VN: Let's Hear It for Some Upsets

VN: Let's Hear It for Some Upsets
VN: Let's Hear It for Some Upsets
Photo: USA Today Sports Images

by Bill Connelly

We'll say this for Florida State: they might not be as good as they used to be (okay, they clearly aren't as good as they used to be), but any game in which they play will probably be all sorts of exciting. Not Washington exciting, but still fun. They've played six games (out of seven) decided by 10 points or less. And the only blowout they've played was a road win over a top-10 team (BYU). They threatened to lay a giant egg last night, but they pulled it together and knocked North Carolina off in the fourth quarter. Figure them out, we dare you.

As is the case in this space, let's take one last look at last weekend, then move on to this weekend.

Box Score of the Week

When is a win only a win and not a sign of great things to come?

Colorado 34, Kansas 30

In last week's Varsity Numbers, this game was picked as a potential upset simply because Kansas had been getting by without actually playing like a top-25 team. Last Saturday night in Boulder, they outgained Colorado by a relatively significant margin (7.2 EqPts) and still found a way to lose thanks to two gift-wrapped turnovers and blown opportunities.



Close % 94.4%
Field Position % 49.3% 38.2%
Leverage % 68.0% 67.7%
EqPts 28.0 20.8
Close Success Rate 43.3% 42.7%
Close PPP 0.34 0.31
Close S&P 0.768 0.733
EqPts 3.7 11.9
Close Success Rate 37.5% 42.5%
Close PPP 0.14 0.30
Close S&P 0.514 0.724
Line Yards/carry
2.12 2.86
EqPts 24.3 8.9
Close Success Rate 45.1% 42.9%
Close PPP 0.40 0.32
Close S&P 0.848 0.747
SD/PD Sack Rate
8.3% / 9.1% 20.0% / 0.0%
Success Rate 54.9% 43.5%
PPP 0.38 0.34
S&P 0.933 0.772
Success Rate 25.0% 40.9%
PPP 0.35 0.24
S&P 0.601 0.652
Number 2 3
Turnover Pts 13.45 13.49
Turnover Pts Margin
+0.04 -0.04
Q1 S&P 0.542 0.438
Q2 S&P 0.665 1.155
Q3 S&P 0.757 0.571
Q4 S&P 1.229 0.776
1st Down S&P 1.196 0.749
2nd Down S&P 0.487 0.722
3rd Down S&P 0.501 0.712
Projected Pt. Margin
+7.2 -7.2
Actual Pt. Margin
-4 +4
  • First of all, the repercussions of this game for Kansas are rather devastating. They have now played the two most winnable games of their Big 12 slate -- a home game against Iowa State and a road game against Colorado. Due to a rough South schedule (Oklahoma at home, at Texas Tech, at Texas), the Jayhawks almost needed to win out against their own division to have a chance at the Big 12 Championship Game. Instead they barely held on to beat Iowa State and then gave the game away against Colorado.
  • How does a team perform 11.2 points below their projected value? Easy: blown opportunities. As fun as it is to tear a game down and build it back together with EqPts, a team still has to execute when given the opportunities the numbers say they have. When they handed Colorado the ball deep in their own territory with two turnovers, the Buffaloes scored two touchdowns. When Colorado did the same for Kansas, the Jayhawks settled for two field goals. And when they twice advanced the ball inside the Colorado 20 in the last five minutes, they came away with zero points, turning the ball over on downs the first time and running out of time the second.
  • A lot is being made of the performance of Tyler "Don't Call Me Taylor" Hansen, who stepped in for ineffective-for-two-years Cody Hawkins at quarterback for Colorado. While he extended a series of plays with evasive maneuvers, and while the Buffs caught fire for 24 points in the second quarter (14 of which were set up by Kansas turnovers), there is not a lot here that suggests long-term success for Hansen and the Buffaloes offense. Kansas rode into Boulder having given up over 1,300 yards in three games against Duke, Southern Miss and Iowa State. Colorado managed 322, and while it was enough to win, it was still below average considering the standard Kansas has set on defense this year. He may have given the Buffaloes a nice change of pace, but odds are that defenses will figure him out and pen him in, just like they did last year when he played this exact same change-of-pace role.
  • Kansas' remaining six games: the three aforementioned South battles, plus a home game against Nebraska, a trip to Kansas State, and a border battle with Missouri at Arrowhead Stadium. Not a gimme win in the bunch. The Jayhawks' North title chances may indeed have evaporated Saturday.
  • The Big 12 North really might be terrible.

S&P+ Top 25

We are almost to the point of getting S&P+ data updated weekly on Football Outsiders, honest. But for one more week, let's take a quick look at the updated Top 25. Once again, these rankings are quite volatile and heavily-dependent on strength of schedule. The current mix is 80 percent 2009 figures, 20 percent preseason projections. Again, the decision was made to phase the projections out of the equation very slowly because of the ratings' overall volatility.

S&P+ Top 25 (After Seven Weeks)
Rk Team S&P+ Last
Chg Off.
Rk Def.
1 Florida (6-0) 270.4 1 0 125.2 6 145.2 3
2 Alabama (7-0) 267.3 2 0 117.3 17 150.0 1
3 Arkansas (3-3) 261.9 5 +2 145.8 1 116.1 24
4 Penn State (6-1) 256.4 8 +4 118.2 15 138.2 5
5 Tennessee (3-3) 251.1 3 -2 121.3 11 129.8 10
6 TCU (6-0) 248.7 11 +5 138.6 2 110.1 28
7 Miami (5-1) 245.3 10 +3 120.0 13 125.3 17
8 Nebraska (4-2) 244.9 4 -4 104.5 42 140.5 4
9 Oklahoma (3-3) 244.4 6 -3 109.6 32 134.8 6
10 LSU (5-1) 243.5 7 -3 116.4 19 127.0 12
Rk Team S&P+ Last
Chg Off.
Rk Def.
11 Texas (6-0) 242.6 16 +5 112.7 24 129.9 9
12 Wisconsin (5-2) 240.4 14 +2 118.1 16 122.3 18
13 Ohio State (5-2) 238.4 9 -4 112.0 25 126.4 15
14 Oregon (5-1) 236.0 15 +1 109.2 33 126.8 13
15 Boise State (6-0) 235.5 21 +6 108.9 34 126.7 14
16 Clemson (3-3) 233.2 29 +13 83.9 104 149.3 2
17 Iowa (7-0) 232.2 24 +7 101.0 61 131.3 7
18 Virginia Tech (5-2) 232.1 12 -6 116.0 20 116.1 23
19 Missouri (4-2) 230.3 20 +1 102.8 51 127.5 11
20 Cincinnati (6-0) 229.8 25 +5 128.4 3 101.4 45
Rk Team S&P+ Last
Chg Off.
Rk Def.
21 Nevada (3-3) 227.9 19 -2 125.5 5 102.4 41
22 Georgia (4-3) 225.1 18 -4 123.3 9 101.8 44
23 BYU (6-1) 224.2 13 -10 121.1 12 103.1 38
24 Fresno State (3-3) 223.3 22 -2 121.5 10 101.8 43
25 USC (5-1) 222.8 17 -8 97.5 72 125.3 16

Biggest Movers of the Week

The rises and falls were not quite as significant as they were last week, but some teams did still manage rather significant movement.

Notable Rises

Texas Tech (15 spots, from 56th to 41st). The fun part about a volatile system like this is how it rewards some teams for winning more than it punishes others for losing (or vice versa). Case in point: Texas Tech thoroughly humbles Nebraska in Lincoln, and while the Cornhuskers only drop four spots, from 4th to 8th, in the standings, the Red Raiders make a pretty hefty jump. Tech looked excellent in making the plays to fluster the Nebraska offense, building the lead with just enough offensive firepower, and running out the clock with a more conservative offensive approach. They are only 41st due to some severe issues in the running game during the first month of the season, but they have looked outstanding for consecutive weeks now.

Clemson (13 spots, from 29th to 16th). Clemson's 38-3 win over Wake Forest was statistically domination in almost every facet. They averaged almost 8 yards per pass and 6 yards per rush, and they held the Demon Deacons to 178 total yards, 3.7 per pass and 1.8 per run. They built a crushing 31-3 halftime lead and coasted home. If they continue to play like that, they are easily the best contender for the ACC Atlantic Division and could await a worn-out winner of the Miami-Georgia Tech-Virginia Tech battle royale in the ACC Coastal.

South Carolina (13 spots, from 47th to 34th). Strength of schedule is a mighty thing. The Gamecocks remained mostly competitive with Alabama (Mark Ingram's "Hello, world" rushing aside), and they are benefiting from having played four SEC conference games while conferences like the Big Ten and Big 12 are only now in full conference swing.

Georgia Tech (12 spots, from 58th to 46th). Paul Johnson's team looked outstanding in their win over Virginia Tech, but they still have quite a way to go to impress the S&P+ formulas.

Other major rises: San Diego State (74th to 53rd), Akron (102nd to 85th), Ohio (98th to 84th), Notre Dame (66th to 55th), Iowa State (94th to 83rd).

Notable Falls

North Carolina State (14 spots, from 64th to 78th). A week after losing to Duke by 21 (at home!), the Wolfpack looked listless in allowing 480 yards and 52 points to a Boston College offense that managed 217 yards combined against Clemson and Virginia Tech earlier this year. Ouch.

Texas A&M (14 spots, from 40th to 54th). However, for losing by 48 points to a previously listless Kansas State team, you the honorary rank of 120th.

Pittsburgh (12 spots, from 62nd to 74th). Pittsburgh actually got punished for Rutgers' weak schedule. The Scarlet Knights are hovering low in the S&P+ rankings thanks to a series of poor showings and the presence of two terribly weak FCS teams on the schedule, and Pitt only beat them by a score of 24-17 last week.

BYU (10 spots, from 13th to 23rd) . TCU fans salivating at a chance to get on ESPN College Gameday would have been awfully upset if BYU had messed around, lost to San Diego State, and watched as Gameday headed to Starkville, or Tuscaloosa, or Columbia instead. The Cougars ended up winning relatively comfortably, 38-28, but it takes a much bigger win than that over the Aztecs to impress the ratings.

Other major falls: East Carolina (93rd to 107th), Louisiana-Lafayette (101st to 114th), Kansas (46th to 59th), Wake Forest (60th to 72nd), Baylor (79th to 90th).

Random Golf Clap

To Joe Posnanski, for writing the best Joe Paterno profile ever written. Joe Poz has been just about the best sports writer in the game for a while now, and he hammered home something that could have easily been too cheesy to enjoy. Joe Pa is such a unique figure, and this profile captures all of his contradictory features in loving detail. Hiring Posnanski was the smartest thing Sports Illustrated has done in a long time. And that's saying something considering they hired a former roommate of yours truly.

And a brief golf clap to Kansas State. They may only be leading the Big 12 North because they had the luxury of playing Iowa State and Texas A&M, two of the Big 12's three worst teams (along with Baylor), but it is rare that a team goes from losing by 52 points (66-14 to Texas Tech) to winning by 48 (62-14 over Texas A&M), and it deserves at least brief applause. Texas A&M now plays Texas Tech. Surprisingly, the Vegas line is not Tech minus-100. The transitive property fails us again.

Random Mini-Rant

Can we please get replay officials bigger monitors? Twice during Saturday night's Missouri-Oklahoma State game, fumble calls were challenged. First, Missouri's Jasper Simmons appeared to fumble a third-quarter kickoff. The replay showed pretty conclusively that Simmons' knee was down, and the ground aided in causing the fumble. The review upheld the fumble call. Then, late in the game, with Oklahoma State up comfortably, Keith Toston appeared to get stripped as he was falling to the ground. ESPN's replay showed the ball was coming loose before he hit the ground. Ruling on the field: He was down. Ruling from the replay booth: He was down. Neither made a significant impact on the game itself (two killer drops in the first half did Missouri in), but during both reviews, cameras showed the poor, lonely replay official in his booth, looking at replays on a monitor that could not have been more than 12-16 inches. This is 2009. You can find a ridiculous deal on 40-inch high-definition televisions. If you are going to have an instant replay system, which is highly recommended, you need to make sure your official has the necessary tools to make the replay worth your while. Don't waste everybody's time if the poor guy can't actually see well enough to make the correct call.

This is also probably rant-worthy, but we'll pass. It's extremely well-covered territory.

Random Reasons to Love College Football

The third Saturday in October.

Another reason: Some time in late-December, Idaho will be playing a football game, likely in a place like Albuquerque, Honolulu, or Boise. They will be playing a team like Air Force, Southern Miss, East Carolina, or UTEP. And they could not be more excited. Keep the playoff talk coming -- it probably would be a great thing for college football. But do not get rid of the bowls. They are one of the things that make college football so uniquely great.

Random Top 10

This section is in honor of the 1,420,934 autumn leaves currently residing in this writer's driveway (after we raked on Sunday).

"Ashes in the Fall" by Rage Against the Machine
"Autumn Sweater" by Yo La Tengo
"Dead Leaves and the Dirty Ground" by The White Stripes
"Fall" by The Watson Twins
"Falling Free" by David Gray
"My Cosmic Autumn Rebellion" by Flaming Lips
"November Has Come" by Gorillaz
"November Rain" by Guns N' Roses (you knew that one was coming)
"October" by U2
"October Song" by Amy Winehouse

Upset Watch

We only nailed one of three upset picks last week -- Colorado over Kansas. That brings the three week total to 7-5, which is still a pretty stout record. This week's major upset possibilities:

Maryland over Duke. Spread: Maryland +5.5 | S&P+ Projection: Maryland by 7.8.
Virginia over Georgia Tech. Spread: Virginia +5.5 | S&P+ Projection: Virginia by 8.5.

As jarring as it may be to see Duke favored over any conference team by more than five points, this one is relatively justifiable. Their last two times out, they took Virginia Tech to the wire before losing by eight, and they trounced N.C. State by 21 in Raleigh. They are hosting a 2-5 Maryland team who, in the last two weeks, has managed to give up 42 points to Wake Forest, then score less than 10 against Virginia. Only a strange upset of Clemson is giving them any chance of a rebound this year. But from start to finish, the Terps have still played a bit better than Duke, and since the S&P+ projections do not factor momentum into the equation, the Terps are still projected to come out on top. Meanwhile, Virginia's "Who needs offense?" resurgence faces a pretty tall task against a Georgia Tech team coming off of its best win of the year. But hey, the S&P+ projections saw Virginia's October comeback coming, so who's to question it now?

South Florida over Pittsburgh. Spread: USF +6.5 | S&P+ Projection: USF by 2.4.

A series of teams are taking turns positioning themselves to take on Cincinnati atop the Big East. South Florida, Connecticut, West Virginia, and Pittsburgh have all looked good enough to be the primary challenger at one time or another. Thanks to Pittsburgh's drop in the ratings (explained above), USF now gets the nod in a battle that contains two of the more interesting freshmen in the country, Pitt's Dion Lewis and South Florida's B.J. Daniels.

UCLA over Arizona. Spread: UCLA +8.5 | S&P+ Projection: Arizona by 4.3.

It wouldn't be an upset watch without at least one Pac-10 game. If you can figure out what to do with either of these teams, you are ahead of the S&P+ ratings. Arizona appears to be putting things together. They moved to 2-1 in conference last week by taking Stanford down in a 43-38 track meet. Sophomore Nick Foles completed 78 percent of his passes for 415 yards and three touchdowns. Meanwhile, UCLA's defense looked strong until last week's 45-26 shellacking at the hands of California (a game which included this ridiculous run). This appears to be a strength vs. strength (Arizona offense vs. UCLA defense) and weakness vs. weakness (UCLA offense vs. Arizona defense) matchup.

Closing Thoughts

To reiterate what I mentioned in yesterday's 7DA, this year does seem to have stabilized a bit in terms of huge upsets, but there really is almost no way that all nine of the BCS Top 15 teams playing on the road tomorrow come back home with wins. There are a ton of road favorites this week, and at least one of them is going down. And honestly, I kind of hope quite a few of them do. It's time for college football itself to steal the spotlight back from officiating and injuries, and a rash of upsets is the best way to do just that. So, to Mississippi State, Tennessee, Michigan State, Washington, Virginia, Baylor, and (of course) Missouri, I say ... go team!


22 comments, Last at 25 Oct 2009, 12:59pm

1 Re: VN: Let's Hear It for Some Upsets

Virginia's 8.5 pts better than Ga Tech? I have a piece of advice. Turn of your computer and go buy yourself a television.

3 Re: VN: Let's Hear It for Some Upsets

Why don't you just change the name from "the S&P+ Top 10" to "the SEC Top 10"?

4 of the top 5? C'mon, son.

15 Re: VN: Let's Hear It for Some Upsets

I'd love to have seen some of those bowls take place when the Big 12's timing-heavy offenses didn't have to take six weeks off between games. It was not the best showing from the conference in late-December/early-January, but I can guarantee that the layoff was at least part of the problem.

6 Re: VN: Let's Hear It for Some Upsets

I find the rankings better than the conventional wisdom, which is one large point int heir favor. Conventional wisdom on college football is typically horrible and it is always depressing to see so many people defend it.

8 Re: VN: Let's Hear It for Some Upsets

Even when you have teams who hold two score, head-to-head wins over other teams with a better record and tougher schedule but are ranked lower?

I agree that conventional wisdom is often a bad thing in college football but there's something to be said for common sense and watching some of these teams play.

Arkansas and UT are absurdly high. Iowa won at PSU has played a tougher schedule and is 13 spots lower. They just won at Wisconsin by 10 and have a better resume and schedule. Nevada and Fresno over USC? Clemson?

9 Re: VN: Let's Hear It for Some Upsets

That is not what the system is measuring, and honestly one result means so little it is almost meaningless. There is barely enough data in college football to do ANYTHING with it. This system is at least a system, instead of a bunch of silly hunches and biases.

The whole point of something like this is to say "What if we ranked teams by plausible system X?". And then you get a result and go, wow that is interesting now maybe I see some things I didn't see when I just assumed that the 6 no loss teams were the best six teams and the 7 one lost teams were 7-13 etc.

99% of the criticism is see of this DVOA and FEI boils down to. "This imperfect system of numbers is not giving me the results I think a perfect system (my personal gut feeling) should!. You people are horrible idiots how can you not see how flawed result x sub 18975 is!" Completely ignoring the fact that result x sub 18975 is interlaced with thousands of other results...

10 Re: VN: Let's Hear It for Some Upsets

And celebrating something just because it's "a system" doesn't make any sense either.

Any system in which a team with 2 losses (Wisconsin) is rated above the undefeated team they lost to at home by multiple scores (Iowa) needs to be reevaluated. If Penn State is ranked #4 and Wisconsin is #12, Iowa must have two pretty impressive wins, right? Where are they ranked? At #17, just behind a 3-3 Clemson team that lost to 2-5 Maryland (with the Terps only other win coming against James Madison).

We all understand that rating college football teams is a difficult process, at least in part because so few of the teams actually play each other. But when a system seems to be ignoring head-to-head results and produces rankings that seem this bizarre, it's an easy target. Why shouldn't it be subjected to criticism?

11 Re: VN: Let's Hear It for Some Upsets

I think you are placing too much emphasis on head to head results. Just because Iowa played better than Wisconsin and Penn St. on the Saturdays that they played them doesn't make the system flawed because those two teams are ranked ahead of them. I don't know whether this system is any good or not, but I do know it's designed to look at the whole season. Sure, Iowa beat Penn St. and Wisconsin. But they also needed a blocked field goal to beat Northern Iowa and only beat Arkansas St. by three.

13 Re: VN: Let's Hear It for Some Upsets

It seems to me that you are dismissing what should always be most actual game played on the field between two teams (at Wisconsin's home stadium, by the way). Are you that devoted to this system? Ranking the teams in alphabetical order by winning percentage is a system too.

Even with the close wins over Northern Iowa and Arkansas State, Iowa is ranked #3 in the BCS computers so I don't think questioning the methodology is out of line. Look, the Hawkeyes could certainly lose against Michigan State in East Lansing tonight and I expect they'll probably lose a game at some point during the season. I don't think they're the best team in the country. But results should matter. Explain to me how a team wins road games by multiple scores against two top-15 opponents (according to this methodology), has yet to lose, and can reasonably be ranked behind those teams.

Remember, all 3 teams are from the same conference so they have similar schedules (except Penn State and Wisconsin don't play, giving each team a break). The only team Wisconsin has beaten convincingly is 1-5 Wofford.

12 Re: VN: Let's Hear It for Some Upsets

And it's easy for PSU to put up great numbers when they've played no one besides Iowa. A game which they lost at home by 2 scores.

I'm not saying it should just be based on record or the way the conventional polls tend to work. I'm all for different systems but I agree with Derek. We shouldn't celebrate a system just because it's a "system" that we think is unconventional and outside the box. We have a team that has 3 losses AND has played a weaker schedule, ranked 12 spots higher than a team that's undefeated with a tougher schedule.

Like I said I like the read and it's great that he and others have gone to so much work to try to find something better. It's just that I can look at Sagarin and it at least jives with what I see and and common sense.

Have you seen Nevada play, especially against someone with a pulse? A bunch of domination against bad teams should not override bad play against decent teams. And that appears to be the case with many teams in the S&P

14 Re: VN: Let's Hear It for Some Upsets


I should emphasize that I certainly appreciate the effort that goes into creating something like S&P (and the fact that no ranking will ever satisfy everyone). I'm sure the writers at FO are always looking for ways to improve the formula.

16 Re: VN: Let's Hear It for Some Upsets

I'll try to respond to a few different pieces of criticism in one comment here.

The rankings are crazy. Arkansas #3???

Valid point. Very valid point. This is the first year I've been in a position to record S&P+ rankings in a week-to-week fashion, and it's produced some interesting results. But we are just barely over halfway through the season. The sample size problem that is always an issue with college football stats is doubled. Before you criticize the current results, you should probably take a look at the year-end results, because obviously that's what counts the most.

Here are the final 2008 rankings.

1. USC
2. Florida
3. Oklahoma
4. TCU
5. Texas
6. Penn State
7. Boise State
8. Ohio State
9. Alabama
10. Missouri

TCU and Mizzou are the only truly strange rankings there, and even those two are rather explainable. TCU lost only to the #3 team (and held them to their worst offensive performance of the regular season) and an undefeated Utah team (#16) on the road via last-second score. The S&P+ system does not punish you for tight losses to really good teams...especially on the road. Meanwhile, Missouri's losses came to #3, #5, #14 (OSU), and #20, and they benefited from killing the teams they were supposed to kill and destroying #21 Nebraska.

While we're at it, here's the 2007 final Top 10...

1. Florida
2. USC
3. Ohio State
4. West Virginia
5. LSU
6. Virginia Tech
7. Oregon
8. Oklahoma
9. Missouri
10. South Florida

Another couple of strange results--Florida's near-perfect offense made up for some defensive issues, and USF's late-season collapse probably wasn't punished enough--but again, nothing crazy. No 6-6 teams up there or anything. If you have a problem with the final results, then fine. But these ratings still have to work themselves out a bit, and midway through the season, things are probably going to look strange. Case in point: Arkansas and Tennessee in the Top 5 simply for having played Florida close. At this stage in the season, take these ratings more as suggestions than rock-solid evaluations. These say that Arkansas, Tennessee and Nevada (and yes, I've seen Nevada play multiple games this season--they're probably not a Top 25 team, but they're fast and athletic, and they're getting hot now that they're getting out of their own way) might be better than you think, while teams like Iowa and USC might not be as good as you think, the end.

Head-to-head should matter more.

I agree in theory, but how exactly to you execute that in practice? How should head-to-head have been applied last year to the Oklahoma-Texas-Tech 3-way tie in the Big 12 South? And in Iowa's case...well, the Northern Iowa and Arkansas State games count every bit as much as the Penn State and Wisconsin games. The whole year counts, and while Iowa did indeed beat PSU, Wiscy, Michigan, etc., they have also struggled at times. The obviously deserve credit for winning, and if they continue to do so, they'll continue to move up the ratings, but in every "competitive" play they've been involved in this year, they've only played like the 17th-best team.

17 Re: VN: Let's Hear It for Some Upsets

Fair explanation Bill.

Like I said I appreciate everyone who puts in the time and work to try to find something better. Philosophically I'm with you for the most part. I'm just not sure the current ranking is a good measure of what's happened thus far OR a good predictor of what's likely to happen.
We'll see how it plays out and I'm sure it will be easier to agree with in December.

20 Re: VN: Let's Hear It for Some Upsets

Still Alive I don't know if people are missing that so much as they just don't philosophically agree.

I've read the arguments and understand all of the "luck" variables that come into play in a tight football game and the sample size that we're dealing with. The team in ChiTown's argument obviously is competitive and nearly on par with those top 7 teams.

I just can't reconcile the disregard for actually winning or losing that many times. It's still the essence of the game.

22 Re: VN: Let's Hear It for Some Upsets

You have a point, danb, and I certainly think that in ranking teams for a playoff or for the title game, losses should hurt.

However, most statistical rankings are looking to be predictive, as well, so losing close games doesn't have as much an effect. Losing seven games to top-eight opponents by one point each means it's pretty likely you'll beat just about anyone you play going forward, and also means you would probably win three or four games against those same seven teams if you played them again.

21 Re: VN: Let's Hear It for Some Upsets

It is very difficult to accurately rank college football teams, let alone provide precise estimates of how much better one team is than another. One major issue not being raised is whether a system that attempts to characterize how the data fit is over-fitting the data and thus has little predictive value. As was said, you can fit anything with 4 parameters.

The real question is: how good are the current strength rankings, which when compared for two teams should provide some probability distribution of scores when the two teams played, at predicting future results? That is, when USC plays at Oregon, what is the probability of the observed final score differential given that the strength rankings predict Oregon should win by a mean score of 13.2pts? If USC wins convincingly (say, 14pts), either it's i) an extremely fluke event (i.e. p < 0.05), or it's ii) within the normal range of expected results (i.e., p > 0.05 and variance around 13.2pts is so high as to make predictions relatively meaningless), or iii) the strength rankings aren't accurately reflecting under-lying strengths of teams at all (notice that iii is not mutually exclusive from i or ii). I suspect that we'll never really know, given the paucity of data available.

Why not include some sort of confidence intervals, instead of point estimates (ML or LS or whatever)? It would be interesting to see what the plausible distribution of Iowa's S&P+ rating is, for example? This is probably already available, and I've just missed it.

I'd also like to note that comparing final standings does two things: i) it demonstrates that various objective and subjective systems converge on a similar set of rankings, and ii) there is no way to determine how much better one ranking is than another at predicting the next game (i.e. the season is over). Just because systems converge on a similar answer does not mean that they are converging on the best answer.

Thanks for doing all this work--it's a lot of fun to think about!