2010 Compu-Picks Ratings - College

The Compu-Picks model rates the top and bottom teams in college football after week twelve

As of the end of Saturday's games (Tuesday's game is counted for next week because MiamiOH played this past Wednesday, and it makes record-keeping difficult to have teams with two games in the same week), these are the top 25 and bottom 20, along with conference ratings (straight average of the scores for each team in the confernece), the top ten performances of the year, and the top five / bottom five home-road splits. Remember that this is a predictive model, designed to pick games and show how good a team actually is. Its results can be very different from what you'll see elsewhere. The workings of the model are confidential (it is, after all, designed to make winning picks), but I'm happy to answer questions about the models' results.

Team Ratings

Rank	Team	League	Score	Schedule Rank	BCS Rank
1	Oregon	Pac-10	0.92	9	1
2	Boise State	WAC	0.88	36	4
3	Texas Christian	Mountain West	0.85	35	3
4	Stanford	Pac-10	0.85	3	6
5	Alabama	SEC	0.68	17	11
6	Auburn	SEC	0.67	11	2
7	Virginia Tech	ACC	0.67	33	16
8	Ohio State	Big Ten	0.64	51	8
9	Nebraska	Big 12	0.58	29	15
10	Arkansas	SEC	0.58	14	12
11	Missouri	Big 12	0.57	15	14
12	Oklahoma	Big 12	0.56	19	13
13	Oklahoma State	Big 12	0.56	44	9
14	Louisiana State	SEC	0.54	21	5
15	South Carolina	SEC	0.54	12	18
16	Arizona	Pac-10	0.50	10	21
17	Wisconsin	Big Ten	0.47	46	7
18	Florida State	ACC	0.46	27	22
19	Texas A&M	Big 12	0.45	18	17
20	Nevada	WAC	0.44	83	19
21	North Carolina State	ACC	0.42	31	23
22	Southern California	Pac-10	0.42	7	NR
23	Miami (Florida)	ACC	0.39	13	NR
24	Utah	Mountain West	0.39	59	20
25	Oregon State	Pac-10	0.39	5	NR
29	Michigan State	Big Ten	0.34	55	10
101	Louisiana-Monroe	Sun Belt	-0.45	94
102	Florida Atlantic	Sun Belt	-0.49	103
103	Tulane	C-USA	-0.50	96
104	Nevada-Las Vegas	Mountain West	-0.51	67
105	Colorado State	Mountain West	-0.51	77
106	North Texas	Sun Belt	-0.53	117
107	Rice	C-USA	-0.55	99
108	Kent	MAC	-0.58	113
109	Bowling Green State	MAC	-0.59	111
110	San Jose State	WAC	-0.61	62
111	Western Kentucky	Sun Belt	-0.63	116
112	Middle Tennessee State	Sun Belt	-0.65	120
113	Ball State	MAC	-0.65	119
114	Louisiana-Lafayette	Sun Belt	-0.66	105
115	New Mexico	Mountain West	-0.73	88
116	New Mexico State	WAC	-0.74	95
117	Memphis	C-USA	-0.75	90
118	Eastern Michigan	MAC	-0.79	110
119	Buffalo	MAC	-0.79	104
120	Akron	MAC	-0.91	109

League Ratings

League	Rating	OOC Schedule Rating
Pac-10	0.37	0.15
SEC	0.32	-0.20
Big 12	0.23	-0.05
ACC	0.14	0.07
Big Ten	0.12	-0.15
Indep	0.10	-0.08
Big East	-0.02	-0.09
WAC	-0.05	0.03
Mountain West	-0.06	0.07
C-USA	-0.23	0.00
MAC	-0.43	0.01
Sun Belt	-0.47	0.05

Top Ten wins of the Year

Game Rank	Team	Opponent	Location	Score
1	Stanford	California	AWAY	48 - 14
2	Texas Christian	Utah	AWAY	47 - 7
3	Oregon	Stanford	HOME	52 - 31
4	Arkansas	South Carolina	AWAY	41 - 20
5	Missouri	Texas A&M	AWAY	30 - 9
6	Stanford	Washington	AWAY	41 - 0
7	Auburn	Arkansas	HOME	65 - 43
8	Florida State	Miami (Florida)	AWAY	45 - 17
9	Oklahoma	Florida State	HOME	47 - 17
10	Nebraska	Kansas State	AWAY	48 - 13

Top and Bottom 5 Home-Road Splits

1	California
2	Connecticut
3	Nevada-Las Vegas
4	Northern Illinois
5	Louisiana-Monroe
116	Navy
117	Western Kentucky
118	Stanford
119	Central Michigan
120	Syracuse

Some thoughts on the list:

1) Last week, I posted the compu-picks top twenty-five and bottom fifteen and this week I'm expanding to the top twenty-five (plus Michigan St, since I'm sure people are curious about where they fit in) and bottom twenty. I will continue to slowly expand the list as the season goes on, and will provide a full list at year's end.

2) The picks had another good week, going 17 - 11 to put together an overall ATS record of 91 - 76 in the five weeks of ATS picks. After an atrocious week 8 pickset, each of the last four weeks has been in the black. The expectation is 55% ATS, and it's not there yet... but it's getting pretty close. 54.5% is a step in the right direction, hopefully it jumps over 55% with this week's picks.

3) After three out of the four weeks I've posted comments, at least one teams the model thought that the BCS overrated got exposed:
After week 8: Auburn, Wisconsin, Michigan St - Michigan St loses 37-6 at Iowa
After week 9: Auburn, Utah, Oklahoma, Wisconsin, Michigan St - Utah loses 47-7 at TCU and Oklahoma loses 33-19 at Texas A&M
After week 10: Auburn, LSU, Wisconsin, Michigan St, Mississippi St - Mississippi St loses 30-10 at Alabama (this really wasn't that bad; Bama is a really good team, and it was on the road for the Bulldogs... but Bama and MSU were only seven spots apart in the BCS before the game)
After week 11: Auburn, LSU, Wisconsin, Michigan St, Mississippi St - no teams got exposed, though LSU and Michigan St came pretty close

Of course, there have been ones going the other way, most notably Oregon St's implosion two weeks against Washington St (followed by a great win against USC). But it certainly seems like there have been more noteworthy bad showings by teams the model thought overrated than those it thought underrated.

4) After last week's games, one team has jumped into the mix at the top: the Stanford Cardinal, off of their dominating win at Cal, who had been extraordinarily good at home so far this year. That win jumped into the #1 spot of best wins of the year, leapfrogging TCU's dominating win at Utah a couple weeks ago. What's amazing about Stanford this year is how consistently dominant they've been against a VERY hard schedule (11 out of 11 1-A games were against AQ teams, 6 of their 10 1-A games so far have been on the road; and the Pac-10 has been VERY good in its non-conference results this year) - six of their 10 1-A games so far have been dominant wins: 35-0, 68-24, 37-14, 41-0, 42-17, and 48-14. Only a bare handful of teams have been more dominant than Stanford this year, and none of them have done it against tougher schedules (and only Oregon's schedule really comes close). To be perfectly honest, I can't think of a single team rated below them that I'd take to beat them on a neutral field. I am totally fine with the system having them at #4.

5) Wisconsin continues to be a fascinating resume. If you only look at their first five 1-A games (including Minnesota), they're maybe a top 40 team, but not even in the ballpark of a top 10 team. If you only look at the most recent five 1-A games (after Minnesota), they're easily a top ten team. So the real question here is, did they flip the switch mid-season and are simply a totally different team than they were at the start of the year? Or are they an up and down team, just randomly up for a few games recently and down for a few games early? The evidence continues to mount that they're actually the first option, just a much better team than they were early in the year. So the real question here is, should they be judged solely based on their last few games? If you're rating how good a team is playing right now, they're a top 10 team. However, it's still half of their resume we're talking about here. Yes, they're playing far better currently, but you can't just totally throw out their mediocre performances for five whole games. I think the model is underrating them a bit... but I can see the argument. I'd be tempted to throw them into the top 10, though I think "fair" is more like 12 or so.

6) Boise has now jumped ahead of TCU after a absolute slaughter of Fresno St. The Nevada game is huge for them, since it's their last chance to make a public appeal, but right now it looks like they should be fairly safe ahead of TCU barring a poor showing at Reno. It's not a huge difference, but this late in the year it's more and more difficult to overcome differences, since resumes become increasingly well-defined with each successive week of play. In terms of schedule strength, as the table shows it's almost exactly a wash. The key difference is that Boise has been more dominant overall (actually they're rated #1 in this department).

7) I've been asked quite a bit about both Alabama and Auburn's ratings in past week, especially since the model has Alabama rated (slightly) higher. As I've noted before, the model definitely has an error range to its ratings (and, for the record, I believe that ALL models do as well, whether they realize it or not), and there's no question that the difference between these two teams is well within the model's error range. So I'm not going to argue why Alabama is higher than Auburn, since they're really tied.

Instead I'm going to go into why the model thinks that they're basically tied, given that Auburn's schedule was rated tougher and Alabama had two losses. The biggest issue here is the high number of close to very close wins for Auburn. 8 points, 7 points, three games of 3 points each (one in overtime)... that's a pretty large sample size of close games. Meanwhile Bama has consistently dominated; of their 10 1-A games, six have been 20+ point wins, and seven have been double-digit wins (it's four and five for Auburn). Moreover, one of Bama's two losses was a close three-point loss, which means that they only had one decisive loss... and it's not like South Carolina blew them out.

The ultimate question here is, how much are close wins really worth? Obviously you can't just plug in margin and be done with it... but when you're rating a team it's still nowhere near worth the same thing as a blowout win. And this is especially true for an overtime win; overtime games are such crapshoots that a win is barely worth more than a tie (and a loss barely worth less). At some point, you have to come up with a rule set to value close wins. Compu-picks has one such system, and it says that Auburn's relatively large number of close wins is in aggregate less impressive than Alabama's relatively large number of blowout wins, coupled with a close loss and one obviously bad performance. Is that right? Not necessarily; there are plenty of interpretations out there, and I'm not going to stand here and pretend that this is somehow the only way of looking at it. But is it defensible for a predictive system? I believe it is.

And, for the record, so does Vegas; Alabama is currently a 4.5 point favorite at home against Auburn, which means that the books also think that Alabama is a bit better (since HFA is generally considered to be worth in the ballpark of 3-4 points). Now, that by itself doesn't make compu-picks right about it, but it does highlight the fact that plenty of other people are rating these two teams VERY closely together. You can also take a look at Sagarin's predictor system (link here), which has Alabama 5th with a 93.27 score, and Auburn 11th with a 86.40 score, which basically means his system thinks Bama is a touchdown better on a neutral field, which is a bigger pro-Bama conclusion than even Vegas has come up with. Again, that doesn't mean that they're right and compu-picks is wrong, but it does reinforce the idea that when comparing these two teams, compu-picks is either in line with other systems out there or, in Sagarin's case, far less in the Bama corner than another well-regarded predictive system.

8) The following teams are ranked materially higher by the model than the BCS: Boise St, Stanford, Alabama, Virginia Tech, Nebraska, Arizona, USC, Oregon St (compu-picks had NC St ranked last week, but now that the BCS also has them ranked the difference is no longer material). Putting aside Virginia Tech for now (as noted in previous weeks' notes, this difference is partially due to not counting AA games... though 16th is probably too low for how well they're playing right now), let's look at the other teams.

Boise, Stanford and Alabama have been talked about above, so I won't go into them again.

Virginia Tech is an interesting case. Part of this difference is that the model doesn't count AA games (which means that the JMU loss doesn't count against them here), but just as important is the fact that they've gone on a tear through the ACC, 7-0 and extremely likely to make it 8-0 as a heavy favorite against Virginia this coming weekend. Compu-picks may have them too high (although to be honest I'm not sure), but the BCS has them too low.

Nebraska just dropped seven spots (#8 to #15) for a very close road loss to a top 20 Texas A&M team. You know how many spots compu-picks dropped them? Zero. Yes, last week they were basically tied with #8 (Ohio St) and now they're basically tied with #10 (Arkansas), and their rating did go down a bit (0.60 to 0.58), but they really didn't move much of anything. It might have been reasonable for the system to drop them a spot or two, but the fact that an extremely respectable road loss could drop them like a rock in the BCS rankings is an indictment of the BCS far more than their rating holding steady is an indictment of compu-picks. If ever you wanted a good example of the fact that the BCS actively punishes tough schedules, there you go. Winning that game would have maybe bumped them a spot, while losing that game killed them. And you wonder why some teams (and some leagues for that matter) schedule the most cupcake-laden non-conference slates they can get away with?

Arizona is the same type of story as most other teams that compu-picks thinks is underrated: very tough schedule, tendency towards blowout wins and close losses (except at Stanford). Despite what the BCS thinks, there simply aren't 20 teams out there better than the Arizona Wildcats.

USC isn't eligible to be ranked by the BCS. If they were eligible, they'd very possibly be ranked. Four losses is never fun, but they've had a tough schedule, they've beaten a top 25 Arizona team and a nearly top 25 Hawaii team, both on the road, and two of their losses have come to elite Oregon and Stanford teams (the Stanford loss was a nail-biter as well).

You can kill them and kill them, but they still come back. Oregon St returns to the compu-picks top 25 after a dominant thrashing of USC, right on the heels of an ugly loss to Wazzu. Presuming they lose to Stanford and Oregon, they're going to be one of the best ever 5-7 teams. They will have played each of the top four rated teams by compu-picks, three of which were on the road, which means that they'd be 5-3 against everyone other than the national elites. Of those other three losses, one was by one point in overtime, one was by three points in regulation, and the third was the ugly 17-point loss. Meanwhile they've also logged a pair of dominant wins, 35-7 and 36-7, though the other three wins were all close. Ultimately, this team is MUCH better than their 5-5 record shows.

However, if they don't make a bowl game it's ultimately their own fault. The current college football structure rewards weak schedules and punishes hard ones, and by scheduling a far too aggressive non-conference slate (TCU and Boise were both expected to be top 15 at the least, and they were both road games to boot), they put themselves in a position where they could fail. And that's what's happened to them this year. If they'd ditched one of the Boise/TCU games and instead scheduled a home game against, say, Buffalo, or Middle Tennessee St, or Marshall, they'd be already at six wins and heading to a bowl. They scheduled like they were an elite program, able to navigate all the landmines without being vulnerable to injuries and/or bad luck in close games, and barring an upset against Stanford or Oregon, they're going to end up home for holidays because they couldn't overcome injuries and bad luck in close games.

8) The following teams are ranked materially lower than the model than the BCS: Auburn, LSU, Wisconsin, Michigan St.

Auburn and Wisconsin have been discussed above, I won't rehash them here.

LSU's schedule may not be at the very top of the list, but it's been strong and absolutely in line with their neighbors in the rankings. Where they're getting hurt is, unsurprisingly, their large number of close to very close wins. 6 points (UNC), 6 points (WV), 2 points (Tenn), 4 points (Florida), 3 points (Bama), and 7 points (Ole Miss) has pretty much defined their resume. They dominated Miss St, which was a very good showing, but other than that it's been the Auburn loss, a bunch of close to very close wins, and comfortable wins against Vandy and ULM, neither of which are going to move the dial much. They'll have a good chance this week at Arkansas to define their resume, one way or the other.

Michigan St just rose two points after barely beating a bad Purdue team at home. See the above Nebraska section on how the BCS punishes tough schedules and rewards easy ones. Moreover, in terms of season-long resumes, Michigan St has not been especially dominant (especially after they got waxed at Iowa), and not had much of a schedule to date (Notre Dame was actually the best by far of a really bad set of non-conference opponents, which continues to hold them down, though not as much now that they're well into the Big Ten schedule). Even if you think compu-picks is too low on them, putting them into the top 10 is flat-out ridiculous. They're basically a homeless man's LSU, with the same tendency towards way too close wins, a lack of good performances against teams not named Wisconsin, a crummy schedule, and the massive beating Iowa laid on them a few weeks ago.

9) The SEC's OOC schedule strength number is a bit misleading, since they still have four games against the ACC (though one of those games is against Wake); that number is going up even before bowl season. For that matter, the ACC's OOC schedule strength number will also go up due to those same four games (though one of those four games is Vandy).

10) This isn't directly to do with the list, but here's a couple fun lists of results:

Texas 20, @ Nebraska 13
Nebraska 48, @ Kansas St 13
Kansas St 39, @ Texas 14

@ Michigan St 34, Wisconsin 24
Wisconsin 31, @ Iowa 30
@ Iowa 37, Michigan St 6

Alabama 24, @ Arkansas 20
Arkansas 41, @ South Carolina 20
@ South Carolina 35, Alabama 21

If you try to apply "head to head is the only thing that matters" logic to this list, your head will explode. You can tease out certain information from these lists (Wisconsin and Alabama had both of their games on the road, they get a bonus; Iowa, Nebraska and Arkansas had losses much closer than their wins, therefore they get a bonus; etc.), but what it really does is highlight that each of these results was JUST ONE GAME. To properly evaluate a team, you need to evaluate the whole resume, not pretend that a single result means everything and the rest almost nothing just because of head to head "logic". That's why Compu-Picks doesn't give ANY special consideration to head to head results. You are what your resume says you are. Period.

Technical notes about the lists:

1) Conference ratings are straight averages of all of the teams in the league. There is no "central averaging" (like Sagarin does), or over-weighting the top teams, or anything like that. Such approaches would yield different numbers, and could potentially change the order of some of the leagues.

2) Games against AA teams are not counted. There are many good arguments both for and against counting such games (see this link for an interesting analysis of the issue). I have elected not to count these results in the Compu-Picks model. As is the case almost every year, this means that one or two especially surprising AA upsets don't make it into the numbers, skewing the results to a fair degree for a couple of teams. I believe that this is a more than acceptable tradeoff given the substantial issues that counting AA games would create, but you are certainly welcome to disagree with my decision on this matter.

3) As mentioned here, the purpose of this system is to make picks, not to create a list used for rankings. As such, I evaluate the system solely on the basis of how good a job it does making picks. I do not evaluate the system on the basis of whether or not it agreed with AP polls, BCS rankings, the BCS computers, or any other such list out there. In fact, the system has a long and established history of being substantially different than those sources. I am fine with these differences. To be honest, I publish these lists because I find them interesting and thought-provoking, and because I believe it is a good thing to introduce an approach that doesn't simply regurgitate the same avenues of thinking as you can find in most places.

4) The system is noisy, especially earlier in the year. This is why I start with only a top 10 / bottom 10 list, and slowly expand it. While I believe that the numbers are reasonable, I certainly accept that they're not perfect. If you believe that a specific team is over- or under-ranked, you may well be right. I bring this up because if you're going to criticize the system for being wrong about a team, I'd appreciate it if you explain why you think the system is substantially wrong, rather than just marginally so (if it's just one or two slots off, especially well before the end of the year, I'd consider that well within a reasonable error range).

5) Some people have expressed curiosity about compu-picks' schedule ratings. Essentially, it's the average difficulty of all 1-A games that a team has played to date, which basically means the average rating of all of a team's 1-A opponents, then adjusted for things like home-field advantage. This, of course, varies enormously from the NCAA's official schedule ratings, which simply look at the opponents' winning percentages and that's it. Needless to say, the NCAA's approach is silly and deeply flawed; it's better than picking numbers out of a hat, but no serious analyst should rely on it at all when trying to evaluate how difficult a team's schedule actually was.

2010 Compu-Picks Blog

Questions, comments or suggestions? Email me at [e-mail]