated if you only use W-L, but I'd have to think about that.

by bradluen (not verified) :: Tue, 10/09/2007 - 8:23pm

OK, I think I get it now.

1. If the probability of winning is a known function of the difference in rating between the teams, then an ELO system based on W-L is unbiased.

2. If the distribution of MOV (or whatever function of scores you care to use) is a known function of the difference in rating between the teams, then an ELO-type system based on MOV is unbiased.

In soccer, neither condition holds perfectly, but the second is about as close to being true as the first. Hence the World Football ELO ratings (linked in name), which take MOV into account, work well.

In college football, the first condition doesn't hold, and the second condition really, really doesn't hold.

by Bart (not verified) :: Tue, 10/09/2007 - 8:46pm

As a Cal fan who's watched all their games this year, I'm just not comfortable with Cal in the top 5. I'm glad to see them there, of course, but I've been burned too many times by Cal underachieving to fully embrace it. This year, the offense has been somewhat inconsistent and the defense has been decent with a knack for a big play or two in each game. As much as it pains me to say it, I don't think they'd match up well against LSU. Hopefully the defense will improve once some injuries heal; they've been missing a couple starters on the D-line for the last few games.

by kibbles (not verified) :: Tue, 10/09/2007 - 11:09pm

Re #56: Congratulations, welcome to the 1700s (or 1960s) and dealing with developing a chess ranking. You’re asking the exact same question they asked. Thankfully, we can use the exact same mechanism they did.

And you really think that the model which best ranks chess players will also best rank football teams? I'm skeptical for any number of reasons, but one point I'd like to raise is maybe chess players are ranked the way they are because there is no such thing as margin of victory in chess.

Re #56: I don’t believe the polls should reflect a “prediction of how the teams will perform.� They are an evaluation of how the teams have performed.
That's why I asked Russell whether his rankings were supposed to be based on what teams had done to this point, or based on the "who would win 6 of 10 on a neutral field" hypothetical. It's an important distinction to make.

by Pat (not verified) :: Tue, 10/09/2007 - 11:43pm

And you really think that the model which best ranks chess players will also best rank football teams?

Yes. It has nothing to do with chess. You can use it for baseball, basketball, soccer, lacrosse... heck, you could use it for people playing FPSs or some other random thing, I dunno.

The entire idea is based on two things: you have a game, between two players, and one of the two is the victor (or a draw). A stronger player typically beats a weaker player - this can be strict (in which case you end up with a beatpaths-type ranking) or flexible, using a logistic function (like Elo/most maximum likelihood rankings use), where it is assumed that a strong player beats a weak player with some frequency X, and an even stronger player beats a weak player with some frequency Y greater than X.

It has nothing to do with the game itself. It's a method for finding an ordered chain of rankings from a set of pairwise comparisons.

players are ranked the way they are because there is no such thing as margin of victory in chess.

There's not really such a thing as margin of victory in football, either.

You're misinterpreting the basis of using margin of victory in football. The only reason using MOV in football improves predictivity is because it's presumed that beating a team by 14 points means that the team is more likely to win again than if they beat the team by 7 points. That assumption is extremely important! Without that being true, the score margin is totally immaterial: imagine the National Coin Flip League, where games are decided by 10 coin flips. In this case, margin of victory would have no predictive strength, and no one would use it.

The fact that MOV has predictive capabilities means that the assumption has some validity. The problem is that this isn't true in general, and I can prove it: there is no difference in terms of likelihood of future victory between winning by 1 and winning by 3. In addition, you can also look at overtime games: the margin of victory in a game which includes overtime is similarly non-predictive.

But it is true on average (which is why it works on average) but this is a problem when you're judging, rather than predicting, because misjudging a team is the worst thing you can do in a ranking used for playoffs (whereas mispredicting a team just means you'll be wrong). No unbiasable metric in football exists to give a measure of how likely a victory was.

You could imagine weird games where there was a true margin of victory - imagine if teams didn't get a win for being ahead at the end, they simply got "(points scored-points allowed)/(total points)" 'victory points'. In this case you would have to use margin of victory.

Not to mention, even if we use MOV, it should normalized by the # of drives, since that’s the unit that you have to score in.

And there's another reason why margin of victory is biasable: a 7 point victory in an 8-drive game is not more likely than a 14 point victory in a 16-drive game.

So now, it's pace dependent: which means you have to normalize by the number of drives. But now that becomes a problem too, because now you're dividing a small number (in terms of touchdowns, usually about 3-4), heavily quantized, by another small number, also quantized (drives, typically 10-15 or so). This means that your measure of margin of victory is now extremely imprecise (probably +/-40% or more!).

I guess the whole argument against MOV in BCS ranking can be summed up like this: "there is a large difference between saying that an average victory by 14 points is stronger than an average victory by 7 and saying that in a single game, a win by 14 is better than a win by 7."

by Pat (not verified) :: Tue, 10/09/2007 - 11:57pm

Oh, also #60: Statistical rankings go 'boom' with undefeated teams. From a maximum likelihood perspective, the likelihood only increases as you increase the ranking of an unbeaten team (since each of their victories become more and more likely). Therefore, if you look at the ranking probability distribution for an unbeaten team, it looks nothing like it does for teams with one loss. The latter ones look like Gaussians, or something like a Gaussian. The unbeaten teams look like a function which starts at some minimum (basically right above the best team they beat) and rises until it flattens out near the maximum ranking.

Figuring out how to convert the probability distribution for a beaten team into a single number is easy - you take the peak (or average, both of which are likely the same).

Figuring out how to do that for unbeatens is much different - each of the ranking systems do things differently, and frequently they claim "this is the best way to do it" blah blah blah. In my opinion, it's all hooey, and I don't believe that statistically it's anywhere near valid. The error distribution for unbeaten teams is just different. It isn't symmetric, it isn't characterized like beaten teams. You shouldn't list unbeaten teams the same way you do beaten teams. It's just that simple.

I've always been in favor of statistical ranking systems listing the unbeaten teams with exactly the same ranking.

by bradluen (not verified) :: Wed, 10/10/2007 - 1:14am

Ranking unbeatens isn't all hooey. Any ranking system will and should put LSU ahead of Hawaii at the moment, since (I presume) the nth quantile of LSU's distribution will be higher than the nth quantile of Hawaii's distribution for all n.

Now, is there a rigorous way of determining, based on results alone, whether Hawaii should be 5th or 45th? Maybe, but I've yet to be convinced by anything I've seen.

by Alex (not verified) :: Wed, 10/10/2007 - 2:33am

Oh, also #60: Statistical rankings go ‘boom’ with undefeated teams. From a maximum likelihood perspective, the likelihood only increases as you increase the ranking of an unbeaten team (since each of their victories become more and more likely).

I read a post on the PFR blog that deals with this problem and describes an interesting solution: Just introduce a dummy team, and give every team 1 win and 1 loss to that dummy team. The rankings that result from that seem much more sensible, although I'm not really sure how statistically rigorous it is. Still, there's a link to a paper that supposedly explains it, so if you're really curious, you might want to check it out.

by bradluen (not verified) :: Wed, 10/10/2007 - 3:27am

67: Thanks for the link. Colley does something similar (as does the IMDB for their top movie lists). It's elegant, but it's still as arbitrary as other ways of setting a prior: why not two wins and two losses? Or half a win and half a loss?

It doesn't solve the essential problem:

Sagarin ELO: Hawaii #54, highest opponent #114
Colley: Hawaii #30, highest opponent #92
Mease: Hawaii #21, highest opponent #88

Hawaii shouldn't be at the very top, and they should be significantly above UNLV, but it's very hard to work out where in between they should fall in these "the team's true rating is..." systems.

by Pat (not verified) :: Wed, 10/10/2007 - 3:40am

#66: It's more awkward than you think. The shape of the distribution - since it's fundamentally lopsided, unless you introduce some new assumption like "no team is infinitely good", the problem is that the likelihood of Hawaii being much better than LSU is extremely significant! For a normal team, with losses, their constrained distribution implies that while there's a chance that, say, #5 might be better than #4, the chance that #5 is much better than #4 is small, and the chance that #5 is muuuch better than #4 is vanishingly small.

Now, that being said, the rankings should be more about how impressive your season was (see previously) than the chance that you are better than another team (predictivity). So ranking unbeatens against each other is easy - you just take the lower limit.

Ranking unbeatens vs. beatens is really what I meant as being 'hooey.'

#67: It's an interesting idea, but basically what it does is essentially say "no team is strictly better than every other team in the country." It also essentially artificially closes the gap between the unbeaten team and the beaten teams.

The odd distribution of the unbeatens is telling you something. It's telling you that you have no idea what the upper limit on how good this team is. In some sense, you can look at the NFL this season - the Patriots look off-the-charts good. If you stuck in an artificial "dummy" team, it would pull them back to the rest of the pack. But how exactly do you know they're not off-the-charts good?

by Alex (not verified) :: Wed, 10/10/2007 - 12:39pm

The odd distribution of the unbeatens is telling you something. It’s telling you that you have no idea what the upper limit on how good this team is.

True, but you could look at it this way:

It's almost unheard of for a team to have more than two or three consecutive undefeated seasons. I would say that this is evidence that, over a sufficiently long season (say, 50 games), every team would lose at least one game somewhere along the line. So, assuming that the team isn't totally unbeatable isn't that unwarranted. In fact, instead of introducing a dummy team, you could simply use the previous season's games, but weighted so that the entire previous season only counts as a fraction of a game. That way, the only teams that the ranking system would have to put at the top, with infinite rankings, would be teams that were undefeated since the beginning of last season. And such teams probably should be at the top of the rankings anyway, so problem solved. Treating teams as multi-season entities should be enough to solve the problem of where to rank undefeated teams.

I know what you're thinking, and you're right. In addition to the arbitrary weighting of the games from the previous season, the fact that roster turnover significantly affects the team's true strength from season to season is a problem here. But still, there is some roster turnover within each season due to injuries, benchings, etc. So, you could look at it as (the seniors graduating at the end of one season) = (the seniors all had season ending injuries at the same time). Not the most elegant solution, obviously, but it seems like an approach that could make things work.

In some sense, you can look at the NFL this season - the Patriots look off-the-charts good. If you stuck in an artificial “dummy� team, it would pull them back to the rest of the pack. But how exactly do you know they’re not off-the-charts good?

Well, they weren't undefeated last season, and the team is mostly made up of the same people now as it was then, so I'd say that the hypothesis that they are still a beatable team is much more plausible than the hypothesis that the players added to the team in the draft/free agency took the team from good to off-the-charts incredible. It's not ironclad proof, but I think it's enough.

by Trogdor (not verified) :: Wed, 10/10/2007 - 5:34pm

"You can use it for baseball, basketball, soccer, lacrosse… heck, you could use it for people playing FPSs or some other random thing, I dunno."

I would imagine some FPS rankings are ELO-based. I know that many RTS and RPG games use ELO for various rankings. Warcraft 3 was the first one I played that used it, and it was an incredibly useful system. When you have random matchmaking and can play 3-5 games an hour, it doesn't take long to develop a good idea of someone's true "skill level". And ELO seems to work better as the sample size increases, so a video game where you can rack up hundreds of data points per player is as good of an environment for ELO as any.

So I guess I have to ask - how good is ELO for college football? You have 119 teams, each playing 12 games, and about 2/3 of those are conference games. So you have small groups playing most of their games within the group (so hierarchy within the group should be pretty clear, although the ELO won't differ much from straight record), and only a handful of inter-group games to sort everything else out. Is that enough? With 119 teams, how many games do you need for the ELO to really work? Is 1 game per 10 competitors enough?

by Pat (not verified) :: Wed, 10/10/2007 - 9:32pm

#70: The team could be beatable and still be significantly farther away from the rest of the pack. The dummy team ties the entire league, in some sense, by saying "no one's that bad that they can't beat dummyteam" and "no one's that good that they can't lose to dummyteam". While the assumption that a team can't be infinitely good might be warranted, inserting a dummy team does more than that.

how many games do you need for the ELO to really work? Is 1 game per 10 competitors enough?

It's not very good at all. Calling all maximum likelihood rankings "ELO" is a bit odd - the idea is essentially the same, the implementation is much different. Elo's original ranking for chess was designed to be kindof "keeping score" in the chess world - it didn't matter that it converged slowly. But in general, even for a full ML ranking, you could shove teams around two or three spots and the total overall likelihood wouldn't change noticeably.

That's a strong reason why the BCS reduced the strength of the statistical rankings, incidentally. They're just not precise enough - there's simply not enough information.

by Pat (not verified) :: Wed, 10/10/2007 - 9:34pm

I guess my point between ML and an Elo ranking is that Elo is designed to not converge until many games have been played (if you're 'keeping score' you don't want someone to jump from last to first to middle), but maximum likelihood is designed to converge as fast as the information allows. Still takes a while, and the precision isn't great.

by Alex (not verified) :: Wed, 10/10/2007 - 11:25pm

While the assumption that a team can’t be infinitely good might be warranted, inserting a dummy team does more than that.

Right, that's why I'd just use the team's previous season instead of a dummy team. That way it's at least relevant to that particular team in some way, and it allows for teams to be way ahead of the pack if they were very good the year before.

by oljb (not verified) :: Thu, 10/11/2007 - 1:00am

I don't know if anyone just watched the Navy-Pitt game, but Wannstadt is a moron. 2nd OT, Navy had hit a field goal on its possession. Pitt has it with a 4th and goal at the two yard line. Bostick tosses a fade, incomplete. Game over.

I'm not convinced that the odds of the Pitt getting those two yards were better than the odds of Pitt winning in a subsequent OT if they had tied with with a FG.

by kibbles (not verified) :: Thu, 10/11/2007 - 11:56pm

Re #75: How's that any different from LSU going for it on 4th-and-short in the red zone when down by 3 last week? Why is Wannstadt an idiot and Miles a genius, other than the fact that Miles' gamble worked?

by oljb (not verified) :: Fri, 10/12/2007 - 11:16am

A few reasons. This wasn't as short a fourth down, if I recall correctly. And the play-call was suspect; Pat Bostick had been having a good night, but he hadn't been throwing fades every pass play, nor had he been dealing with making a completion bunched up in the end zone. Pitt had good success on the ground with Shady McCoy, so perhaps it would have been more defensible to call a rush to pick up those yards.

The flipside is neither team had been able to stop the other consistently the whole night. There's no reason to think this was Pitt's only realistic chance to win. They would have not been in any weaker of a position going into the 3rd OT.

I don't think Miles is a genius, necessarily, for that decision in the first place. Part smart, part lucky. But if we are getting into a discussion of the relative coaching abilities of Wannstadt and Miles, I think there is plenty of evidence outside these two plays to suggest that Wannstadt is a truly sub-par coach, at least for how the game is played today.